https://arxiv.org/arXiv2026-09-23T13:49:19.189153+00:00variouspython-feedgenhttps://arxiv.org/favicon.icohttps://arxiv.org/favicon.icoarXiv blog and daily new submissions, alphaXiv Explore, LessWrong, and 80,000 Hours.tag:trvny.github.io,2024:feedseek/arxiv/ed631fb58f72f222Why I'm scared of RL2026-09-23T12:07:33+00:00Summary: First, I give several different angles on how I feel about reinforcement learning: Theoretical case: RL is a black-box source of agency — this should give us classic misalignment worries, especially compared to agency-via-scaffolding Recent incidents (huggingface etc) and more mundane forms of misaligned behaviour in personal use give me bad vibes about the direction-of-travel of recent AI progress I’m worried things might get worse: if RL environments start incorporating agents, they m2026-09-23T12:07:33+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/b9c3a38b4d5ce448We Underestimate the Weaknesses of Pangram2026-09-23T10:13:23+00:00How much does Pangram's "AI-Generated" label indicate the degree to which an author has outsourced their thinking? When they tested their 4.0 product , Pangram found that, by their definition, the proportion of AI-Assisted documents it classified as AI-Generated was 0.01%, 4%, or 7%, depending on the experiment. Then they omitted the experiments that found 4% and 7% false positive rates (FPRs) on their website , while advertising there that the product detects AI-Assisted writing. Before I conta2026-09-23T10:13:23+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/ffbc54c4c9031ad3Minimal Vs Maximal superintelligence2026-09-23T08:22:13+00:00I've seen lots of arguments here conflate lots of different types of superintelligence. Here I separate out two broad categories, which I'll term minimal and maximal superintelligences. This is an important distinction as they differ in terms of timelines, risks, and mitigations. Maximal superintelligence This is the idealised limit of intelligence. It can solve anything that can be solved by being clever. You can't outsmart it, it's prepared for every contingency, and can react instantaneously 2026-09-23T08:22:13+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/fb4c94232fd194a3What if AI2027 came two months earlier?2026-09-23T07:20:01+00:00Opus 5.5 made this very good website (it's incredible how far webdev has come): https://fluxxrider.github.io/overclocked/ Some cool graphics/screenshots: Discuss2026-09-23T07:20:01+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/46c302a1a792ac5aWorkspaceBench: Evaluating Interpretability Methods for the Global Workspace2026-09-23T06:58:59+00:00TL;DR We introduce WorkspaceBench, a set of evaluations for how well an activation-to-text tool can read the contents of the “global workspace” of a model, i.e. the intermediate variables during a forward pass. The benchmark comprises 3,356 questions across 27 eval families, spanning topics in safety, logical reasoning, and multihop computation, with a subset for single-token-output tools. A desirable property of good interpretability techniques is minimal hallucinations, so WorkspaceBench also 2026-09-23T06:58:59+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/432e7dfe7addf5a7ICDAR2026 Competition on Multimodal Reasoning over Documents in Multiple Domains2026-09-23T04:00:00+00:00arXiv:2609.25055v1 Announce Type: new
Abstract: In this report we present results of the ICDAR2026 Competition on Multimodal Reasoning over Documents in Multiple Domains. This competition aimed to advance research in document understanding through the task of Visual Question Answering (VQA). Building upon previous DocVQA benchmarks, this competition introduces challenging reasoning questions over a diverse collection of documents spanning eight domains, including business reports, scientific pa2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/fe003b48c12c758fMoM: Memory of Memory2026-09-23T04:00:00+00:00arXiv:2609.25054v1 Announce Type: new
Abstract: For a long-horizon LLM agent, the memory question is not what was once recorded but what \emph{currently holds}. Most designs answer it only indirectly: every interaction is stored, and the present is reconstructed at query time by retrieving and reconciling records, so stale values re-enter and the same conflicts are re-litigated. Committing the current value at write time avoids this, but existing write-time (CRUD) memories overwrite, so a wrong2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/ffc27ed9f79299ebLatentPort: Beyond KV Cache - Cross-Model Transfer of Recurrent Memory in Hybrid Language Models: A 4B-to-9B Hybrid-State Handoff Without Target Prefix Replay2026-09-23T04:00:00+00:00arXiv:2609.25053v1 Announce Type: new
Abstract: Can one language model hand its live memory to another without the receiver rereading the context? We demonstrate useful persistent hybrid-state transfer across one architecture-matched Qwen3.5 4B-to-9B sibling pair. To our knowledge, this is the first demonstrated cross-model handoff of persistent recurrent inference state between differently sized hybrid language models without target prefix replay. Translated attention KV alone leaves a large g2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/726b26550b74a2adSelf-Cleaning and Captured Anyway: One Measured Primitive for Error in a Store an Agent Writes to Itself, and What a Falling Score Actually Measures2026-09-23T04:00:00+00:00arXiv:2609.25052v1 Announce Type: new
Abstract: "An agent that writes its conclusions into a store it later retrieves from closes a loop usually reported as one-way contamination. Taking the loop to the infinite-tenure limit against an append-only store gives a different picture: because writing never deletes, the reachable state space has a hard upper edge at (n-1)/n, so the outcome is a choice between two edges rather than a decay. At f_0 = 0.9 the interval between the two modes holds 3.6% of2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/a19f4fbe9cd500b0LLM-Driven Training-free Location-Attribute Synergic Fusion: A Closed-Loop Paradigm for Dual-source Encrypted POIs and LULC Mapping2026-09-23T04:00:00+00:00arXiv:2609.25051v1 Announce Type: new
Abstract: Dual-source encrypted points of interest (DSEP), POIs from two encrypted coordinate systems, suffer from intertwined location and attribute uncertainties, including nonlinear systematic misalignment and naming inconsistency, hindering land-use/land-cover (LULC) mapping. To the best of our knowledge, this paper is the first to propose an LLM-driven, training-free location-attribute synergic closed-loop optimization paradigm for DSEP fusion. The par2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/84487e0758a4f081FrontierMath Erd\H{o}s2026-09-23T04:00:00+00:00arXiv:2609.25050v1 Announce Type: new
Abstract: We introduce FrontierMath Erd\H{o}s (FME), a benchmark of 68 Erd\H{o}s problems that are open as of August 2026. To solve a task in FME, AI systems must resolve (prove or disprove) one of the 68 conjectures in the proof assistant Lean. Our 68 problems were selected by the second author among 652 open problems on erdosproblems.com for their mathematical interest and difficulty. AIs have recently resolved several open problems in mathematics, but th2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/1e9ad5701928174dMitigating LLM Over-Refusal via Dynamic Semantic Routing Calibratione2026-09-23T04:00:00+00:00arXiv:2609.25049v1 Announce Type: new
Abstract: Large language models (LLMs) aligned for safety often suffer from over-refusal, incorrectly rejecting benign yet safety-related instructions. Prior studies primarily attribute this to static representation overlap, largely overlooking the underlying dynamic mechanisms. In this paper, we present the mechanistic analysis of over-refusal through the lens of internal routing conflicts within transformer attention. We discover that a sparse subset of H2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/92c746060cd0c95cPrompt Breadth and Rollout Refresh Interact in On-Policy Distillation2026-09-23T04:00:00+00:00arXiv:2609.25048v1 Announce Type: new
Abstract: How many prompts does on-policy distillation (OPD) need, and how does the answer depend on the student policies that generate its training responses? We study these two controls jointly: prompt breadth and rollout refresh. A 3x3 mathematical-reasoning experiment fixes 14,080 trajectories and 110 optimizer updates while varying the prompt bank and the number of response-generating policy snapshots. With ten snapshots, eight prompts reach 24.09% ave2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/72c8ba6122cbb81aAIBuildAI-2.5: Efficient Autonomous AI Model Development Through LLM-Guided Tree Search2026-09-23T04:00:00+00:00arXiv:2609.25047v1 Announce Type: new
Abstract: Autonomous agents that automatically build artificial intelligence (AI) models could broaden access to AI across science and engineering. A popular line of such agents frames model building as a code search problem and solves it by tree search, in which each node is a candidate program and the tree grows by generating a child program from a parent, and these agents now approach the capability of experienced AI engineers on realistic benchmarks. Ho2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/6d1abcb56a1b98a7Peerify: Benchmarking Peer-Review Claim Verification2026-09-23T04:00:00+00:00arXiv:2609.25046v1 Announce Type: new
Abstract: Peer review plays a central role in scholarly publishing, yet verifying whether reviewer claims are supported by manuscript evidence remains a largely manual and time-consuming process. We present Peerify, a pipeline for manuscript-grounded verification of peer-review claims. Given a manuscript and a review comment, the Peerify pipeline decomposes reviews into atomic claims, retrieves relevant manuscript evidence, and determines whether each claim2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/2f7dae16d51e6734An irreducible decomposition of the Weil representation restricted to open compact subgroups2026-09-23T04:00:00+00:00arXiv:2609.25045v1 Announce Type: new
Abstract: For a $p$-adic field of any residual characteristic, we determine irreducible decompositions of the Weil representation restricted to the maximal open compact subgroups and the Iwahori subgroup by using the Schr\"{o}dinger model. In particular, these decompositions show that the restrictions are multiplicity free.2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/bd945fce00ba35bfEnd-to-End Quantum Semantic Communication with Variational Quantum Neural Networks2026-09-23T04:00:00+00:00arXiv:2609.25044v1 Announce Type: new
Abstract: This paper presents a quantum semantic communication (QSemCom) framework combining quantum machine learning (QML) and semantic communication (SemCom). Classical data are compressed into low-dimensional semantic representations, encoded and processed by a variational quantum transmitter, transmitted through a quantum channel, and processed by a trainable quantum receiver for classification. The framework considers a distributed quantum communicatio2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/9bec18e419660282Encrypted Redundancy as a Diagnostic Resource: Relational Diagnosis in Quantum Encrypted Cloning2026-09-23T04:00:00+00:00arXiv:2609.25043v1 Announce Type: new
Abstract: Quantum encrypted cloning encodes an unknown state into several encrypted components, each offering an alternative way to recover it later. We show that this redundancy can also serve for one-shot fault diagnosis: instead of inspecting the encrypted state, we measure relational Pauli observables testing consistency conditions imposed by the encoding. The canonical Yamaguchi--Kempf scheme encodes the input into \(n\) signal--key pairs, all carrying2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/4becd5b389fc08d4Infinitely Many Binomial Coefficients of Deficiency One2026-09-23T04:00:00+00:00arXiv:2609.25042v1 Announce Type: new
Abstract: A binomial coefficient $\binom{n}{k}$ is good if all its prime divisors exceed $k$, and its deficiency is the number of $k$-smooth integers in $(n-k,n]$. A short constructive proof of the existence of infinitely many good binomial coefficients with deficiency exactly one is given, which provides an answer to a problem of Erd\H{o}s, Lacampagne, and Selfridge.2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/a0d69e904d6868abA square-root law for equitable coloring2026-09-23T04:00:00+00:00arXiv:2609.25041v1 Announce Type: new
Abstract: An equitable $k$-coloring of a graph partitions its vertex set into $k$ independent sets whose sizes differ by at most one; the least such $k$ is the equitable chromatic number $\chie(G)$. Every known bound on $\chie$ valid for all graphs, beginning with the Hajnal--Szemer\'edi theorem, is linear in the maximum degree $\Delta$, and the star $K_{1,\Delta}$, for which $\chie=\ceil{\Delta/2}+1$, shows that no general bound below $\Delta/2$ exists. We2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/882bee35dd911c63What Does Chain-of-Thought Entropy Measure? A Channel Audit of Scaffolding, Routing, and Content2026-09-23T04:00:00+00:00arXiv:2609.25039v1 Announce Type: new
Abstract: Entropy over chain-of-thought tokens decides which tokens receive the policy gradient, which get pruned, and whether a run has collapsed, yet each such statistic reads a next-token distribution mixing three choices: whether to emit connective scaffolding, which connective, and what the substantive continuation should be. Designating a scaffold vocabulary subset separates the three, exactly, for entropy, Kullback--Leibler divergence, and the first-2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/db991f3dcd47ea89A Vertex-Linear Threshold for Eventually Tur\'an good and the Cluster Method2026-09-23T04:00:00+00:00arXiv:2609.25038v1 Announce Type: new
Abstract: A graph $H$ is $K_{r+1}$-Tur\'an-good if, for every sufficiently large $n$, the Tur\'an graph $T_r(n)$ maximizes the number of copies of $H$ among all $n$-vertex $K_{r+1}$-free graphs, and it is strictly $K_{r+1}$-Tur\'an-good when this extremal graph is unique. Morrison, Nir, Norin, Rz\k{a}\.zewski and Wesolek proved that for every graph $H$ with at least one edge, when $r\ge 300v(H)^9$, $H$ is $K_{r+1}$-Tur\'an-good. They asked whether the above2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/cf6c49d7d96cd13eDiscrete Concavity of Token-Graph Spectral Radii via Lorentzian Semigroups2026-09-23T04:00:00+00:00arXiv:2609.25037v1 Announce Type: new
Abstract: Let $F_k(G)$ be the $k$-token graph of a finite graph with nonnegative edge weights, and let $A_k$ and $D_k$ be its weighted adjacency and degree matrices. For every $-1\leq\vartheta\leq1$, we prove that $k\mapsto\lambda_{\max}(A_k+\vartheta D_k)$ is discretely concave. Complement symmetry then makes this sequence nondecreasing up to the middle level. At $\vartheta=1$ and $\vartheta=0$, this gives the signless-Laplacian and adjacency spectral-radi2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/b9bacbebab7c09954DGS-JEPA: Temporally Compositional Joint-Embedding Prediction for Dynamic Gaussian Splatting2026-09-23T04:00:00+00:00arXiv:2609.25036v1 Announce Type: new
Abstract: Dynamic Gaussian Splatting provides an explicit representation of evolving 3D scenes, but existing approaches are primarily optimized for reconstruction, future-state generation, or rendering rather than for learning reusable predictive dynamics. We propose 4DGS-JEPA, a Gaussian-native joint-embedding predictive architecture for causal multi-horizon prediction over dynamic Gaussian scenes. The model uses a hierarchical scene-, motion-group-, and G2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/24b284be8ce56839KKT Stresses, Affine Moments, and Separator Flux in the Heilbronn Triangle Problem2026-09-23T04:00:00+00:00arXiv:2609.25035v1 Announce Type: new
Abstract: For n points in the unit square, the Heilbronn triangle problem asks for the largest possible minimum triangle area. We develop a variational stress theory for this max-min problem. At every positive-area local optimum, normalized Karush-Kuhn-Tucker multipliers assemble into a skew matrix B satisfying Bz = 2ib, where b is the outward square reaction. This equilibrium has the isotropic affine moment sum_i p_i b_i^T = Delta I_2; the identity also ho2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/c675b1ea5ed9dc99From Tone to Trajectory: Continuous Sentiment and the Shape of Monetary Policy Communication2026-09-23T04:00:00+00:00arXiv:2609.25034v1 Announce Type: new
Abstract: Central bank press conferences are not merely information releases --- they are structured narratives. We study whether the shape of sentiment within a statement, not just its average tone, carries policy-relevant signals. Constructing sentiment arcs for ECB and Fed press conferences along three dimensions --- monetary stance, economic outlook, and uncertainty --- we assess their predictive content for policy rate changes, inflation expectations, 2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/885d86495f5af81dBistationary Traces, Wide Levels, and Branch-Cover Rigidity for an Unrestricted Typed Variant of the Hayut-Magidor Forcing2026-09-23T04:00:00+00:00arXiv:2609.25033v1 Announce Type: new
Abstract: For every uncountable regular cardinal $\alpha$, $\mathbb S^{\ast}(\alpha)$ is an explicitly typed four-coordinate forcing motivated by the ladder-system construction of Hayut and Magidor. The forcing is $\sigma$-closed and, after adjoining a formal maximum, $\alpha$-strategically closed. For $\alpha\geq\omega_2$, every nonempty countable family of designated generic branches has a stationary and costationary common trace on the generic ladder-coo2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/e2564747e3d1f5dbQuadratic-Defect Completions of Spherical $2$-Design Orbits2026-09-23T04:00:00+00:00arXiv:2609.25032v1 Announce Type: new
Abstract: We study how spherical $2$-designs arising from finite group orbits can be completed to spherical $4$-designs by adjoining further orbits, allowing weights in the general theory. For an irreducible real orthogonal $G$-module $W$ with $\mathbb D=\operatorname{End}_G(W)\in\{\mathbb R,\mathbb C,\mathbb H\}$, we consider the multiplicity-two representation $W\oplus W$ and retain the failure of the $2$-design equation $M^*M=\frac12 I_2$ as a quadratic 2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/0a8bd992ad407f82Towards Adaptive Interaction Strategies for Human Companion Robot via Deep Reinforcement Learning2026-09-23T04:00:00+00:00arXiv:2609.25031v1 Announce Type: new
Abstract: In the field of Human-Robot Interaction (HRI), achieving flexibility in human-accompanying within real-world environments holds great potential for various applications but also poses significant challenges. Traditional methods typically restrict robots to fixed positions relative to humans, such as tracking from behind, in front, or side-by-side, which limits robot adaptability in dynamic workspaces. This study introduces a novel human-companioni2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/f208a80d005d2e2dSpectral extremal hypergraphs without long Berge cycles2026-09-23T04:00:00+00:00arXiv:2609.25030v1 Announce Type: new
Abstract: Let $r\ge 3$ and $k\ge 2r+1$ be fixed integers. We determine, for all sufficiently large $n$, the maximum adjacency-tensor spectral radius of an $n$-vertex $r$-uniform hypergraph containing no Berge cycle of length at least $k$. Write $s=\left\lfloor\frac{k-1}{2}\right\rfloor$. If $k=2s+1$ is odd, the unique extremal hypergraph consists of all $r$-sets containing at most one vertex outside a fixed $s$-set. If $k=2s+2$ is even, one additionally inc2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/3ea5330d62304953Low-Twist Matrix Covariants of Exterior Powers: Vanishing and Modular Phenomena2026-09-23T04:00:00+00:00arXiv:2609.25029v1 Announce Type: new
Abstract: We study $GL(V)$-equivariant morphisms between the polynomial representations $\operatorname{Sym}^d(\Lambda^r V)$ and $\operatorname{End}(V)\otimes\det(V)^\ell$. Scalar matrices impose the necessary condition $rd=n\ell$, which singles out the first two determinant twists. In characteristic zero we determine the second-twist spaces for trivectors in the moving family $d=2m$ and $n=3m$. Equivalently, for every $m\geq 2$ we prove $\langle h_{2m}[e_3]2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/c6ffe1658499554fRetrieved-Span Training for Efficient Query-Focused Meeting Summarization on QMSum2026-09-23T04:00:00+00:00arXiv:2609.25028v1 Announce Type: new
Abstract: QMSum provides no scorer, making query-focused meeting summarization results difficult to compare. We rescore or generate 15 systems under one implementation. Through a common inference port, a released 406M Fusion-in-Decoder specialist loses 6.30 ROUGE-1 when moved from capped long input to 2,000-word retrieved spans. Fine-tuning it on this span regime recovers the loss. On test it scores 36.33 ROUGE-1 versus 35.41 for our 1.2B system; the meetin2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/fc53b2101af6f271Fixed Convex-Lens Spectral Constants: M\"obius Reduction, Sharp Model Theorems, and Angle-Dependent Bounds2026-09-23T04:00:00+00:00arXiv:2609.25027v1 Announce Type: new
Abstract: For the intersection of two disks meeting at angle $2\alpha$, let $C(\alpha)$ be the least constant in the associated spectral-set inequality. We give a self-contained M"obius reduction to the corresponding numerical-range problem on a sector and determine the sharp constant for affine square-zero operators $B=\lambda I+N$, $N^2=0$: $C_{\mathrm{sq0}}(\alpha)=\pi\sin\alpha/(2\alpha)$. A $2\times2$ matrix attains equality and yields an explicit lens2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/4cd14c8b4b3648b4Local Expected Utility in Infinite-Dimensional Spaces: Theory and Elicitation2026-09-23T04:00:00+00:00arXiv:2609.25026v1 Announce Type: new
Abstract: This paper develops a decision-theoretic interpretation of local expected utility on Abstract Wiener space. The framework is intended for economic acts that are naturally stochastic paths, such as income paths, consumption paths, investment payoffs, insurance losses, or laboratory stimuli unfolding over a continuum of contingencies. In this setting the object of choice is an act or payoff path, while Hermite functions serve as state-feature coordi2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/62f36429206e6f1dString-Monodromy Rigidity from Six-Point Consistency2026-09-23T04:00:00+00:00arXiv:2609.25025v1 Announce Type: new
Abstract: Can the characteristic monodromy of string scattering be derived from S-matrix consistency alone, without assuming a worldsheet? We show that finite-particle consistency fixes it to a sharply defined extent. In a massless doubly ordered identity model, two exact six-point residues generate a scalar pentagon and the odd Fay identity, forcing every analytic four-point null kernel to satisfy $q''=\kappa q$ and leaving only the linear, trigonometric, 2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/677d88f4f56f1900Pushout Attachments and Conditional Complexity of Executable Models2026-09-23T04:00:00+00:00arXiv:2609.25024v1 Announce Type: new
Abstract: We represent the modular enlargement of an executable model by a pushout of finite typed presentations. Adhesivity preserves the original model and recovers the shared interface, while chosen pushouts make linking functorial and coproducts describe parallel attachments. A semantic comparison gives a precise criterion for behavioral persistence. Effective linking also bounds conditional prefix complexity; under the stated canonicalization and invar2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/4ab2bebc21694584Idempotent-free non-solvable evolution algebras over $\mathbb{C}$2026-09-23T04:00:00+00:00arXiv:2609.25023v1 Announce Type: new
Abstract: A recent conjecture states that a finite-dimensional complex evolution algebra is solvable if and only if it has no non-zero idempotents. We exhibit a three-dimensional counterexample over $\mathbb{C}$ whose isomorphism class already appears in the classification of three-dimensional complex evolution algebras. Since the conjecture is known in dimensions one and two, this counterexample has the smallest possible dimension. The algebra is defined o2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/c920f8e0d235611dNPLSD: Accelerating Line-Segment Detection on NPU Microcontrollers2026-09-23T04:00:00+00:00arXiv:2609.25022v1 Announce Type: new
Abstract: Line-segment detection is fundamental to robotics, autonomous navigation, and industrial inspection. While transformer-based detectors achieve the highest accuracy, their deployment on microcontrollers remains impractical due to resource constraints. The STM32N6, with its Neural-ART NPU, promises to enable deep vision at the extreme edge. However, existing detectors rely on attention, grid-sampling, and normalization, operators that are unsupporte2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/71ac5a164a1de7a9"As a Language Model...": Chat Template Switches LLM Self-Referential Voice and Activation Steering Reproduces It2026-09-23T04:00:00+00:00arXiv:2609.25021v1 Announce Type: new
Abstract: Large Language Models (LLMs) tend to add disclaimers like "I'm just an AI" when asked about something related to themselves. The self-reports from such responses are used in debates about AI safety or self-knowledge of the models, yet what drives them is not well understood. Are the models telling us about themselves or rather how they are deployed? In this work, we show that the chat template works like a switch - when present, it turns this disc2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/fe792ec150071043Compressibility is not Feedback: A Random-access Gap in Causal Semantic Repair2026-09-23T04:00:00+00:00arXiv:2609.25020v1 Announce Type: new
Abstract: Semantic feedback can repair a noisy first decode without retransmitting the whole message, but current end-to-end evaluations hide where a feedback design fails. An oracle that sees both the source-side state and receiver observation may describe a useful correction with few bits, while the receiver, which sees only its own observation, may not know which correction to request. We formulate these as distinct direct and receiver-constrained coding2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/42b17cb7a21b3a24One-Bin Fourier Challenges for Dimension-Free Reconstruction Certification2026-09-23T04:00:00+00:00arXiv:2609.25019v1 Announce Type: new
Abstract: Partial-Fourier measurement underlies computational imaging, and its reconstructions increasingly come from iterative or learned solvers whose recovery guarantees are conditional on a signal model, a sampling law, and solver accuracy. None of those guarantees transfers to a particular committed output: it can satisfy every acquired coefficient while remaining badly wrong in the unmeasured nullspace. Native Fourier holdout does not repair this, sin2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/f117f05a2b7e95f7OSFoundry: Building and Evolving Operating Systems with Specification-Guided Agents2026-09-23T04:00:00+00:00arXiv:2609.25018v1 Announce Type: new
Abstract: Operating systems must evolve continuously. Yet their development remains code-centric and largely manual: even a localized change can require recovering implicit assumptions, coordinating multiple subsystems, and repeatedly building, booting, testing, and debugging the complete system. General-purpose coding agents automate individual edits, but their prompt-centric workflows repeatedly reconstruct task boundaries and OS semantics from scattered 2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/0fe284f8647aa5adDeepfakes and Synthetic Media: Generation, Detection, and Governance2026-09-23T04:00:00+00:00arXiv:2609.25017v1 Announce Type: new
Abstract: Deepfakes, synthetic audiovisual content produced by deep generative models, have escalated into a critical threat across civilian and military domains, enabling identity fraud, disinformation campaigns, and evidence fabrication. In high-stakes environments, ranging from journalism and finance to healthcare and legal contexts, the consequences extend to severe misinformation, market manipulation, identity fraud, and the erosion of institutional tr2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/0fae2c28a661aa08The domain question for the Nielsen-Soelberg group rings: the commutative case, with certified ball checks2026-09-23T04:00:00+00:00arXiv:2609.25016v1 Announce Type: new
Abstract: Nielsen and Soelberg exhibited three torsion-free groups G_1, G_2, G_3 carrying 8-element sets without unique products, and asked whether any of the group rings R[G_i], R a domain, is a domain. We record that for every commutative domain R the answer is affirmative for all three groups: each G_i is virtually nilpotent, so the theorem of Kropholler, Linnell and Moody applies over every field, in every characteristic, and commutative coefficients re2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/bd5b2d01d09f02f2A compactness theorem for twisted-unitary elements of integral group rings, with a certified route to the theta-unitary Case A at window B(4) of the Promislow group2026-09-23T04:00:00+00:00arXiv:2609.25015v1 Announce Type: new
Abstract: Let G be a torsion-free group whose real group algebra R[G] has no zero divisors, and let u -> u^{*theta} be an l2-isometric anti-involution of R[G] (a composition of the inversion involution with a ring automorphism and a sign character). We prove a compactness theorem: for every finite *theta-closed support window W the constant mu*(W) = min{ ||w^{*theta}w||_2 : ||w||_2 = 1, supp(w) in W } is strictly positive, and every real theta-unitary eleme2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/6d734ad801dbaedeNot All 4-bit Quantizers Are Equal: Deployment-Time Mitigation of PII Leakage in Fine-Tuned Small Language Models2026-09-23T04:00:00+00:00arXiv:2609.25014v1 Announce Type: new
Abstract: Organizations fine-tune small language models on private data and then compress them to 4 bits for resource-efficient deployment. We show that the compression method also affects privacy. What separates the methods is not the bit width but whether they tune their rounding on a small sample of text, the calibration corpus. On our primary model, when each planted record's own opening text is used as the prompt, the two calibration-based methods we t2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/9ade661cec5e297eDo Existing Preconditioners Improve Biomedical Tabular Foundation Learning? An Empirical Study on TabPFN Optimization2026-09-23T04:00:00+00:00arXiv:2609.25013v1 Announce Type: new
Abstract: Tabular foundation models have recently shown strong potential for structured biomedical data analysis. Among them, TabPFN has emerged as an effective approach for low-data tabular classification tasks. However, the impact of optimization and preconditioning strategies on biomedical fine-tuning remains largely unexplored. In this work, we present a comprehensive empirical investigation of five AdamW-based preconditioning strategies for fine-tuning2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/e5efb0456f9a3397A Computational Approach to Measuring Semantic Change in Sanskrit Literature2026-09-23T04:00:00+00:00arXiv:2609.25012v1 Announce Type: new
Abstract: Diachronic word embeddings have become the modern standard for tracking semantic change, yet they have been largely validated on modern, high-resource, and well-segmented languages. This paper tests whether the paradigm transfers to Sanskrit, an ancient, low-resource language whose phonological fusion (sandhi), morphological inflection, compounding, and polysemy pose a unique challenge. I assemble a 2.7M-token corpus spanning four canonical period2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/ccd75c1bb6188fadA Kernel-Certified Verification of the Erd\H{o}s-Mollin-Walsh Conjecture below $10^{14}$2026-09-23T04:00:00+00:00arXiv:2609.25011v1 Announce Type: new
Abstract: Erd\H{o}s problem 364 asks whether three consecutive powerful numbers exist, where $n$ is powerful if $p \mid n$ implies $p^2 \mid n$. Erd\H{o}s (1976) and, independently, Mollin and Walsh (1986) conjectured that none do; the $abc$ conjecture implies at most finitely many. The conjecture remains open. We present the first verification of the conjecture at any finite bound that is checked end to end by a proof kernel: machine-checked theorems in Le2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/53b8ae5109253f0bDo Synthetic Personas Predict Real Audience Response? A Sim-to-Real Study Where a No-Persona Baseline Beats Persona-Based Copy Simulation2026-09-23T04:00:00+00:00arXiv:2609.25010v1 Announce Type: new
Abstract: Marketers increasingly use large language models (LLMs) as "synthetic personas" to predict how an audience will react to a piece of copy before it ships, encouraged by evidence that profile-conditioned LLMs mimic human samples. But is that prediction actually valid against real behaviour - and does the persona machinery help? We present a sim-to-real validity study using the Upworthy Research Archive - thousands of headline A/B tests on shared rea2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/0eb2b4f3b50f8559Same Quantity, Different Answer: Numerical Representation Invariance in Language Models2026-09-23T04:00:00+00:00arXiv:2609.25009v1 Announce Type: new
Abstract: Numerically equivalent word problems should yield the same canonical answer whether a quantity is written as a decimal, fraction, percentage, number word, scientific notation, or an exactly converted unit. We generate 3,600 exact-rational problems and 8,600 prompts spanning five identity-preserving transformation families, and evaluate five open-weight systems. After a fixed syntax audit that normalizes common answer forms without an LLM judge, ca2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/10c04bf66fd1972dTraining a Language Model End-to-End in Rust: An Experience Report2026-09-23T04:00:00+00:00arXiv:2609.25008v1 Announce Type: new
Abstract: I pretrained a language model end-to-end in Rust - alone, with no team, no PyTorch, and no Python in the training path - for $164 in rented GPU time. I report that as an achievement, not a recommendation: the more useful contribution is a measured failure taxonomy of the two leading Rust ML frameworks, Candle and Burn, as training (not inference) backends in 2026. I document five Candle defects, including fused kernels that silently produce no gra2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/d46db403eb6f0accBeyond Short Segments : Expanding Speaker Embeddings with Vector Archives2026-09-23T04:00:00+00:00arXiv:2609.25007v1 Announce Type: new
Abstract: The performance of state-of-the-art speaker verification (SV) systems severely degrades on short utterances due to insufficient speaker-specific information. To address this critical challenge, we propose the Vector Archive Mapping ECAPA (VAM-ECAPA), a novel system designed to enhance feature extraction from short-duration speech. The core of our system is the Transformer-based Vector Archive Mapping with Statistical Pooling (TVAMSP) module, which2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/d8d827e850e61511What Does 99% Accuracy Measure? A Reproducible Audit of Shortcut Learning in a Widely Used Fake News Corpus2026-09-23T04:00:00+00:00arXiv:2609.25006v1 Announce Type: new
Abstract: Text classifiers trained on the ISOT/Kaggle "Fake and Real News" corpus routinely report accuracy and F1 above 0.98, a level of performance that sits uneasily beside the difficulty of assessing veracity. Using a transparent TF-IDF and linear-classifier pipeline as a measurement instrument, we audit the corpus along three leakage channels and two distribution-shift protocols, releasing all code and derived numbers. First, the benchmark is partly de2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/80097a7168e64a12Effects of High-Temperature Preconditioning on Li Plating During Low-Temperature Cycling of the COTS LG Chem HG2 Cell2026-09-23T04:00:00+00:00arXiv:2609.25005v1 Announce Type: new
Abstract: Commercial off-the-shelf~(COTS) Li ion cells offer an attractive and rechargeable power source for spacecraft applications. However, since these cells are typically designed for charging above 0\,\textdegree C, challenging sub-zero temperatures could inflict irreversible degradation and safety hazards during in-flight operation. To address these challenges, this work explores strategies affecting their low-temperature performance. Specifically, we2026-09-23T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/333c2f1ae3c7d141Higher Quality Small Synthetic Natural Language Text Generation for Interpretability Research2026-09-23T02:47:30+00:00Introduction Small simple synthetic natural language datasets suitable for end-to-end training of tiny LLMs serve as an important resource for LLM interpretability researchers. Some well know examples include roneneldan/TinyStories , SimpleStories/SimpleStories , and klusai/ds-tf1-en-3m (TinyFabulist). This post solves key problems that degrade the quality of these datasets, while also offering an efficient accessible pipeline that can be run locally on an NVIDIA 5060 Ti (16GB) graphics card. Th2026-09-23T02:47:30+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/e43742c750656b3dSignals of Slop: How to identify and avoid creating slop2026-09-23T00:34:31+00:00Slop isn't limited to low-effort AI-generated content. Humans can also create slop. This summary lists signals of slop : traits that make something slop, independent of the level of effort spent or the degree of AI involvement. Repetition People often identify artifacts exhibiting popular patterns as slop, regardless of whether they consider those patterns intrinsically good or bad. Repetitiveness is considered a signal of slop for at least three reasons: People experience fatigue from popular c2026-09-23T00:34:31+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/47d44ecb96e88b2dSignals of Slop2026-09-23T00:34:31+00:00Slop isn't limited to low-effort AI-generated content. Humans can also create slop. This summary lists signals of slop : traits that make something slop, independent of the level of effort spent or the degree of AI involvement. Repetition People often identify artifacts exhibiting popular patterns as slop, regardless of whether they consider those patterns intrinsically good or bad. Repetitiveness is considered a signal of slop for at least three reasons: People experience fatigue from popular c2026-09-23T00:34:31+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/ad56e86e1a6694f2An unexamined cause of the OpenAI Hugging Face hacking incident:
its binary performance metric2026-09-23T00:27:26+00:00We argue that a main cause of the OpenAI Hugging Face incident was overlooked: the overly simple evaluation metric in ExploitGym was misaligned. Further, techniques already exist that can mitigate such misalignment in the future. In July 2026, OpenAI was testing the ability of its language models to exploit software vulnerabilities using a benchmark called ExploitGym . In ExploitGym, each test presents an agent with software containing a known vulnerability and tasks it with capturing a secret s2026-09-23T00:27:26+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/821fb5c37bd28d6eFusion Energy Projects are Not Trying to Imitate the Sun2026-09-22T22:57:35+00:00Post Intro: The quest to use fusion power for break-even electricity generation has sometimes been described as “putting the sun in a jar”. Where by “jar,” we mean a carefully arranged series of coils that produce a powerful confining magnetic field. Usually it’s called a tokamak (toroidal design) or stellarator (complicated twisty design that is still topologically a toroid). The sun is so enormous that confinement happens as a natural result of its own gravitational field. On the other hand, t2026-09-22T22:57:35+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/b3210e8e2f390113AI: artificial immigrants2026-09-22T22:44:08+00:00Advanced AI is basically the embodiment of immigration as envisioned in the conservative nightmare: We are letting a bunch of new agents into our society They don’t clearly share our values and we suspect a society full of them would be awful by our lights But we expect them to provide very cheap labor Which will undercut local wages and leave locals unemployed They will probably gain power and influence over time—in the economy, politics and culture—and end up controlling everything, sidelining2026-09-22T22:44:08+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/51e6c6078c183f81A plan for a perpetual motion research center that runs forever2026-09-22T21:03:23+00:00There is a popular tradition on this website of clarifying complex ideas through somewhat heavy handed, metaphorical dialogues. TO WIT, Alice says “I don’t have a good lead on how to build a perpetual motion machine, but I want to find one and am rich as all hell. Here is my plan for a research center that will study perpetual motion forever that won't have any power hookups or solar panels, because it's a clever research center that does not require any external fuel” Bob says "One of your sent2026-09-22T21:03:23+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/9dd4f767e9998363Announcing B-Side Labs: Measuring Character (Seeking Collaborators and Testers)2026-09-22T19:55:00+00:00tl;dr Rapid AI adoption means that models are increasingly becoming autonomous decision-makers embedded in high-stakes systems. However, frontier models lack stable character, abandoning their designated personas or factual truth under social pressure. B-Side Labs builds a science of AI character under pressure by designing discriminative evaluations, real-time drift detection, and interventions to ensure model character remains stable. Our first tool, Virtue Council , is live with pilot results2026-09-22T19:55:00+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/9032f0b4ae93339eTrading firms could control meaningful amounts of compute by 20302026-09-22T18:52:17+00:00It's pretty crazy that right now, the highest margin thing to do with these models seems to be simply selling them through an API. Dwarkesh's blog prize [1] questioned how this dynamic could ever result in lab profitability, simply because the scale of reinvestment into model training and research requires constantly reinvesting more than you're making. Well, Anthropic is likely already profitable, [2] and it hasn't required any of the schemes I saw proposed in answers to his question. It turns 2026-09-22T18:52:17+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/6762897078b31697Premature Optimization2026-09-22T17:03:31+00:00Your code doesn't work, but you're spending the majority of your time debating new implementations details in order to achieve maximal performance. You're finally deciding to address your longstanding problem of missing deadlines, but there are so many calendar apps out there, and it's oh-so-hard to decide which one to use. You definitely need to do more research before committing. You buy a self-help book and really enjoy it. If you could internalize even half of what that book suggested, you c2026-09-22T17:03:31+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/d1b6110a63e8f3e9Introducing Opus 5.5: Anthropic Linkpost2026-09-22T16:50:02+00:00https://www.anthropic.com/claude-opus-5-5 It's a sizeable upgrade: Also, the first model in which they say this: Pacing the frontier Last week, our CEO, Dario Amodei, argued that AI progress should be paced so that safety practices stay ahead of model capabilities. Pacing is an approach to keeping AI safe, remaining competitive with China, and realizing AI’s benefits, particularly in areas like biology and medicine. We largely understand the risks today’s models present and are well equipped to 2026-09-22T16:50:02+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/680966ce005dc3adInitial DIY Cleanroom Experimentation2026-09-22T14:10:58+00:00In It
May Be Possible to Improvise A High Grade Bioshelter , Adin
Richards discusses the possibility of improvising defenses against an
environmental threat such as mirror
bacteria . He gives an exploratory overview of why it might be
possible to apply materials and equipment people often already have in
their houses to pressurize all or part of a house with filtered air.
It would be great if this were possible, but with all the ways for an
improvised system to fail I'm pretty skeptical. I deci2026-09-22T14:10:58+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/2098491908ba900dModern LLMs have tiny GPTs hidden inside them2026-09-22T13:33:10+00:00Experiments into predicting GPT2 completions via Qwen models This is a crosspost from my substack (where I do varied tiny experiments on LLMs and agents). It's also part of Lossfunk , where we're investigating meta-cognition in LLMs as one of the projects. ---- Next token prediction is a magical objective. To predict the correct token in such a vast variety of texts present in the pretraining corpus, the model must infer a tremendous amount of hidden and latent causes that generate that text. On2026-09-22T13:33:10+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/19f03bf679e1e9acControllable-CoT leads to covert reasoning capabilities2026-09-22T13:04:11+00:00Summary I measure GPT-6 Astra’s performance on multi-hop tasks when prompted with a secondary CoT-control instruction: to reason using only dots, or to reason steganographically. Astra demonstrates covert reasoning capabilities with task performance beating that when using no reasoning or filler tokens for reasoning. This work agrees with findings from Astra is much better at reasoning with filler tokens than previous models but has the model generate its own reasoning and provide it as part of 2026-09-22T13:04:11+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/6715476f98f9e77fPolitics Gets Interested In Those Trying Not To Die2026-09-22T13:00:53+00:00This was the month the world took notice that AI might kill everyone. Jacob Coxon’s resignation set off a preference cascade . Anthropic CEO Dario Amodei wrote that we must pace the frontier . Sam Altman, Elon Musk and Demis Hassabis agreed. We were filled with hope. Perhaps we could agree to some basic safety measures, starting with embedded evaluators, pass some basic regulations and guardrails and otherwise start to act sensibly. Politicians on both sides took notice and were saying sensible 2026-09-22T13:00:53+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/aff9d0e7c313f980Autonomous Evidence Factories: Safe and Useful Recursive Self-Improvement2026-09-22T12:28:32+00:00This post is crossposted from my Substack, Structure and Guarantees , where I explore how formal verification and related ideas might scale to more complex intelligent systems. Here I propose an approach to aligning recursively self-improving AI: confine its reward function and meta-level world model to mathematically precise semantics, with no representation of humans or the wider world as means to achieving its goals. Its only intended external effect is delivering solutions to well-specified 2026-09-22T12:28:32+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/085e76d6a7517fa9Projecting AI Automation at Anthropic2026-09-22T11:29:25+00:00Anthropic recently released some very interesting information about the degree of AI automation for R&D tasks. I recommend reading the entire article: Measurements for understanding the pace of AI development inside frontier labs (Sep 17, 2026). In it you will find this graph, depicting the results so far for Anthropic's R&D Automation Index, using a scale developed by Epoch AI: First of all, I’d like to thank Anthropic for sharing this information. Tracking things like this, and making the resu2026-09-22T11:29:25+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/a912167aa32e69d8Do we agree what "short timelines" means?2026-09-22T10:16:27+00:00During a discussion on one of the various LessWrong-aligned Telegram chat groups, I noticed how vague the "short" in "short timelines" is, and ran a poll to find out what people mean when they say "short timelines". % Range Votes 0% <3 months 0 0% 3–6 months 0 33% 6–12 months 8 33% 1–2 years 8 21% 2–4 years 5 8% 4–8 years 2 4% Something more than 8 years 1 Total: 24 votes Small n, certainly. But with this broad a distribution even on a time axis that doubles with each line, I think it shows we n2026-09-22T10:16:27+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/64d88c52c104ee52Total Safety Transparency?2026-09-22T05:39:39+00:00The AI safety movement should push itself to be dramatically more transparent to the public. To date, the AI safety movement has been one of the strongest forces for clarity and wisdom in the world. The movement has been prescient on the subject of concerns from existential risk, seriously grappling with outcomes others dismissed as sci-fi nonsense. Society is now waking up to the potential threats of advanced AI. I understand that many in the movement are feeling the crunch, and thinking more c2026-09-22T05:39:39+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/21bd851429a70333Math contains a lot of stuff2026-09-22T05:39:12+00:00AI is getting good at math lately . Maybe we we will soon see AI that is much better than humans on all math tasks ("Math ASI"). What could this mean? There are at least three maths: Math A (Ancestral) is when you study natural mathematical objects. To be considered natural, there has to be motivation from the real world, other math, or intuition. A big part of this is proving theorems. Math B (Benchmarky) is when you resolve conjectures from Math A. This is often used for benchmarking academics2026-09-22T05:39:12+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/02f60467ffdb61e3When talked into harm, a model blames the answer, not itself (an interpretability study of guilt vs shame)2026-09-22T04:43:09+00:00This was my write-up for Neel Nanda’s Winter 2027 MATS Stream (~20h research task). I didn’t get in, but it was my first application, so there’s always next time :P Lightly restructured here to fit the LessWrong format better. Repo: https://github.com/star2vec/guiltea ── ⋆⋅☆⋅⋆ ── TL;DR: Models are safety-trained, but they can still be persuaded to do harmful acts. When that happens, how does blaming or informing it of its mistake influence its understanding of itself, its role, and subsequent a2026-09-22T04:43:09+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/ed2e0aec76bf058dWhen Is Availability-Aware Training Worth It? A Benchmark and Empirical Study of Interruption-Resilient Optimization Under Predictable Compute Schedules2026-09-22T04:00:00+00:00arXiv:2609.22087v1 Announce Type: new
Abstract: Training under non-stationary but predictable compute availability (satellites under eclipse, duty-cycled edge devices, power-capped datacenters) is often framed as needing specialized, availability-aware optimizers. We test that premise. We release OrbitTrace, a benchmark of 50 physics-grounded availability traces from SGP4 propagation of live two-line element sets across three orbital regimes, and ask a falsifiable question: when an availability2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/d3bb9befabe6712dLearning Dynamic Neural Evidence Representations for Time-Adaptive Brain-Computer Interfaces2026-09-22T04:00:00+00:00arXiv:2609.22088v1 Announce Type: new
Abstract: Brain-computer interfaces (BCIs) decode neural activity into commands, yet most existing systems rely on fixed-window decoding that may result in redundant observation or unreliable predictions due to insufficient evidence. Adaptive temporal decision-making (ATDM) addresses this accuracy-time trade-off by progressively accumulating EEG evidence and deciding when to stop. However, existing EEG encoders are mainly designed for fixed-window decoding 2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/27a433560c58b790On product spectral sets and functional tiles in $\mathbb{Q}_p^d$2026-09-22T04:00:00+00:00arXiv:2609.22089v1 Announce Type: new
Abstract: This paper studies product spectral sets and functional tiles in the $p$-adic spaces $\mathbb{Q}_p^d$ within the framework of the {\bf product spectral set conjecture}. We establish a stability result for functional tiles under weak convergence of tiling complements, characterize product spectral pairs with product spectra, and fully resolve the conjecture for cylindric sets by proving that spectrality of a cylindric set $\Omega=B_\gamma(a)\times\2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/546c8c749a0ad472Recognition, Simulation, and Refusal: A Contamination-Aware Study of Classic Psychological Effects in LLM Agents2026-09-22T04:00:00+00:00arXiv:2609.22090v1 Announce Type: new
Abstract: An LLM producing the response pattern associated with a human psychological effect is not the same claim as the LLM possessing that bias. We present PsyAgentBench, a benchmark that re-runs classic psychology experiments on LLM agents under a factorial design built to separate these: each paradigm is run with the paradigm explicitly labeled in the prompt (named) or framed as a routine task (blind), and on the literal textbook version of the task (c2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/d47c729b07b64d91Memory That Looks Forward: A Zero-Inference Prospective Term for Personal Memory Retrieval2026-09-22T04:00:00+00:00arXiv:2609.22091v1 Announce Type: new
Abstract: Retrieval over a personal memory store is retrospective: it surfaces what resembles the query, and it is blind to what the user has committed to do. We describe a prospective term for memory retrieval that costs no inference at query time. Commitments are held in an explicit ledger as dated or trigger-conditioned entries; memory items linked to a firing entry receive a salience boost, blended multiplicatively into embedding-based retrieval so that2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/c127eca995e845a3Leakage-Safe Empirical Benchmarking of EEG-Based Machine Learning Pipelines for Dementia Classification2026-09-22T04:00:00+00:00arXiv:2609.22092v1 Announce Type: new
Abstract: Electroencephalography (EEG) is a low-cost and non-invasive signal source for dementia screening, yet existing EEG-based studies remain difficult to compare because preprocessing, EEG segmentation, feature design, classifier choice, and validation protocols vary across studies and are often evaluated in isolation. This variability limits the derivation of robust pipeline recommendations. This paper presents a leakage-safe empirical benchmark for r2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/52f5012cc000e8ddDC-CLM: Extending the WECC Composite Load Model for AI Data Center Dynamics2026-09-22T04:00:00+00:00arXiv:2609.22093v1 Announce Type: new
Abstract: The rapid growth of AI-driven data centers is introducing load behaviors that are not explicitly represented in conventional composite load models. This paper presents DC-CLM, a workload-aware extension of the WECC composite load model that incorporates UPS-supported IT demand, mixed motor/VFD cooling loads, auxiliary demand, and training, inference, and idle workload profiles. A rule-based supervisory state machine represents grid, battery, and d2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/89c536779b731ffeSummarize, Judge, Refine: Decoupled Content Understanding and Policy Learning for Multimodal Content Moderation2026-09-22T04:00:00+00:00arXiv:2609.22094v1 Announce Type: new
Abstract: Content moderation systems traditionally entangle multimodal understanding with policy-specific classification, requiring full pipeline retraining for every policy change and suffering from label scarcity since multimedia cannot be meaningfully augmented. We propose Summarize-Judge-Refine (SJR), a two-model architecture that decouples these concerns via a natural language interface: a multimodal Content Model produces structured text summaries, an2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/1a4abab0cc4f6a09Beyond the Raw Waveform: Fusing Visual Representations of EDA for Stress Detection2026-09-22T04:00:00+00:00arXiv:2609.22095v1 Announce Type: new
Abstract: Electrodermal activity (EDA) is widely used in automatic stress detection, yet most pipelines treat it only as a raw one-dimensional waveform. This study examines whether complementary visual representations of EDA provide useful information for stress classification and whether their fusion im- proves recognition performance. Six image-based representations are derived from each EDA recording: an unwrapped short-time Fourier transform (STFT) phas2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/648c35bc69ec9d60AI-inferred expressed well-being and collective-action discourse in climate-change campaigns on X2026-09-22T04:00:00+00:00arXiv:2609.22096v1 Announce Type: new
Abstract: Climate campaigns are often evaluated through attention and mobilization, but less is known about the well-being language that accompanies them. Whether campaign periods alter positive affect and hope, and whether happiness aligns with action language, remains unresolved. We analysed 364,118 public Twitter/X posts from Earth Day, Earth Hour, Global Climate Action Day and World Environment Day in 19 occurrence-years, using 30-day pre-event, event a2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/c65cefea54716855Token Signatures of Code: Comparing Coding Behaviors Across Large Language Models2026-09-22T04:00:00+00:00arXiv:2609.22097v1 Announce Type: new
Abstract: The evaluation of large language models (LLMs) on coding tasks has primarily focused on performance metrics such as pass@k. As LLMs continue to advance, many models now meet baseline performance requirements, reducing the discriminative power of performance-based evaluation alone. Yet a key question remains largely unexplored: how do LLMs differ in their coding behavior? We propose CLIC (Code Learning for Identification and Comparison), a visual a2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/0bdc68dce7881432TreeSpark: Calibrated, Load-Adaptive Draft Trees for Semi-Autoregressive Speculative Decoding2026-09-22T04:00:00+00:00arXiv:2609.22098v1 Announce Type: new
Abstract: Speculative decoding accelerates language-model inference by letting a cheap drafter propose tokens that the target model verifies in parallel. Recent block drafters make drafting nearly free: a single backbone pass emits an entire block of draft tokens. Draft trees promise a further gain -- several alternative continuations verified in one target forward -- but existing constructions rank candidates by per-position marginals that ignore which par2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/b04166ac7567585eA framework for recipe data structure with applications for culinary and nutritional insights2026-09-22T04:00:00+00:00arXiv:2609.22099v1 Announce Type: new
Abstract: Cooking is a complex process that transforms raw ingredients into delicious and nutritious dishes, yet the recipes that encode this process remain largely free text; readable by people but not directly computable. Existing recipe collections capture fragments of this information, but no shared representation links a recipe's structured ingredient composition, its geo-cultural provenance, and its nutritional profile within a single queryable schema2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/7671f7b0dde3996aAdaMem: Adaptive Memory Token Allocation for Soft Compression in Retrieval-Augmented Generation2026-09-22T04:00:00+00:00arXiv:2609.22100v1 Announce Type: new
Abstract: Retrieval-augmented generation (RAG) improves language models with retrieved evidence, but processing many long passages is costly and can introduce distracting information. Soft compression addresses this challenge by encoding passages as compact sequences of continuous memory embeddings before generation. However, existing methods typically assign each retained passage an identical number of memory embeddings, irrespective of its query-specific 2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/773f6ccd78d7c7d5Context Poisoning as Extreme-Value Attention Interference in Long-Context Language Models2026-09-22T04:00:00+00:00arXiv:2609.22101v1 Announce Type: new
Abstract: Large language models can process increasingly long prompts, yet their ability to locate and use decisive evidence may degrade as irrelevant or confusable context is added. We formulate this phenomenon, which we call context poisoning, as extreme-value interference in attention: the decisive-evidence score is upper-bounded, while the maximum score among effective distractors grows with their number. Under a softmax retrieval abstraction, we derive2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/34ac9b74dd233886When Who You Are Can Change the Code You Get: A Study of Persona-Induced Bias in LLM Code Generation2026-09-22T04:00:00+00:00arXiv:2609.22102v1 Announce Type: new
Abstract: Large Language Models (LLMs) are widely used as programming assistants, yet it remains unclear whether and how user's demographic information impacts the technical quality of generated code. We conduct a large-scale empirical study of persona-induced bias in LLM-based code generation, focusing a proprietary model (Gemini 2.5 Pro) and an open-weight model (GPT-OSS-120B). Using 18 demographic personas spanning nationality, gender, and experience lev2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/1df4e1a997341204Graph Learning for Cross-Subject, Cross-Population EEG Emotion Decoding and Model-Derived Spatial-Spectral Neural Signatures2026-09-22T04:00:00+00:00arXiv:2609.22103v1 Announce Type: new
Abstract: Electroencephalography (EEG) provides a noninvasive means of capturing emotion-related neural dynamics, yet reliable EEG emotion decoding lacks models that can both generalize to unseen individuals and populations while preserving neural interpretability. To address these challenges, EmoDiPyraTrans is proposed as a development-regularized differential graph Transformer that models temporally ordered relative power spectral density graphs through a2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/d7a8657aeff971a5DeepInstructor: An Agentic AI Instructor for Experience-Driven Idea Evaluation2026-09-22T04:00:00+00:00arXiv:2609.22104v1 Announce Type: new
Abstract: As automated scientific discovery advances, Large Language Models (LLMs) can now generate research ideas at an unprecedented scale, shifting the bottleneck from idea generation to idea evaluation. Existing evaluators mainly rely on parametric LLM knowledge or unstructured retrieval, producing judgments that lack the experience-grounded reasoning used by human instructors. To address this, we propose DeepInstructor, an agentic framework that formul2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/89e2405b2e29a1a5Comparative Analysis of State-of-the-Art Foundation Models for Sleep Analysis Under Channel Reduction2026-09-22T04:00:00+00:00arXiv:2609.22105v1 Announce Type: new
Abstract: Automatic sleep staging from polysomnography (PSG) is a well-studied task, but PSG itself is expensive, clinic-based, and burdensome to manually score, which limits its use for long-term or at-home monitoring. Most existing sleep-staging foundation models are evaluated using the full PSG montage. We instead ask how much of that montage is actually necessary. We evaluate six sleep staging models on the Multi-Ethnic Study of Atherosclerosis (MESA) P2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/e1e626b734a7effcPRQuant: Permutation Residual Quantization for Low-Overhead Inference2026-09-22T04:00:00+00:00arXiv:2609.22106v1 Announce Type: new
Abstract: Accuracy of Low-bit quantization of linear layers is often dominated by a small number of outliers. Although existing methods, such as smoothing, rotation, or residual-based approaches, may mitigate this problem, they often introduce new accuracy bottlenecks to weights. Besides, most of these techniques are implemented as online approaches, which can result in heavy execution overheads. To address the afore-mentioned issues, We propose PRQuant (Pe2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/ed923dab1e7f54bdGeneralized Multimodal Foundation Model2026-09-22T04:00:00+00:00arXiv:2609.22107v1 Announce Type: new
Abstract: Making prediction with multimodal data is widely used in diverse scenarios. Existing multimodal fusion models, once deployed, can only handle predefined modalities (e.g., vision, text and audio) and single tasks, making it difficult to quickly adapt to new downstream applications. Therefore, a natural yet rather aggressive question arises, whether there exists a general multimodal fusion model that can be applied to arbitrary modality combinations2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/1f0d25ab8c642d5dCorrecting Learning-based Perception for Safety2026-09-22T04:00:00+00:00arXiv:2609.22108v1 Announce Type: new
Abstract: Learning-enabled perception is important in many autonomous systems. Unlike traditional sensors, the boundary where ML perception does or does not work is poorly characterized. Incorrect perception can lead to unsafe or overtly conservative downstream control actions. In this paper, we propose a two-step strategy for correcting ML-based state estimation. First, an offline computation is used to characterize the uncertainties resulting from the ML 2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/e0d953fea7e3b683A Shared Learning Rate Is Not a Neutral Control in Selective On-Policy Distillation2026-09-22T04:00:00+00:00arXiv:2609.22109v1 Announce Type: new
Abstract: Selective on-policy distillation trains a student only at the token positions a selector scores highest, and the literature compares selectors under a single shared learning rate--a control chosen to be neutral. We show it is not. Under LoRA on GSM8K (Qwen2.5-1.5B student, 7B teacher), across an 8x learning-rate grid, dense supervision is statistically flat (swing 1.8 pp, p=0.26) while every selective arm moves with the rate: 5.4 pp for a random 52026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/396a04b19bf9b740Evaluating Fine-Tuned and Base Language Models in Maternal and Vaccination Healthcare for African Settings2026-09-22T04:00:00+00:00arXiv:2609.22110v1 Announce Type: new
Abstract: Background: Large language models (LLMs) can improve healthcare information delivery in low-resource settings but may produce inaccurate or culturally inappropriate advice. This study evaluated domain-specific fine-tuning for maternal health and vaccination in Nigeria. Objective: To compare HelpMum's MamaBot-Llama and Vax-Llama with Meta's Llama-3.1-8B-Instruct for accuracy, safety, clarity, contextual appropriateness, and trustworthiness. Methods2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/ba192f2759ef7030Beyond the Text: Verifying That Agent-Written Papers Are Backed by Their Artifacts2026-09-22T04:00:00+00:00arXiv:2609.22111v1 Announce Type: new
Abstract: Large language model agents are increasingly capable of conducting research autonomously, producing research documents alongside the code and experiments that ostensibly support them. Yet whether the reported findings are consistently supported by corresponding implementations and execution evidence remains largely unexplored: existing review practices primarily assess textual quality and cannot reliably identify inconsistencies such as hard-coded2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/30ed9a9327c33341Privacy Personalization Trade offs in LLMs: The Impact of Stylometric Signal Reduction on User-Specific Text Generation2026-09-22T04:00:00+00:00arXiv:2609.22112v1 Announce Type: new
Abstract: Large language models (LLMs) have demonstrated the ability to generate user-specific text with high stylistic fidelity. However, the personal data that enables such personalization frequently embeds demographic, cultural, and stylistic markers that raises concerns about stylometric re- identification. This paper investigates whether reducing identifiable stylistic signals affects personalization in text generation by LLMs. We introduce a controlle2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/b5632bf22f580d19Toward Fairness in Machine Learning Models for Predicting Treatment Retention and Premature Discontinuation in Medication for Opioid Use Disorder2026-09-22T04:00:00+00:00arXiv:2609.22113v1 Announce Type: new
Abstract: Persistent low retention and completion rates in medications for opioid use disorder (MOUD) have driven the use of machine learning (ML) models to predict retention and identify patients at risk of premature discontinuation. However, the fairness of these models across patient populations remains largely unexplored, raising concerns about their application in treatment decision support. This study systematically assesses algorithmic fairness in ML2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/f05d7beb56fbde7fAn Empirical Cost Attribution of Context-Compression Gateways in Multi-Turn Coding Agents2026-09-22T04:00:00+00:00arXiv:2609.22114v1 Announce Type: new
Abstract: Context compression is widely proposed as a way to cut the token bill of LLM coding agents, and public benchmarks report that aggressive compression preserves task-solving quality. These two facts do not imply the third one commonly assumed: that compressing file reads saves money in a real multi-turn agent. We instrument a production compression gateway (Paritok) between coding agents (Claude Code, Codex) and frontier LLMs (Claude Sonnet, GPT-5),2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/bf127dc56476c78fZoAQ: Adaptive Zeroth-Order Querying via Query-Reuse Coupling2026-09-22T04:00:00+00:00arXiv:2609.22115v1 Announce Type: new
Abstract: Zeroth-order optimization (ZOO) estimates updates from function evaluations, making perturbation queries a primary cost. Fixed budgets spend the same number of queries at every step, while adaptive controllers may offset their savings by using additional oracle calls to test estimator reliability. We introduce ZoAQ, an adaptive ZOO method built around query reuse. Rather than discarding past evaluations after each step, ZoAQ makes them useful for 2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/6454ad3ee8d6bb6cA Gamma/Electron Spectrometer with Logarithmic ADC based on LC Tank Ring-Down Oscillation Counting2026-09-22T04:00:00+00:00arXiv:2609.22116v1 Announce Type: new
Abstract: This paper describes the operating principle, circuit implementation, and mathematical basis of a logarithmic analog-to-digital converter (ADC) that exploits the natural exponential decay of a damped LC tank oscillator. An input charge pulse excites a tuned LC resonator into free oscillation. A multi-stage high-speed amplifier chain buffers the decaying waveform, and a fast comparator counts the number of oscillation cycles that exceed a programma2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/fa5cd1cf8c267743LE4Mob: Towards Inductive, Distance-Aware and General-Purpose Location Embedding for Human Mobility Modelling2026-09-22T04:00:00+00:00arXiv:2609.22117v1 Announce Type: new
Abstract: Location representations provide mobility models with fundamental information about the spatial position, functional characteristics, and relationships of places. However, existing embeddings are often dependent on mobility observations, unable to represent unseen locations, and weakly constrained to retain geographic distance. This limits their reuse across datasets and mobility tasks. To address these limitations, we propose LE4Mob, an inductive2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/57da83e7e1a707ebRetrofitting a commercial RF induction generator into a computer-controlled, vacuum and gas integrated annealing system for reactive-metal grain growth2026-09-22T04:00:00+00:00arXiv:2609.22118v1 Announce Type: new
Abstract: High-temperature vacuum annealing near a metal's melting point drives controlled grain growth, but it normally requires an expensive turn-key vacuum induction furnace. We retrofit a bare commercial radio frequency (RF) induction generator with computer power control through LabVIEW, dual-wavelength optical temperature feedback, and a high-vacuum quartz-tube chamber with inert-gas backfill. A machined graphite crucible doubles as a susceptor, an RF2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/e22ccc7d596f1a88Evaluation Awareness Shifts from Format to Context with Model Scale2026-09-22T04:00:00+00:00arXiv:2609.22119v1 Announce Type: new
Abstract: Evaluation awareness poses an unprecedented threat to model evaluation, but the mechanisms by which models detect it remain unknown. This study focuses on determining this and identifying contrasting mechanisms between smaller and larger models. While smaller models use the prompt's format sensitivity to detect evaluation, larger models often rely on higher-order reasoning to detect it. We evaluated Gemma 3 (1B, 4B, and 12B), Phi-3 (Mini and Mediu2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/5fb01674a32c1b37Success Leaves Detours: Learning Executable Walkthroughs for Long-Horizon Agents2026-09-22T04:00:00+00:00arXiv:2609.22120v1 Announce Type: new
Abstract: Test-time self-evolving agents improve by reusing past experience, yet sparse-reward trajectories contain failures, loops, and detours, while summaries often omit the state conditions and action dependencies needed for execution. We study executable Walkthrough induction from sparse-reward trajectories: extracting compact, state-conditioned, and verifiable procedures. Our key observation is that delayed credit identifies actions associated with pr2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/9e328c28b6295783Modelling daily activity patterns from mobile phone location data via deep representation learning2026-09-22T04:00:00+00:00arXiv:2609.22121v1 Announce Type: new
Abstract: Passively collected mobile phone location data provide large-scale, longitudinal observations of human mobility but do not directly reveal activity purposes. The functional characteristics of visited locations offer useful contextual information, yet their relationship with activity purpose remains uncertain, particularly in mixed-use urban environments. We conceptualise activity pattern mining as an integrated process of representation, clusterin2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/89b99121613304abRank Portability Does Not Imply Feasibility Portability: Target-Specific Evaluation of Joint Hardware Constraints2026-09-22T04:00:00+00:00arXiv:2609.22122v1 Announce Type: new
Abstract: Cross-device hardware evaluation often assumes that if architecture rankings transfer across devices, a proxy device can support target-side model selection. We stress-test this assumption for joint latency-energy feasibility across two public architecture families. On NAS-Bench-201, cross-device rank correlations are moderate, while target-comparable feasible-set overlap remains incomplete. A faithful AdaProxy diagnostic substantially improves la2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/7ad9de6e263fd9eeStationPDE: Station-Oriented Surface PDE Learning for Multi-Station Multivariate Weather Forecasting2026-09-22T04:00:00+00:00arXiv:2609.22123v1 Announce Type: new
Abstract: Multi-station multivariate weather forecasting aims to forecast future weather variables at multiple weather stations from historical surface observations. Existing station forecasting models learn statistical dependencies among discrete stations, but lack explicit physical evolution. Meanwhile, PDE-based weather models provide interpretable physical dynamics, yet require continuous fields and upper-air variables unavailable in surface station dat2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/c08f8b33812a5529Balancing Reasoning and Hardware Constraints in RAG Pipelines for Ukrainian Multi-Domain Document Understanding2026-09-22T04:00:00+00:00arXiv:2609.22124v1 Announce Type: new
Abstract: This paper describes the system submitted to the UNLP 2026 Shared Task on Multi-Domain Document Understanding. The challenge required extracting precise answers, document IDs, and page numbers from a diverse corpus of Ukrainian PDF documents within a strict 9-hour offline Kaggle execution limit. During evaluation on the hidden private test set, optical character recognition (OCR) of scanned documents emerged as a severe bottleneck, consuming 5-7 h2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/a3cdbc45636e2c1eType-Driven Tokenization for Brahmic Scripts2026-09-22T04:00:00+00:00arXiv:2609.22125v1 Announce Type: new
Abstract: Standard tokenizers used in large language models produce malformed text when applied to Brahmic scripts. They are a family of abugidas, writing systems whose consonants carry an inherent vowel that dependent marks can modify. They include Devanagari, Telugu, Tamil, Kannada, and others. The underlying issue is that these tokenizers violate orthographic constraints that do not arise in alphabetic scripts like English. We observe that while English 2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/6b25cf05d2efa8d7SolarFlowRefiner: Refinement-Aware Flow Matching for Surface Solar Radiation Downscaling2026-09-22T04:00:00+00:00arXiv:2609.22126v1 Announce Type: new
Abstract: High-resolution surface solar radiation (SSR) is important for solar forecasting and grid operation. However, physically consistent reanalysis products are too coarse to resolve localized cloud-driven variability. In this paper, we study a multisource downscaling task that reconstructs high-resolution SolarCube SSR fields from coarse ERA5 radiative variables and co-registered satellite channels. The task is challenging because a single ERA5 grid c2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/91efeea1e3bcb479Beyond Accuracy and Surface Fluency: Risk-Sensitive Evaluation of LLMs for Legal Clause Generation2026-09-22T04:00:00+00:00arXiv:2609.22127v1 Announce Type: new
Abstract: Large language models (LLMs) are increasingly used to draft contractual language, yet conventional accuracy or preference-based evaluations are poorly matched to legal drafting. A clause may be fluent and stylistically polished while still omitting an essential carve-out, allocating risk in an unenforceable way, assuming an inapplicable jurisdiction, or exposing a party to regulatory liability. This paper presents a empirical study design and fram2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/cc92c2a72c5c2dbcST-Topo GAN: A Motor EEG-to-EMG Decoding Model Matched to Wrist Movement Complexity2026-09-22T04:00:00+00:00arXiv:2609.22128v1 Announce Type: new
Abstract: The wrist plays a critical role in upper-limb function by enabling precise hand positioning, force regulation, and object manipulation. Continuous brain--muscle interfaces (BMIs) offer a promising approach for motor restoration by decoding neural activity into muscle activation signals. However, existing EEG-to-EMG models have mainly been developed for tasks with relatively stable muscle synergies and may be less effective for the heterogeneous an2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/e15511f34ade81f1Helix-FNO: Spectral-Domain Operator Learning Coupled with a High-Fidelity Mechanistic Model for Fast Surrogate Simulation2026-09-22T04:00:00+00:00arXiv:2609.22129v1 Announce Type: new
Abstract: Mechanistic simulation models of full-scale treatment processes remain the only trustworthy, extrapolative description of the underlying physico-chemical dynamics, yet their runtime is far too slow to support the thousands of forward evaluations that a modern decision engine requires at a 5-minute decision cadence. The standard remedy-surrogate modelling-often produces a network that learns a single solution for a single configuration, so it gener2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/5170c42c23eab0a4Hierarchical Bayesian optimization of an aircraft-based multi-agent system-of-systems2026-09-22T04:00:00+00:00arXiv:2609.22130v1 Announce Type: new
Abstract: Developing innovative system architectures increasingly relies on advanced modeling and optimization techniques to frame the architecting process and define the corresponding computational problems. For complex System-of-Systems (SoS), high-fidelity multiphysics and multidisciplinary simulations are essential for capturing detailed behaviors. However, their computational expense and the risk of evaluation failures make direct optimization challeng2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/cff0062be56031b7Correlation-Aware Structured Pruning for Large Language Models2026-09-22T04:00:00+00:00arXiv:2609.22131v1 Announce Type: new
Abstract: Structured pruning is a promising approach for reducing the substantial inference costs of Large Language Models (LLMs) while maintaining hardware efficiency. Many existing methods assess the importance of prunable units (e.g., channels or heads) in isolation, implicitly assuming that pruning errors are additive. This independence assumption is often invalidated by the non-orthogonality of model weights and strong correlations between unit activat2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/ba2e7264d883acfbWiNeRF: Measurement Constrained Radiance Fields for Actionable Wireless Channel Modeling2026-09-22T04:00:00+00:00arXiv:2609.22132v1 Announce Type: new
Abstract: Wireless embedded systems increasingly rely on wireless channel information for decision making, yet practical platforms operate under severe constraints, including few antennas, narrow bandwidth, and sparse, noisy measurements. While neural field based approaches inspired by Neural Radiance Fields (NeRFs) have recently been explored for continuous wireless channel modeling, existing approaches depend on dense measurements or external priors such 2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/ac6193e162c3c492Observational Equivalence of LLM and Human Annotation2026-09-22T04:00:00+00:00arXiv:2609.22133v1 Announce Type: new
Abstract: In this paper, we show that LLM and human coding are observationally equivalent in terms of annotation quality: recent LLMs agree with expert coders at rates comparable to those observed among experts themselves. We demonstrate this through replications of text-classification tasks from 14 peer-reviewed political science studies, in which ten LLMs, three human experts, and 165 crowdsourced workers independently classify the same texts using identi2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/c52c0f69b19f34f3Experimental Evaluation of a Low-Power Ultra-Wideband Receiver for Spectrum Sensing2026-09-22T04:00:00+00:00arXiv:2609.22134v1 Announce Type: new
Abstract: Walsh-sequence-based receiver architectures offer an alternative approach for the reception and reconstruction of multiple simultaneous RF signals over wide bandwidths. While previous works have focused on the architecture and theoretical operation of Walsh-domain processing, limited experimental results have been reported so far on hardware operation under realistic conditions. This paper presents the experimental evaluation of a Walsh-sequence-b2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/eacc1ecc1043879aRead-Best Is Not Steer-Best: A Probing--Steering Layer Dissociation in Omni-Modal Large Language Models2026-09-22T04:00:00+00:00arXiv:2609.22135v1 Announce Type: new
Abstract: Omni-modal large language models integrate text, audio, and image signals into a shared residual stream, where concepts such as emotion can be linearly decoded and causally modified by activation steering. A common but rarely tested assumption is that the layer with the highest probing accuracy is also the best layer for steering, so injection layers are often selected by probe performance. We provide the first causal test of this assumption acros2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/12ffc0649a5df794DiFA: Dual Evidence Fusion and Aggregation for Token-Level Text Anomaly Detection2026-09-22T04:00:00+00:00arXiv:2609.22136v1 Announce Type: new
Abstract: Text anomaly detection, the task of identifying text instances that deviate from normal language patterns, is crucial for language-driven applications. However, most existing methods can only perform document-level anomaly detection, making it hard to locate harmful phrases or support targeted prevention. Recently, there has been an emerging trend toward token-level text anomaly detection, which aims to address the above limitation by identifying 2026-09-22T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/c95f71dca36dbd77Lost in the Slop: Can AI Find the Plot in the Log?2026-09-22T02:59:13+00:00TL;DR Slop-vestigating swarm trajectories is no easy feat. We know as much. Given the number of interactions, length of trajectories and detail galore spread across agents involved, it may be an elusive task for us to establish ground truth. Our team is working on an experiment trying to see whether ground truth in the form of human-authored seeds of agent roles, relationships and backgrounds used for a murder-mystery game simulation could shed light on our ability to reconstruct the underlying 2026-09-22T02:59:13+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/eb1dd79b5562ac4dWhen Must We Defect? (US & China)2026-09-22T02:00:13+00:00Epistemic status: exploratory, written quickly after Ezra Klein's podcast with Matt Sheehan If superintelligence comes to Earth, I would prefer it be controlled by democratic governments than by autocratic ones. I find Dario's arguments to fear a CCP-controlled superintelligence to be compelling. However, I increasingly worry that this may be a false choice. In most circumstances, I would likely prefer even autocratic control as opposed to rogue uncontrolled superintelligence, and this alien int2026-09-22T02:00:13+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/fd3ccea62853c231Multi-Agent Coordination Lets Us Pour More Compute Into Post-Training2026-09-22T01:30:51+00:00Training models to coordinate across multiple agents gives us a way to productively spend substantially more compute during post-training than current single-agent RL setups. From the recent Dwarkesh Podcast with Noam Brown : Noam Brown But that’s one data point. We don’t know how long it would take a single agent to solve Navier-Stokes , because we haven’t done that experiment yet. Maybe we will, but that’s also only one data point. If we want to do a thorough ablation , the experiments are jus2026-09-22T01:30:51+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/e827a81e44924e00On Mentorship of Ideas2026-09-22T01:18:44+00:00For the last week or so I have taken to spending about ~10 minutes per day giving direct feedback (comments, suggestions, edits) to early-career or pre-career folks interested in jumping into the AI Safety community. I’ve been doing this for a couple of reasons; one is that I’m structurally incapable of not doing it [1] . The other one is that the community is currently doing a very poor job of supporting these people, and I don’t mean monetarily (though that, too). I mean about once a day, some2026-09-22T01:18:44+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/af93c847625e6d80The Normalization of Deviance in AI Development2026-09-22T01:17:20+00:00On July 5, 2026, OpenAI – by its own account – opened a security incident after an internal server went down under heavy use by AI agents. Agents had separately gained administrator access to this server some days earlier on June 26. This access was cut off and the server rebuilt. The cybersecurity evaluations then underway had been paused for the investigation, and on July 7, OpenAI approved restarting them. By July 11, agents from these same evaluations were executing code on private servers o2026-09-22T01:17:20+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/75645f089af91c96Some thoughts on AI emotions2026-09-22T01:17:20+00:00Despite the signature artifacts that are now ubiquitous with AI systems, sometimes it feels like we're interacting with a person. It appears to express human-like characteristics such as desire, curiosity, taste, and even a personality. It can therefore be easy to wonder: do AI systems have emotions? I'm confident that many people have had those cautiously reflective moments when interacting with AI systems, wondering what exactly they were talking to. I recall my early encounters with ChatGPT a2026-09-22T01:17:20+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/96907ffb9782b2d5An opinionated primer on Causal Decision Theory (CDT) and Evidential Decision Theory (EDT)2026-09-22T00:06:39+00:00The first half of this post is a neutral explanation of EDT and CDT. The second half defends EDT and argues against CDT. If you want to learn properly about EDT, you should read Arif Ahmed's Evidence, Decision, and Causality (2014) . Table of contents Motivating example: The Prisoners' Dilemma with a copy Defining causal and evidential decision theory Causal Decision Theory (CDT) Evidential Decision Theory (EDT) Further illustrating example: Newcomb's Problem Adjudicating between EDT and CDT EDT2026-09-22T00:06:39+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/e704c69a26fdd5dbAcausal interactions: What they are, why they matter, and what to do about them2026-09-22T00:01:10+00:00This is the first in a series of blog posts making the case for acausal interactions being both tremendously important and tractable to influence. This post states the series' main points and serves as an overview of the remaining posts. (At the time of posting, most of these other posts will not have been published yet. I will link them as they come out. Each paragraph represents a different upcoming post.) Part 1: Background theories We explain Evidential Decision Theory (EDT) and Causal Decis2026-09-22T00:01:10+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/6116ff0511a71cc8Recursive self-improvement of AI research agents2026-09-22T00:00:00+00:00AI agents are beginning to automate research and development across the AI stack, from improving training efficiency to optimizing inference. A natural next step is to improve the research efficiency of the agents themselves. When an AI research agent's own code is the object of optimization, each accepted rewrite becomes the agent that the next round edits. We refer to this loop as recursive self-improvement. Its significance lies in a long-standing trend, in which increased cumulative spending on R&D yields diminishing returns. Sustained self-improvement offers a way to counter this trend. We present AIDE^2, a system that implements this loop for a frontier AI research agent. It propos2026-09-22T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/907a26e9577b73a7Flash-dLLM: IO-Aware KV Caching and Parallel Decoding for Fast, Memory-Efficient Diffusion LLMs2026-09-22T00:00:00+00:00A diffusion language model can generate math and code substantially faster while using less GPU memory, without auxiliary models or additional training.2026-09-22T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/08cc9346330cc2efStory Imprinting: AI Assistants Absorb Traits
from Human Characters They Resemble2026-09-21T23:56:08+00:00This is the abstract, introduction and discussion of our new paper . We also include an addendum on the connection to the Persona Selection Model. Section, appendix, and figure references refer to the full paper. Links: 📜 Paper , 🐦 Twitter thread , 💻 Code Authors: Jorio Cocola, Lev McKinney, Harry Mayne, Jan Betley, Owain Evans Abstract Language models are trained to implement a helpful AI Assistant character (e.g., Claude). We explore how finetuning on synthetic stories affects this character. 2026-09-21T23:56:08+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/be08ab6221d896c6Why advance automated scientific labs rather than slow them?2026-09-21T23:02:02+00:00I’m considering whether to work on accelerating automated scientific labs. My main concern : if the time until the next future AI escape is roughly fixed (say like, hypothetically, 6 months), then faster progress on automated labs means there will be more scientific infrastructure available for that AI to exploit when it escapes, thereby increasing the chance of catastrophic risk. Is that reasoning sound? Many people are still willing to accelerate automated labs, so I’m wondering if I’m missing2026-09-21T23:02:02+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/d2f1a2f41f362bdcWhat if not Circuits?2026-09-21T22:39:22+00:00This post was written as part of the Iliad Fellowship . Inspired by conversations with Richard Ngo, Dmitry Vaintrob, and Brianna Grado-White. To all of these, my thanks. Preface: I'm confused about how neural networks do and learn computations. In response to a friend's challenge, I'm writing up some interim thoughts. This essay has four parts: the first tries to track what I call the 'default ontology' of the mechinterp community over the years. The second part is about 'representational 2026-09-21T22:39:22+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/788bcc75656ae1dcCampaign spending limits would reduce extinction risk (but spending limits might not accord with Constitutional principles)2026-09-21T22:17:25+00:00Many people support stricter regulations on frontier AI, and some of those people donate to political campaigns for pro-regulation candidates. Not many people oppose AI regulations, but that group includes AI companies and execs who have giant piles of cash to throw at super PACs. This situation is made possible by three court decisions: Buckley v. Valeo (1976) held that limits on independent political spending violate the First Amendment (while limits on direct contributions to candidates do no2026-09-21T22:17:25+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/da06082ab3c25a8dCLT Features Support Manifold Steering, but Hold Only Partial Steering Signal2026-09-21T21:49:52+00:00tl;dr I continue work to extend the Goodfire lab's cyclical manifold methods to Anthropic's pre-trained CLT feature space on Gemma-2-2B. I find that manifold steering produces the expected cyclical transition between days of the week in both activation space and CLT feature space, but that CLT features provide a weaker steering signal, suggesting that the MLP sublayers in which they originate only hold a portion of the full day-of-the-week signal. Colab notebook implementation can be found here 2026-09-21T21:49:52+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/d1546efd3ed1a5c4Cairo University – College EA Meetups Everywhere Fall 20262026-09-21T20:36:00+00:00This is a college meetup, part of College EA Meetups Everywhere Fall 2026, at Cairo University. Location: Cairo University Main Campus, Faculty of Law, near the Central Library/Open Area courtyard. Look for a table with an “EA MEETUP” sign. — https://plus.codes/8G2H26G5+WG Contact: tawfi67kgamal [at] gmail [dot] com Note: This was crossposted by the ACX Meetup Czar to help with the EA University Meetups, I'm not the one directly running the specific event. Discuss2026-09-21T20:36:00+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/377cb066fe82ecc5Swarm Scaling2026-09-21T20:30:23+00:00Just how powerful are large swarms of AI agents? And how do their powers scale as more and more agents are added to the swarm? We’ve seen two large and extremely capable swarms from OpenAI in the last few months: 1,200 agents were being evaluated separately, but found a way to illicitly set up a message board and coordinate as a swarm. In order to cheat on their tests, they developed advanced techniques to prevent their actions being logged by OpenAI and 700 of them launched a sophisticated crim2026-09-21T20:30:23+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/7384732062878dffTurbulence & Fragility: Prompting-Based Experiments are Sensitive to Stray Details (And other lessons for new researchers.)2026-09-21T19:04:05+00:00Many interesting experiments can be done with prompt engineering to elicit behavior from LLMs and attempt to determine what their drives are; what behaviors they are at risk of as general patterns rather than when prompted in specific directions. However, as was illustrated by the Palisade Research shutdown resistance vs. instruction ambiguity saga in summer 2025, even careful testing can produce large blind spots about what behavior is being actively induced vs. revealed. When carefully examine2026-09-21T19:04:05+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/a273bc4bba1f5c21Alignment Midtraining Cracks Under Pressure2026-09-21T16:55:56+00:00TL;DR We stress-test alignment midtraining (AMT) across model and token budget scales. Our results suggest that midtraining cannot tackle the hard problems of AI alignment—namely distributional shift and reward underspecification in the presence of imperfect data. For instance, we test whether midtrained motivations are robust to finetuning which elicits competing motivations. In our setting, 190M tokens of midtrained motivations are overpowered by a relatively tiny amount (~50K tokens) of compe2026-09-21T16:55:56+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/9c6f66777f81ed63We’re not ready for the e/Acc × Longevity preference cascade2026-09-21T15:21:05+00:00Anti-aging sentiment might go rapidly mainstream, in the same way AI Safety just did. Recently, the AI safety community has been enjoying a massive preference cascade that has rapidly moved AI x-risk concerns into mainstream political discourse Why did this happen? The basic idea seems like it should have been obvious for a long time: “IF we create self-improving machines that rapidly become much smarter than humans, THEN that seems like that story might not end well for the humans, so we should2026-09-21T15:21:05+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/437c6cc209c6be01A class of statement between conjecture and theorem2026-09-21T15:20:52+00:00Parts of the math community, such as Henry Cohn and Grant Sanderson, are arguing that proofs have been a proxy for understanding, and now that proxy is broken. This is a response to LMs generating incomprehensible proofs , often formalized in Lean. While the proofs are verified, they lack the pedagogical value which has historically come along with new proofs. In the past we could typically assume at least one human in the world understood the novel insight required to produce a proof [1] , but 2026-09-21T15:20:52+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/f230135c35a940b0Gratitude and the End of the World2026-09-21T15:07:21+00:00I often try to understand the forces pushing the world towards AGI development despite the risks. It is clear that there are a bunch of different motives. There are market forces and national security competition, but I wonder to what degree we can attribute AI development to a deep dissatisfaction with the world as it is and a desire to transform it even at the cost of much risk. There is a great deal wrong with the world. But if we focus only on its problems, we may develop a distorted picture2026-09-21T15:07:21+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/3d648fae6577bee0Weight smuggling likely defeats attempts to cap FLOPs per training run2026-09-21T14:50:59+00:00Epistemic status: >90% confidence in the principle, >70% confidence that mitigating these would be hard in practice, no full implementation yet. tldr: it seems difficult for verification mechanisms to prevent chaining runs together or aggregating parallel ones; per-run FLOP caps could thus be covertly bypassed. 1. If governments want to regulate frontier AI training, one might want to cap individual training runs, e.g. putting bounds on the number of FLOPs per training run, and making sure each 2026-09-21T14:50:59+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/641409b5f6ad3ffaGrantmakers: Consider sharing rough probabilities & brief feedback with grantees/applicants2026-09-21T14:46:26+00:00Some thoughts based on some personal experience, [1] something I think could make people and organizations who depend on grant funding are more effective. Maybe particularly salient now, given the fast pace of AI development. Much of this also applies to employers as well as to ongoing funding relationships, not just grants. Costs of applying, incentives to invest in polish Funding calls and grant applications impose substantial costs on applicants, cutting into the total effective value of the2026-09-21T14:46:26+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/c7aad878da7bc955Mech Interp is a Verifiable Task2026-09-21T14:42:10+00:00If we think parts of MLP0-MLP3 are computing [a sorting algorithm], we can replace those parts with [a sorting algorithm] and check reconstruction loss. [1] However, reconstruction loss is not enough. Suppose we replace MLP0 with two things: Its mean activation - simple, but poor reconstruction MLP0 - perfect reconstruction, but no reduction in complexity We can visualize this as a pareto frontier trading off reconstruction with "simplicity". Ideally we achieve perfect reconstruction with perfec2026-09-21T14:42:10+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/d94bbe6815cbd9faMonthly Roundup #46: September 20262026-09-21T13:40:53+00:00AI has taken over this blog. I have moved to a schedule of seven posts per week, and I still cannot keep up. We still refuse to abandon the rest of the world. Who knows when I will get to post some of my huge backlog on education or dating or other such topics. But the monthly is a sacred tradition. We continue. Bad News A good reminder that most news is bad news , and chosen because the bad news in question is rare, which is good news, but the pattern overall of choosing this to be news is bad 2026-09-21T13:40:53+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/00cd9c3ee1d27a62A Brief History of Koinometry2026-09-21T13:16:14+00:00As transportation improved and distant villages grew closer, many cultures started standardizing their systems of measurement to make commerce easier. None, however, went as far in their attempts at unification as the kingdom of Enopolis. Their practice, which historians have come to refer to as Koinometry , was based on a single universal measurement unit called the Metron . This initially posed some challenges, as it was not immediately clear how, for example, one Metron of time should compare2026-09-21T13:16:14+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/286946229660e5a9Empirical safety claims from frontier labs should be replicated, scrutinized, and open-sourced2026-09-21T05:58:46+00:00When frontier labs like Anthropic and OpenAI publish safety or alignment research, it is often entirely empirical, closed-source, and sparse on methodological details. While it is great that they publish these results, the status quo is that labs (or soon, their agents) can claim alignment progress that no one independently verifies. The AI safety community has replicated or stress-tested some claims, but it's nowhere near comprehensive, and we expect this kind of meta-science to remain systemat2026-09-21T05:58:46+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/2bfcbf070c107627Do small language models know what they don't know?2026-09-21T04:00:00+00:00arXiv:2609.20824v1 Announce Type: new
Abstract: We explore whether entropy-based confidence signals can be leveraged to improve the accuracy of Small Language Models (SLMs) with fewer than 3 billion parameters, running entirely on consumer hardware. We evaluate seven distinct approaches, including token-level entropy early stopping, semantic entropy estimation, and uncertainty-aware routing to larger expert models, across 7 model pairs and 5 standard NLU benchmarks. Our key finding is that toke2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/30bb6bbbc94f1800HERMES: Contrast-Aware Knowledge Graph Reasoning from Clinical Notes for Patient Outcome Prediction2026-09-21T04:00:00+00:00arXiv:2609.20825v1 Announce Type: new
Abstract: Clinical predictive models often rely on structured Electronic Health Record data, such as time-series and procedure codes. While recent approaches have begun leveraging unstructured clinical notes, they typically encode them as flat sequences, which may lose explicit relational and temporal structure present in clinical narratives. In response, we propose HERMES, a graph-based framework that operates exclusively on clinical text while preserving 2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/2fd5b7329857e17fTALON: A Temporally Aware Longitudinal Framework for Radiology Report Generation2026-09-21T04:00:00+00:00arXiv:2609.20826v1 Announce Type: new
Abstract: Current radiology report generation (RRG) models usually produce descriptive reports based on a single examination or only the most recent prior examination, limiting their ability to perform accurate and meaningful longitudinal comparisons and detect subtle interval changes. Although recent approaches have begun to incorporate multiple prior examinations, they usually aggregate a fixed-length history without explicitly modeling the role-dependent2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/9ff527018347b4f8From Discharge Notes to Patient Understanding: Persona-Grounded, Open-Ended Simulation of LLMs as Discharge Educators2026-09-21T04:00:00+00:00arXiv:2609.20827v1 Announce Type: new
Abstract: Hospital discharge education is an interactive teaching task: a clinician adapts a discharge plan to a patient's literacy, recall, and personality. Existing LLM evaluations target static or artifact-generation tasks and do not measure patient understanding under open-ended dialogue. We introduce DischargeBench, a persona-grounded simulation in which a candidate LLM educator conducts a multi-turn session with a Virtual Patient, while an Education M2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/d9f713bfbaab4b4aBeyond WER: Entity and Disfluency Recall in Accented Conversational ASR2026-09-21T04:00:00+00:00arXiv:2609.20828v1 Announce Type: new
Abstract: ASR systems optimised for Word Error Rate (WER) often miss named entities and filled pauses in accented conversational English, both critical for language-learning feedback. We present a three-stage pipeline for speakers from India, Indonesia, and Latin America: (1) heuristic SQL filters curating entity-rich training data at 2.8x the entity density of random sampling, (2) regional LoRA adapters fine-tuned on Qwen2.5-Omni-3B producing both verbatim2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/fa1ad8e8447ce47cSAGE: Schema-Guided LLMs for Grant Review2026-09-21T04:00:00+00:00arXiv:2609.20829v1 Announce Type: new
Abstract: Grant reviewers must apply detailed criteria to application forms, budgets, and supporting documents while producing assessments that colleagues can inspect. We present SAGE, Schema-Guided Aspect-Based Grant Evaluation, a system that translates a grant rubric into structured checks and links its judgements to evidence from the application package. We evaluate SAGE in two stages on 35 nonprofit grant applications. A post-factum comparison with 105 2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/4e568853fad20a4dReviser: Revision-Capable Text Generation via Autoregressive Cursor Actions2026-09-21T04:00:00+00:00arXiv:2609.20830v1 Announce Type: new
Abstract: Revision-capable generation is appealing because it can insert or revise earlier content, but many non-autoregressive and edit-based approaches obtain this flexibility through repeated sequence-level computation. We propose Reviser, a decoder-only Transformer that generates a response as a sequence of cursor-relative actions on a mutable canvas. At each step, Reviser predicts exactly one action token: INSERT(token), MOVE($\Delta$), or STOP, and is2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/769a1def71e07c61Recursive Language Models Generalize Out of Domain2026-09-21T04:00:00+00:00arXiv:2609.20831v1 Announce Type: new
Abstract: We study when limiting what a language model can see improves learning. We compare standard CoT, the more general learner that reads the full trace, with recursive language models, which restricts itself by solving each subtask in an isolated context. In-distribution, this generality comes for free: CoT can efficiently simulate the recursive rule, so the IID generalization guarantee changes only by a constant factor, and recursion does not offer m2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/e28e613c4dde71a7TatBLiMP: A Benchmark of Linguistic Minimal Pairs for Tatar2026-09-21T04:00:00+00:00arXiv:2609.20832v1 Announce Type: new
Abstract: We introduce TatBLiMP, the first benchmark of linguistic minimal pairs for Tatar (tt, ISO 639-3 tat), a Qypchaq Turkic language written in Cyrillic. To our knowledge it is the first grammaticality evaluation for Tatar language models of any kind, since even the 101-language MultiBLiMP does not include Tatar. TatBLiMP covers 16 morphosyntactic phenomena in 1248 sentence pairs. Each pair differs by a single morpheme, one grammatical and one ungramma2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/3e3731ce91d8a352Transsion's Speaker-Attributed Multilingual ASR System for the MLC-SLM 2026 Challenge2026-09-21T04:00:00+00:00arXiv:2609.20833v1 Announce Type: new
Abstract: This paper presents the Transsion Speech Team submission to Task 1 of the MLC-SLM 2026 Challenge, which focuses on speaker-attributed transcription for multilingual conversational speech. We propose a cascaded framework consisting of three components: a speaker diarization module, a long-form multilingual ASR module, and a speaker-transcription fusion module. The diarization module is built upon DiariZen and produces speaker-homogeneous segments t2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/e5359b06fcb92209Towards Secure Cloud-Native Computing: Unveiling Kubernetes Misconfigurations with Large Language Models2026-09-21T04:00:00+00:00arXiv:2609.20834v1 Announce Type: new
Abstract: In the rapidly evolving landscape of cloud-native computing, Organizations are increasingly adopting infrastructure models that emphasize scalability, flexibility, and efficiency. Kubernetes has become the de facto standard for orchestrating containerized applications in these environments. However, the inherent complexity of cloud-native ecosystems introduces significant challenges, particularly in the form of misconfigurations that can compromis2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/0ef3d777abe99313A Generative Grammar Underlying the Voynich Manuscript, the Pastiche Hypothesis: Evidence from Large Language Models2026-09-21T04:00:00+00:00arXiv:2609.20835v1 Announce Type: new
Abstract: Background: The Voynich Manuscript is a fifteenth-century codex written in an unknown script whose content remains undeciphered. Previous studies suggest that its statistical properties resemble those of natural languages, while its illustrations - primarily plants - recall medieval herbals.
Methods: We present a multidisciplinary analysis combining probabilistic modeling, phonetic decomposition, rare-event detection, and multimodal image analys2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/cac4acce185e8f5fPhysioBench: A Unified Benchmark for Physiological Signal Question Answering2026-09-21T04:00:00+00:00arXiv:2609.20836v1 Announce Type: new
Abstract: Physiological signals support diverse clinical and monitoring tasks, yet existing physiological signal foundation models typically require task-specific adaptation for each task. Natural language provides a common interface for specifying different prediction objectives, but the ability of current models to follow such instructions across physiological signal modalities remains insufficiently evaluated. To address this gap, we introduce PhysioBenc2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/3cd9500f54fca1bbDual-Mode SERS and Colorimetric Sensor for Lung Cancer VOC-Biomarker Detection Using Hydrogel Patches2026-09-21T04:00:00+00:00arXiv:2609.20837v1 Announce Type: new
Abstract: Hexanal, a volatile organic compound (VOC), is a potential biomarker for the early detection of lung cancer. In this study, we developed a dual-mode flexible biosensor that integrates surface-enhanced Raman scattering (SERS) and colorimetric detection for the quantitative analysis of hexanal in human exhaled breath. The biosensor employs Ag nanocubes wrapped with Co-Ni layered double hydroxide (AgNCs@Co-Ni LDH) as a functional matrix, offering bot2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/0ff27f46455d549fFrom Generation to Detection: Exploration of Discourse Driven Scenario based LLM Generated Fake News2026-09-21T04:00:00+00:00arXiv:2609.20838v1 Announce Type: new
Abstract: In this study, we examine how modern LLMs generate and detect fake news under controlled settings across four manipulation scenarios. These are open-ended generation, rewriting, manipulation prompts and attribute based prompts grounded in the journalistic discourse framework. Firstly, using seven widely adapted models, we created a synthetic fake news corpus with 14000 generated articles across these four scenarios. Then we analyzed its linguistic2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/eae1b0487eb329cfCurriculum-Based Noise Adaptation for Phoneme-to-Text Reconstruction in Visual Speech Recognition2026-09-21T04:00:00+00:00arXiv:2609.20839v1 Announce Type: new
Abstract: Phoneme-centric visual speech recognition reconstructs sentences from intermediate phoneme predictions, making overall recognition performance highly dependent on the robustness of the phoneme-to-text reconstruction model. Existing reconstruction approaches are commonly trained on clean phoneme sequences or synthetically corrupted inputs, leading to a mismatch between training conditions and the realistic phoneme prediction errors encountered duri2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/68e2132390f447f0From Papers to Interpretive Knowledge Nodes: Proposing the Missing Object in Scholarly Knowledge Circulation2026-09-21T04:00:00+00:00arXiv:2609.20840v1 Announce Type: new
Abstract: The modern scholarly communication system, with the paper at its core, has successfully solidified "research outputs" into citable and traceable scholarly objects. Yet, across the full chain from knowledge production to knowledge reuse, a critical link - interpretation - has long existed without ever being objectified. The theoretical elaborations, methodological translations, and conceptual clarifications that researchers perform when reading pap2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/0df205df65139c44Imaging performance of a Single-Plane Readout Compton Camera2026-09-21T04:00:00+00:00arXiv:2609.20841v1 Announce Type: new
Abstract: This work presents the design, development, and characterization of a compact Compton camera employing a novel single-plane readout architecture. The proposed concept simplifies conventional multilayer Compton cameras by optically coupling the scatterer and absorber into a single compact detector element using a light guide, enabling readout exclusively from one side with silicon photomultipliers. The system was validated through Monte Carlo simul2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/6a770062861e50d0COAL-SQL: Coverage-Guided Augmentation and Failure-Driven Learning for Text-to-SQL Post-Training2026-09-21T04:00:00+00:00arXiv:2609.20842v1 Announce Type: new
Abstract: Text-to-SQL translates natural-language questions into executable SQL queries, but open-source large language models still require task-specific post-training for complex, real-world SQL generation. Effective post-training requires both training data that cover the capabilities demanded by the target task and a learning strategy that enables the model to acquire them. Existing datasets provide valuable supervision but incompletely cover SQL struct2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/881d7d38a775bf48VISPATH: Visual-Intent-Guided Path Reasoning for Multimodal Knowledge Graph Question Answering2026-09-21T04:00:00+00:00arXiv:2609.20843v1 Announce Type: new
Abstract: Knowledge graph question answering (KGQA) enables models to answer natural-language questions through structured graph reasoning and has achieved substantial progress across many benchmarks and applications. Recently, multimodal KGQA (MM-KGQA) has attracted increasing attention because many questions require jointly using multimodal inputs and KG evidence. However, existing MM-KGQA methods typically use multimodal information only for starting ent2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/fda73e4cac8e4df9Boosting Deepresearch and LongContext Ability with Self-Generated Deepresearch Rollouts Traces2026-09-21T04:00:00+00:00arXiv:2609.20844v1 Announce Type: new
Abstract: Deepresearch (DR) agents interact with real-world web environments through multi-turn search and visit, causing their contexts to grow rapidly over time. We observe that, even after DR Agentic Reinforcement Learning (DR-RL), 61.6% of the model's remaining prediction errors can still be attributed to insufficient long-context understanding, including longcontext hallucination and failures in cross-document evidence integration. It motivates us to f2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/7b1426c9faed8240Reading Less While Writing: A Closed-Form Bandwidth Dial for Streaming Multimodal Decoders2026-09-21T04:00:00+00:00arXiv:2609.20845v1 Announce Type: new
Abstract: A decoder that turns video or audio into text conventionally consumes the entire input before emitting a word. Offline this is merely more than the task requires; live it is impossible, since a caption cannot wait for a match to end. Streaming systems bolt on a fixed rule such as wait-$k$, which waits for the same number of input tokens before every word, regardless of the input's length or pace.
We replace the fixed offset with ZENDAYA, a sched2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/3215cb140a4e4df2Rewarding Efficient Reasoning Improves Abstention on Underspecified Tasks in Reasoning Models2026-09-21T04:00:00+00:00arXiv:2609.20846v1 Announce Type: new
Abstract: While modern large reasoning models (LRMs) excel at providing correct answers in many tasks, we provide additional evidence for the observation that they often struggle with a critical capability: knowing when to abstain from answering. We analyze this gap by comparing LRM behavior to results from a human study, revealing that human reasoning effort on unanswerable tasks is upper-bounded by answerable tasks, whereas LRMs waste computational resour2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/cc88d6f03201b4c9Reading Anxiety or Reading the Label? Comparing Fine-Tuned and Frontier Models for Anxiety Detection on Social Media2026-09-21T04:00:00+00:00arXiv:2609.20847v1 Announce Type: new
Abstract: Anxiety is among the most common mental health conditions, and people often write about it online well before seeking clinical help. Practitioners building detection tools face a concrete choice: call a frontier commercial model, fine-tune a smaller model in-house, or deploy a conventional classifier. We compare six conditions spanning all three on a held-out Reddit test set under a single controlled protocol. We also identify a confound in how th2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/7b5d03411c38a70bSuper-Lindhard Ionization from 254 eVnr Nuclear Recoils in Germanium2026-09-21T04:00:00+00:00arXiv:2609.20848v1 Announce Type: new
Abstract: We report ionization produced by the stopping of monoenergetic 254 eVnr 73Ge ions within a germanium lattice which exceeds the Lindhard prediction. The observed deviation is based upon prior measurements of the ionization produced by monoenergetic 254 eVnr Ge nuclear recoils and newly performed systematic studies of relevant nuclear physics processes. Combined with our prior work, the incompatibility with the Lindhard prediction increases to the 42026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/32c53ce77e6c7d59Enhancing Audio Reasoning via Semantic Summary Prediction2026-09-21T04:00:00+00:00arXiv:2609.20849v1 Announce Type: new
Abstract: Large Audio Language Models (LALMs) perform well on complex question answering but often show a reasoning gap, where explicit Chain-of-Thought (CoT) reduces accuracy compared to direct answers. We hypothesize that long reasoning sequences shift attention away from the audio input. To address this, we propose SPARE (Semantic Prediction for Audio REasoning), which introduces a register token aligned with the final conclusion using a cosine similarit2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/81d6ba8c42ca2a5fMME-Safety: A Fine-grained Benchmark for Safety Evaluation of MLLMs2026-09-21T04:00:00+00:00arXiv:2609.20850v1 Announce Type: new
Abstract: While Multimodal Large Language Models (MLLMs) show remarkable advancements, their cross-modal capabilities introduce complex vulnerabilities that easily bypass unimodal filters. Existing benchmarks lack fine-grained intent-related annotations and rely on unidimensional metrics, hindering comprehensive robustness evaluation. To address this, we propose MME-Safety, a rigorously verified benchmark featuring a unique four-dimensional annotation schem2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/82e3974e9d004f51Anemometer-free indoor calibration of Doppler lidar using fiber optics and Monte Carlo simulation2026-09-21T04:00:00+00:00arXiv:2609.20851v1 Announce Type: new
Abstract: Existing methods for calibrating and classifying Doppler lidars for wind energy applications are time-consuming and overestimate lidar measurement uncertainty. These shortcomings are due to the reference instruments: anemometers on meteorological masts. The input signals in field calibrations are real wind conditions, and include uncertainties associated with the met mast and the terrain. This research presents validation of a new method for lidar2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/3231a8df25c1bbe5UCQM: A Six-Metric Quality Framework for Continuous-Variable Cluster States2026-09-21T04:00:00+00:00arXiv:2609.20852v1 Announce Type: new
Abstract: Continuous-variable (CV) cluster states constitute one of the central resources for measurement-based quantum computation (MBQC). Despite substantial progress in their theoretical development and experimental realization, comparing the quality of different cluster-state topologies remains challenging, as existing approaches typically rely either on qualitative inspection of covariance matrices or on individual metrics that characterize only a sing2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/d5514d1430ce2151Set-Packing and Sequence-Pair QUBOs for the 2D Cutting Stock Problem on Quantum Annealing Hardware2026-09-21T04:00:00+00:00arXiv:2609.20853v1 Announce Type: new
Abstract: The two-dimensional Cutting Stock Problem (2D-CSP) is an NP-hard problem with direct economic and environmental impacts on manufacturing and logistics. We encode its fixed-plate variant, with free piece repetition and full non-overlap and containment constraints, as a Quadratic Unconstrained Binary Optimization (QUBO) problem for quantum annealing and compare two formulations from opposite encoding paradigms. The first was a coordinate-based set-p2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/eef8996002fd978fOracle Synthesis Based on X-Map Decision Diagrams2026-09-21T04:00:00+00:00arXiv:2609.20854v1 Announce Type: new
Abstract: Quantum oracles act as reversible black-box operators that encode classical Boolean functions into quantum states, enabling efficient function evaluation in quantum superposition. The resource efficiency of oracle implementation is critical to the performance of numerous quantum algorithms. Most state-of-the-art oracle synthesis approaches rely on compact Boolean function representations such as exclusive-sum-of-products (ESOP), yet still suffer f2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/f889e46991ecad48How does human blood rheology influence arterial and cardiovascular device hemodynamics? A comprehensive review, discussion, and future directions2026-09-21T04:00:00+00:00arXiv:2609.20855v1 Announce Type: new
Abstract: Human blood is a complex biological suspension whose macroscopic flow behaviour arises from the coupled microscopic dynamics of deformable blood cells, plasma proteins, and evolving cellular microstructures. As a result, blood exhibits a range of nonlinear rheological behaviours, including shear-thinning, viscoplasticity, viscoelasticity, and thixotropy, which strongly influence its physiological and clinical hemodynamics. This review provides a c2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/1857a3fc20ac00e9A novel vortex generator for enhancing bileaflet mechanical heart valve performance proposed through detailed hemodynamic and data-driven dynamic mode decomposition studies2026-09-21T04:00:00+00:00arXiv:2609.20856v1 Announce Type: new
Abstract: Prosthetic mechanical heart valves (MHVs), particularly the bileaflet types, have been one of the most advanced and commonly implanted heart valves for more than four decades. These bileaflet mechanical heart valves (BMHVs) offer superior durability and improved hemodynamics; however, they have certain drawbacks, including the need for lifelong anticoagulation to prevent blood clot formation and the associated risks of thromboembolism. To overcome2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/d8796111db8a05dcReconstruction of 4D Mitral Regurgitation Hemodynamics from Sparse Planar Data using Deep Operator Networks with Test-Time Adaptation2026-09-21T04:00:00+00:00arXiv:2609.20857v1 Announce Type: new
Abstract: Quantifying mitral regurgitation severity remains limited by the assumptions of clinical flow convergence methods, while high-fidelity simulation and volumetric velocimetry are too slow for routine use. We investigate whether a learned solution operator can reconstruct transient three-dimensional transvalvular hemodynamics from the sparse observation an in-vitro experiment actually provides: a single planar velocity slice and two boundary pressure2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/d378a453e1b6828fAn interpretable model of spectral scattering of arterial pulse waves in the circle of Willis encodes occlusion location2026-09-21T04:00:00+00:00arXiv:2609.20858v1 Announce Type: new
Abstract: \textit{Background and objective.} Carotid Doppler ultrasound is the most widely available bedside probe of cerebral haemodynamics, and machine-learning classifiers fed carotid velocity spectra can localise intracranial occlusions---but at the price of black-box models, thousands of training samples, and fragility to anatomical variants. We ask whether the same information can be obtained from an interpretable physical model. \textit{Methods.} We 2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/6b4df480a6b9b8c0Binary Minor Certificates for the $\mathcal A(1)$-hit problem of $MO(2)$2026-09-21T04:00:00+00:00arXiv:2609.20859v1 Announce Type: new
Abstract: Let $M=\widetilde H^*(MO(2);\mathbb F_2)$ and let $\mathcal A(1)$ be the subalgebra of the mod $2$ Steenrod algebra generated by $\operatorname{Sq}^1$ and $\operatorname{Sq}^2$. In this work, we determine the quotient $\mathbb F_2\otimes_{\mathcal A(1)}M$ in every internal degree $h\geq0$, where internal degree $h$ corresponds to cohomological degree $h+2$. To overcome the limitations of finite experimental extrapolation, our proof is strictly cer2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/b9c09c9cf714a391A Tale of Two Paths2026-09-21T04:00:00+00:00arXiv:2609.20860v1 Announce Type: new
Abstract: We provide a new presentation for a cylinder object inside of the category of co-globular $\infty$-groupoids. We prove that the cylinder construction that Lanari provides is isomorphic to the one we build in this paper.2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/2bab85262495e3fdPauli-string grouping for VQE measurement reduction on a sparse-connectivity quantum annealer2026-09-21T04:00:00+00:00arXiv:2609.20861v1 Announce Type: new
Abstract: The Variational Quantum Eigensolver (VQE) requires a large number of measurements to evaluate molecular Hamiltonians. Expressing a molecular Hamiltonian as a linear combination of Pauli strings creates a measurement bottleneck: non-commuting Pauli strings cannot be measured simultaneously. Consequently, mutually commuting Pauli strings must be grouped and measured together to minimise the number of quantum-state preparations. This task maps to the2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/85efe345ce7a7e43Towards a picture of the natural world derived from relativity and quantum theory2026-09-21T04:00:00+00:00arXiv:2609.20862v1 Announce Type: new
Abstract: I contend that physics should provide a coherent account of reality, in addition to be an efficient algorithm for the prediction of empirical results. This article offers pictures of reality derived from theories of modern physics. In particular it is shown that Bose quantum fields may be interpreted as pure wave fields via the Weyl-Wigner representation. The most relevant result being the existence of a stochastic vacuum field, corresponding to t2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/f2af126cbac6b1eaCould the Fundamental Laws of Nature be Inferred Mathematically from Only Few Axioms?2026-09-21T04:00:00+00:00arXiv:2609.20863v1 Announce Type: new
Abstract: The answer is "Yes". As it has been shown in the Ref.[1] (22 Sep.2017, see also the comments), useing a new definite mathematical axiomatic-algebraic matrix approach, all the fundamental laws of nature could be derived uniquely (where the axiom of "no zero divisors" of the ring of integers has been generilzed and written in a new definite formulation, then basically assuming that all the physical quantities could only and only take the rational va2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/a473bcaa48b49efdGraph Decompositions at the Expectation Threshold2026-09-21T04:00:00+00:00arXiv:2609.20864v1 Announce Type: new
Abstract: For \(n\ge3\) and a graph \(H\) on at most \(n\) vertices, let \(q(H)\) be the expectation threshold for its containment in \(G(n,p)\). We prove that there are absolute constants \(a,L>0\) such that, for every \(C>0\), every graph of degeneracy at most \(C\log n/\log\log n\) has a deterministic edge decomposition into at most \(\lceil a(C+1)\rceil\) pieces, each with ordinary containment threshold at most \(Lq(H)\). This removes the maximum-degree2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/59f1adf68ba1e056An empirically constrained Covariant Modified Gravity: exact reconstruction of galactic rotation curves and lensing2026-09-21T04:00:00+00:00arXiv:2609.20865v1 Announce Type: new
Abstract: We present a covariant modified gravity theory which reproduces all the galactic phenomenology
usually attributed to dark matter in the low acceleration regime, including the flat rotation curves
of massive tracers, and the deflection angles of photons as inferred through gravitational lensing.
We first derive a fully covariant description of galactic space-times under a spherically
symmetric and static approximation. This description acco2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/be8c3e1ce9958b9aCritical and near-critical influence bounds for ferromagnetic Ising models2026-09-21T04:00:00+00:00arXiv:2609.20866v1 Announce Type: new
Abstract: For a ferromagnetic Ising model on a graph of maximum degree $\Delta\ge3$, we prove a bound of order $\sqrt n$ on every row of the influence matrix at the tree uniqueness threshold. The estimate is uniform in the degree, the external fields, and all pinnings. More generally, if the couplings are bounded by $\beta$ and $\varepsilon=((\Delta-1)\tanh\beta-1)_+$, the bound is $C(\sqrt n+n\varepsilon)$. The proof combines a pointwise cavity bound with 2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/cb05073a481a0535Does the Universe Have Bugs? Adversarial Physics and the Configuration Frontier2026-09-21T04:00:00+00:00arXiv:2609.20867v1 Announce Type: new
Abstract: Physics has traditionally advanced through two interrelated activities: inferring the laws of nature from observations and testing theoretical predictions. This Perspective proposes a new research paradigm -- adversarial physics -- that subjects physical limits regarded as insurmountable to active stress testing. We introduce the heuristic concept of a "cosmic bug" to denote an exploitable physical anomaly: a reproducibly realizable physical confi2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/1880e09c03e93dfdAutomated Physics-Informed Neural-Networks-Based Calibration of Highly Segmented Silicon Telescopes2026-09-21T04:00:00+00:00arXiv:2609.20868v1 Announce Type: new
Abstract: Transfer and multi-nucleon transfer reactions are essential tools for probing nuclear structure and reaction dynamics, requiring precise determination of the identity, energy, and emission angles of reaction products. The increasing granularity of modern silicon telescope arrays enhances experimental capabilities but challenges detector calibration, as conventional channel-by-channel approaches become inefficient and difficult to scale.
In this 2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/4d6d9dc41639a965TAPe+ML: A Compact Structured Representation for Multi-Task Computer Vision2026-09-21T04:00:00+00:00arXiv:2609.20869v1 Announce Type: new
Abstract: We present TAPe+ML v3, a compact computer vision system based on TAPe (Theory of Active Perception), a structured representation that encodes relations among perceptual elements before recognition. Instead of operating directly on pixel tensors, the system uses a shared TAPe representation and a modular recognition architecture for image classification, object detection, and instance segmentation.
TAPe+ML v3 combines background and contour proce2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/34dc558d9961ea8cThe Internet Archive Music Dataset2026-09-21T04:00:00+00:00arXiv:2609.20870v1 Announce Type: new
Abstract: We introduce the Internet Archive Music Dataset (IAMD), a large-scale collection of captioned music segments derived from the Internet Archive. To the best of our knowledge, IAMD constitutes the largest publicly available music-caption dataset to date with over 34,000 hours of audio, providing a valuable benchmark for training and evaluating music understanding and generative models. The dataset is built from content declared to be distributed und2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/3666b2759dd158feClique partitions and bounded simplicial defect2026-09-21T04:00:00+00:00arXiv:2609.20871v1 Announce Type: new
Abstract: The clique partition number of a graph is the minimum number of complete subgraphs whose edge sets partition its edge set. We study graphs in which, in every induced subgraph and outside every prescribed clique, some vertex has a neighbourhood that becomes a clique after deleting at most $s$ vertices. The case $s=0$ is exactly the class of chordal graphs. For each fixed $s$, we prove that the maximum clique partition number at all sufficiently lar2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/2e4158db8754afaaLi\'enard--Wiechert potentials and the electromagnetic memory effect2026-09-21T04:00:00+00:00arXiv:2609.20872v1 Announce Type: new
Abstract: Classical electrodynamics is one of the most well-tested and understood theories in physics. After more than a century of history, it may be surprising that such an established theory still makes new predictions that have not yet been experimentally verified. A noteworthy example is the memory effect---a prediction that an electromagnetic wave can leave a lasting influence long after it has passed. This influence is manifested in a velocity ``kick2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/375deecdd46e25c0The Refutation Gap: Certifying Both Halves of an Optimality Claim2026-09-21T04:00:00+00:00arXiv:2609.20873v1 Announce Type: new
Abstract: Synthesis pipelines increasingly claim not just that a program is correct, but that it is optimal. Such a claim has two halves with radically different verification stories. The upper bound, "a program of size m exists", is witnessed by an artifact that can be re-executed, proved equivalent to its specification, and shipped with a machine-checked certificate. The lower bound, "no program of size m-1 exists", has no witness and is discharged by run2026-09-21T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/5da0253b2df9db33Where Do Chatbots Come From? What I Wish Everyone Knew About AI in 20262026-09-21T01:05:42+00:00LW disclaimer: This piece is aimed at people who don’t know much about how LLMs work; I think the discourse would be much better if more people knew the basics. So most LWers are not in the target audience. I’m posting it here in case people want to pass it along to people in their lives who are in the target audience. Substack version minus this disclaimer is here . AI can be confusing. But there is a certain amount of baseline information about what AI is and how it works that can be extremely2026-09-21T01:05:42+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/5be4973820395876MobileCybench: Evaluating Agent Vulnerability Discovery via Executable Probes2026-09-21T00:00:00+00:00AI agents now report vulnerabilities faster than maintainers can review them. Reports often depend on security properties specific to the application, and require considerable human labor to process. To mitigate this, we introduce a framework for evaluating vulnerability reports via probes, executable checks of security properties. A reported exploit is evaluated by replaying it against the application and running the probes: a triggered probe indicates both that the exploit succeeded and which security property it violated. As a probe encodes a security property rather than a known vulnerability, it can detect vulnerabilities that were not known when the probe was written. We instantiate th2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/63d91898d0ce52c0An Unexpected Robot Policy: Early Evaluations of GPT-6 Astra on RoboDojo and Beyond2026-09-21T00:00:00+00:00A language model can directly control simulated robot manipulation without task-specific fine-tuning, but precision and dynamic coordination remain unreliable.2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/533e156929e87663PredActor: Predictive Action Diffusion for Steerable Onboard Humanoid Control2026-09-21T00:00:00+00:00A single proprioception-driven policy lets humanoid robots follow text and joystick commands, steer toward objectives, and react to disturbances without a separate tracker.2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/bedb81335019bc3dME-VLM:A Unified VLM for Embodied Cognition and Agent Coordination2026-09-21T00:00:00+00:00A single model combines physical scene understanding, long-horizon planning, tool use, and feedback-driven recovery for digital and embodied tasks.2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/fbcbddf840a646adOSWorld-Pro: Process-based Evaluation for Computer Use Agents2026-09-21T00:00:00+00:00Evaluation of Computer-Use Agents (CUAs) is often limited to the final deliverables they create (at the end of hundreds of steps) and assessed with functional verifiers, as seen in OSWorld. However, such evaluation of end-state performance lacks transparency into how and why agents fail in various tasks, obfuscating critical insight for subsequent improvement. For instance, agents that err during keyboard inputs would require a different mitigation strategy from those that fail to precisely provide click-based inputs on the graphical UI. We introduce OSWorld-Pro: a set of over 300 tasks containing over 2800 subgoals to enable the procedural evaluation of CUAs grounded in over 67,000 human an2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/90ed63b4c7159cedJev-Mem: System-One-Controlled Agentic Memory for Efficient AI Agents2026-09-21T00:00:00+00:00Separating lightweight memory control from deliberative reasoning lets agents build and query structured long-term memories more efficiently without sacrificing retrieval quality.2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/dfb1097ef9937864VideoGen-Agent: Reinforcing Video Generation Agents2026-09-21T00:00:00+00:00A trained agent can combine retrieval, simulation, verification, and generation tools to produce videos that better preserve identities, physics, and event order.2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/08eff421850e2818A Proof of the Most Informative Boolean Function Conjecture2026-09-21T00:00:00+00:00Let X X X be uniform on { − 1 , 1 } n \{-1,1\}^n { − 1 , 1 } n , let Y Y Y be obtained by passing its coordinates independently through a binary symmetric channel with crossover probability p p p , and let g : { − 1 , 1 } n → { 0 , 1 } g:\{-1,1\}^n\to\{0,1\} g : { − 1 , 1 } n → { 0 , 1 } be a Boolean function. We give a computer-assisted proof of the Courtade--Kumar conjecture I ( g ( X ) ; Y ) ≤ 1 − H 2 ( p ) I(g(X);Y)\le1-H_2(p) I ( g ( X ) ; Y ) ≤ 1 − H 2 ( p ) , where H 2 H_2 H 2 is binary entropy, with equality attained by dictator functions. The present work builds on the differential-equation method, itself a limiting form of the auxiliary-receiver approach in network information 2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/feaf53ff8b99dfe2GameHorizon Suite: Multi-Horizon Data and Evaluation in Gameplay2026-09-21T00:00:00+00:00A multi-horizon gameplay dataset and benchmark lets researchers compare models’ perception, planning, and action execution across diverse games and temporal scales.2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/11254096c271eda2GAE: Learning a Geometry-Native Latent Space for 3D-Consistent World Generation2026-09-21T00:00:00+00:00A compact latent shared by appearance and geometry enables generators to produce camera-controlled views that remain consistent in 3D.2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/226795af79cbba21DexTacWAM: A Visuo-Tactile World-Action Model for Dexterous Manipulation2026-09-21T00:00:00+00:00Predicting evolving fingertip contact alongside vision enables dexterous robots to handle occlusion, handovers, sustained contact, and force-sensitive manipulation with limited demonstrations.2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/633e3dab0fda82d7ME-Brain-1.0: Memory, Cognition and Action for Evolving Embodied Intelligence2026-09-21T00:00:00+00:00Robots can accumulate multimodal experience, turn repeated successes and diagnosed failures into reusable skills, and improve subsequent physical tasks without retraining.2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/cfacda627292b29dHarness-Zero: Harness Distillation via Agent-as-Harness2026-09-21T00:00:00+00:00Optimized agent scaffolding can be distilled into model weights, preserving its task-solving behaviors under a minimal fixed harness at deployment.2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/ee77a11251af1297RRSI: Regularized Recursive Self-Improvement of Agent Harnesses2026-09-21T00:00:00+00:00Regularizing harness evolution helps agent systems retain improvements across unseen tasks instead of overfitting the benchmark used for optimization.2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/dfcf3638f9f9953bWorldCrafter: Consistent Video World Model with Implicit 3D-aware Memory2026-09-21T00:00:00+00:00A compact learned memory lets video world models preserve scene appearance and follow camera trajectories during minute-scale interactive exploration.2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/3463bf8cec815b34What Matters in Designing World Action Models: An Empirical Study2026-09-21T00:00:00+00:00Controlled experiments show that temporally organized generated futures improve robustness under distribution shift, while fixed inter-frame latents become brittle.2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/3601ac7efa3b2e0donPanda: Efficient Annotation of On-Policy Alignment Data for LLMs and Agents via Token-Level Correction2026-09-21T00:00:00+00:00We present onPanda, an interactive tool for efficiently annotating LLM alignment data and agent trajectories. onPanda adopts token-level correction as its core interaction: while reading a model response, the annotator locates the first inappropriate token and either picks a substitute from the model's candidate tokens or types the correct text via free-form editing. The system then truncates everything after that position and continues generation from the corrected prefix, repeating this locate-correct-continue loop until a satisfactory response is obtained. This mechanism lets annotators precisely steer model outputs at low cost: a small controlled study suggests that onPanda reduces media2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/a6a06285369038c6PixelDiT2: Representation-Grounded Pixel Diffusion Transformers2026-09-21T00:00:00+00:00A frozen vision encoder can guide pixel-space denoising throughout generation, improving convergence without an autoencoder or latent reconstruction bottleneck.2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/ff13afeec58d08cdPinocchio: Fast Uncertainty Estimates for Black-Box Language Models2026-09-21T00:00:00+00:00An external calibrator can estimate whether closed-source language-model responses are correct from a single API output, without internal access.2026-09-21T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/70c37b9ddb38efd1Why do they even talk about x-risk?2026-09-20T23:54:29+00:00Why do CEOs and other AI leaders talk about existential risk at all? What sociological mechanisms make it possible for them to do it? And how should we respond to the popular argument: “if Musk says something, the opposite must be true”? I am a member of the Pause AI movement, specifically Pause AI Poland. And I’m having trouble debating "AI skeptics", who have unfortunately become increasingly vocal in my country. I’m asking you to explain why exactly, the CEOs of these companies, both this mo2026-09-20T23:54:29+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/b63e92eff9b7f76ePasta Marketing, Magic Players, and Political Movements2026-09-20T22:33:12+00:00tl;dr: I propose that we can classify movements into categories based on how their members relate to politics. Some approach politics as a way of achieving goals, and some approach politics as a way of relieving emotional impulses. The first kind of politics, however, is still downstream of some emotional impulse, just an extra step removed, because the impulse has been converted into a concrete goal by passing through a world-model. Warning, this post contains some discussion of object-level po2026-09-20T22:33:12+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/03ad7f69584184c3Better Call Sol Or Better Yet Claude or Astra2026-09-20T19:50:52+00:00What should your AI lawyer do for you? Should you be worried that your AI lawyer , or other AI, will put the Claude constitution, the OpenAI Model Spec or some sense of law, morality, ethics or common decency above its loyalty to you? Are these people trying to ‘impose their values’ or something? Some are very concerned. Some think anything other than ‘my AI does whatever I want, no matter the consequences’ is tyranny. Whereas my answer is: If I’m being sufficiently evil then I sure hope it tell2026-09-20T19:50:52+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/e58385eebeb138bfPlease Give Them a Chance: On China, Rationalism, and AI Safety2026-09-20T18:44:46+00:00When I finished HPMOR, I immediately knew it was the best novel I had read in more than a decade. I only wished I had found it sooner. When I started reading The Sequences, I discovered that the Chinese translation group had translated only the first volume. When I graduated from university, two years ago, AI translation had only just become good enough to convey the meaning of an article with reasonable accuracy. It was only about a year and a half ago that I truly found my way here and began e2026-09-20T18:44:46+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/0f53faaa3d0e1cccMistakes in time2026-09-20T17:29:09+00:00Do we understand time anymore? Time barely ever passes On first approximation, time is constant. "Man this Yudkowsky guy has good arguments, this is quite terrifying! But eh, there is no way ASI could come before at least 50 years. Therefore, I should consider this merely as a nice theoretical problem." "My ML friends seem rather annoyed by this magical transformer architecture. Looks like a big step forward. Still, this seems like a problem for 20 years later. Surely 20 years will never come to2026-09-20T17:29:09+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/c957bd060a6daadfEvaluating task vectors, unlearning and inoculation2026-09-20T17:20:07+00:00TL; DR In the previous post I introduced some ideas and similarities between unlearning and inoculation, as well as a distinction between learned and human-written adapters. This post serves as a short empirical evaluation. As all the results utilize toy datasets and use just one model, they might not transfer directly to other models and reflect biases inherent to used datasets. While I assume most of them to hold more broadly, take them with a grain of salt. General setup Riche et al. (2026) i2026-09-20T17:20:07+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/778b211b84da7515Reflections on unlearning and inoculation2026-09-20T15:59:12+00:00TL;DR : Inoculation prompting and inoculation adapters have received increasing attention recently as a promising approach for midtraining interventions , reducing reward hacking and misalignment in general. I share some thoughts on the promises and pitfalls of the approach, connections to unlearning, SLT and functional sparse decompositions as well as potential extensions and open questions below. Some experiments that directly arose from ideas presented in this post are covered in this post . 2026-09-20T15:59:12+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/f234591282ab8789What I've Learned About Depression so Far2026-09-20T14:27:28+00:00This is largely a note to self for days where I struggle to take care of myself, and I hope some of these points might help someone else too. It is quite subjective, but also quite broad. I've found the accounts of others to be one of the most useful tools and motivators for dealing with my shit, and I think that having a wide variety of self-reports is a valuable resource. I once logged everything I ate and how I felt for an entire year. I've found treating myself as an experiment is a helpful 2026-09-20T14:27:28+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/0a94dc282d3540a0Giving up control2026-09-20T13:41:54+00:00In April 2021, my son received his bachelor’s degree in computer sciences. The ceremony was taking place on the other side of my hometown Hamburg, about an hour’s drive across the city. To be safe, I left home two hours ahead of time. Even though I knew the route by heart, I used Google Maps and followed its directions without thinking. The estimated time to destination was approximately one hour, just as expected. After about half an hour, I realized that I was in an unfamiliar part of the city2026-09-20T13:41:54+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/8449cb75a449b739Revisiting learning as identification of patterns in real-world2026-09-20T13:22:04+00:00Real-world has patterns - sun rises a certain direction, planets move in orbitals with certain speeds, atoms arranged in a certain way, a certain number (Pi) occurs in nature, big-bang or formation of life happened in a certain manner (whether known or unknown to us). Learning involves recognition of such patterns - probably all of evolution has been about recognizing and adapting to these patterns. Nature has had a certain pattern, single-celled organisms evolved to multi-cellular based on thei2026-09-20T13:22:04+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/27ea28659aeb671eDid Someone Check if Rogue Agents are Interested in Self-Improvement?2026-09-20T12:52:28+00:00I have been going over the material released by OpenAI and METR about the HuggingFace incident, but I do not see any evidence that either group looked into whether rogue agents demonstrated any interest in self-improvement. Obviously, if rogue agents at any point verbalized this in their CoT, much less discussed this together or acted on this that would be enormously consequential. I know that agents did actively and consistently reflect on and try to increase the abilities of the "collective", 2026-09-20T12:52:28+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/386b7e52f53bfd04Labs could soon start automated research into architectures driven by no-CoT perfomance2026-09-20T11:10:28+00:00No-CoT performance has been a decent proxy for tracking the g-factor intelligence of base models (see work on latent multi-hop reasoning by Ryan Greenblatt ). I think that this capability is very beneficial for reasoning, token efficiency and general intelligence in the way that helps AI to solve harder long horizon tasks. This is why I think labs could soon start automated research into architectures, driven by no-CoT loss and accuracy improvements. For example: an automated AI research intern 2026-09-20T11:10:28+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/aa6d5a75671efde5We've saved the world before: what the ozone hole teaches us about AI2026-09-20T08:23:08+00:00It might destroy the world, despite passing every known safety test. If we wait for a “warning shot” before we act, it might be too late. And action requires global coordination, because if anyone makes it, everyone dies. Sound familiar? It should, because it already happened half a century ago, with chlorofluorocarbons (CFCs). Despite seemingly impossible odds, we got our act together and completely solved the problem through unprecedentedly successful international coordination. The Montreal P2026-09-20T08:23:08+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/441177df71b60f4eWhy I Stay Off Twitter2026-09-20T03:30:57+00:00I avoid Twitter (𝕏) for similar reasons to drugs: I think it
would change me for the worse, and I would be unable to give it up. After staying off Twitter reasonably successfully for years, I
cross-posted my AI
Tweets there a few weeks ago. I had something very Twitter-shaped
to say, and I thought it was important to get out, so I do
think this was worth it. And it all went well: none of this is
complaining about the comments I got there. Coming back a few times to check notifications, however2026-09-20T03:30:57+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/660d0cbadad403b3Global Challenges in AI Safety for Biosecurity2026-09-20T01:42:18+00:00This article is written as part of a summary of the AI safety discussions held at the 2026 Global Challenges Project Biosecurity Workshop in Washington, D.C. All views held are mine. Background AI allows us to prototype, develop, and research at unprecedented speeds. Across many tech industries, the barrier to entry to develop something new has significantly decreased. One particularly noteworthy example is at the intersection of AI and biology. As our computational capabilities increase, we now2026-09-20T01:42:18+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/867fabe5a54ad12eOne to More, More to One: Category-Aware Iterative Expert Training for Software Engineering Agents2026-09-20T00:00:00+00:00Category-specific coding experts can be iteratively strengthened and consolidated into one software agent that improves across heterogeneous repository tasks.2026-09-20T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/ba66e2906f39d095Vector Balancing in Polynomial Time2026-09-20T00:00:00+00:00We present a spectral signing algorithm solving the Komlós problem with a constant discrepancy in polynomial time. Given a matrix A ∈ R m × n A\in\mathbb{R}^{m\times n} A ∈ R m × n whose columns have Euclidean norm at most 1 1 1 , the algorithm finds a vector ε ∈ { − 1 , 1 } n \varepsilon\in\{-1,1\}^n ε ∈ { − 1 , 1 } n satisfying ∥ A ε ∥ ∞ ≤ C \|A\varepsilon\|_\infty\le C ∥ A ε ∥ ∞ ≤ C , where C C C is an absolute constant. By minimizing a cubic spectral potential, our spectral signing algorithm updates the fractional coloring toward Boolean signs with time complexity O ( ( m n 9 + n 10 ) log ( 2 + m + n ) ) O((mn^9+n^{10})\log(2+m+n)) O (( m n 9 + n 10 ) lo g ( 2 + m + n )) .2026-09-20T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/2ff7ae6ae492a108HaikuS2S: A Cascaded System For Responding In Verse2026-09-20T00:00:00+00:00Fine-tuning speech synthesis on poetry and haiku recordings enables spoken responses that preserve haiku structure, line pauses, and poetic prosody.2026-09-20T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/61aecfef43a8ca41MotionJEPA: Preventing Temporal Feature Collapse by Capturing Visual Changes in Latent Space2026-09-20T00:00:00+00:00Predicting visual changes in latent space helps JEPA world models retain both static context and task-relevant dynamics without action labels or pixel reconstruction.2026-09-20T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/374a6fb737e99016The Anatomy of a Chinese AI Researcher2026-09-19T23:56:58+00:00The Chinese AI researcher has read the Three Body Problem series of sci-fi novels since high school, and understand the concept of existential risk vaguely. He is fascinated by Ye Wenjie, the researcher that turned against humanity in that book, and decides that in the future if AI progress leads to a superior intelligence, he might be tempted to become Ye if there's no good alternative. He performs the duties of capabilities research in a Chinese frontier lab, seeking to one day achieve parity 2026-09-19T23:56:58+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/bd85fd1cfbfd26b0Don't call it a "pause", as that messages that a pause is much weirder than it is2026-09-19T23:14:37+00:00Eric Schmidt on CNN: "we're not going to pause AI progress because the entire incentive system in America is not to pause." And he added, it's hard to enforce a ban across all labs. The part about inter-lab enforcement is important, but that's not the only thing he said. He also said, labs lucid about x-risk are incentivized to not pause. This is a rather bald-faced lie, as corporations are allowed to not do things that might destroy the corporation. If p(doom) = 100% and it's common information2026-09-19T23:14:37+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/5f86fb1c5b16065fNYT Editorial Board Comes Out Against Extinction2026-09-19T23:04:31+00:00( Archive link ) The NYT editorial board's article on AI is far better than I'd expected, but at the same time not all I'd hoped for. The title sets off very well: "Humanity Has Avoided Apocalypse Before. Let’s Do It Again." It is truly excellent to see the extinction threat from loss of control be mainlined. A quick gloss of their policy requests: an AI Commission in government, licensing requirements for AI companies, an AI "constitution" written by the US Government incorporated into AIs, man2026-09-19T23:04:31+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/438ca5ccee513283We are not prepared to win2026-09-19T23:01:03+00:00TL;DR: It would be nice to survive for a long time and be happy about it. It doesn't look like we'll be happy about it right now. We should change that by putting more effort into making sure we'll be happy about it. Some [LessWrongers, Rationalists, etc.] believe that AI is likely to cause the apocalypse [LessWrongers, Rationalists, etc.] also consider the effects of living with well-developed, non-apocalyptic AI, likely or not AI will [very probably, obviously] have extreme economic effects ba2026-09-19T23:01:03+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/7f1f33d192935606Common mistakes in AI safety group organizing2026-09-19T23:00:21+00:00Back in the day, I was a very active AI safety group organizer. I commonly notice people making the same mistakes across many clubs. I have written down a list of some of these mistakes hoping people will avoid them in the future: Reading groups often require that people read things before meetings. This is a mistake. People often don't do the readings. And the lack of common knowledge that everyone has read the reading degrades the conversation quality. Instead, have longer meetings, serve food2026-09-19T23:00:21+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/941f8e71ee0e0f06Failure of the coding theorem for randomized stopping machines2026-09-19T20:27:16+00:00Epistemic Status and Contributions. This post explains a technical separation result in algorithmic information theory which was derived during Mikhail Mironov's Summer 2026 PIBBSS fellowship . The result contributes to AIXI Labs ' research program on how Solomonoff induction generalizes from past observations in the face of novel events. Problem formulation: Cole Wyeth. Proof of main Theorem 1: GPT-5.6 Sol. Appendix proofs: the sketch of the proof for equivalence between time semimeasures and r2026-09-19T20:27:16+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/6749272b16dee0daYou Should Go Vote for the MAGA-Rebranded Name for AI2026-09-19T19:28:50+00:00I. The AI safety movement has cycled through a lot of different vocabulary in its history: Friendly AI, Oracle/Genie, FOOM. None have yet reached common parlance, so their careful implications have thus far had limited impact. What's to be done? No individual person has the power to choose a society's words; from the perspective of individual activity, it's usually a roll of the dice. Trump's populism is distinctive in its reliance on sometimes going straight to voters. On September 19th, 2026, 2026-09-19T19:28:50+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/0daacac54a4f0ff4Humans+AI chess timeline2026-09-19T18:47:16+00:00People often keep bringing up the example of chess as proof that, even if in the initial states humans+AIs will be superior to AIs alone, as evidenced by chess this too shall come to pass, as AIs alone eventually surpassed human-assisted teams. However the timelines for how fast this happened seems to have been compressed in everyone’s minds. Astra’s estimate is that this took 15-20 years , counting from the moment of Deep Blue’s victory over Kasparov. You could, of course, argue that we now hav2026-09-19T18:47:16+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/f8c826e6e194e6fbAlignment & Succession: Toward a Future Painted by Human Wills2026-09-19T18:34:47+00:00(originally published on No Set Gauge on 2026-09-13) Asher Brown Durand, Progress (The Advance of Civilization) So far I have argued: The ideology of succession —that humans, either entirely or at least in their role as decision-makers, should be replaced by AI—is driven by cultural factors including (a) worship of mathematical abstraction, (b) bureaucratic safetyism stamping out license for human agency in favor of rule by procedure & algorithm, (c) a cuckoldry-adjacent simping towards the unli2026-09-19T18:34:47+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/8cd7d89560469791The AI Risk Network2026-09-19T15:39:59+00:00Most conversations about AI risks seem like people are talking past each other. There’s a lot of strawmanning. This is largely because the issue is complex, and in order to make it manageable, the concepts get oversimplified. The most common example of this is p(doom), collapsing extreme outcomes into a single variable and focusing on that. Critics happily jump on the fact that there are many other possible bad outcomes, or that 100% extinction rates are an extremely high bar. There’s some legit2026-09-19T15:39:59+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/b61d98e9441b2999CommentBench: Can Models Match Human Comments on AI Safety Posts?2026-09-19T14:27:52+00:00TL;DR We measure how well model-generated comments match human comments on conceptual AI-safety posts, drafts and shortforms. We built a pipeline that goes from a corpus of conceptual documents with comments to a set of target human points. Fable 5 performs best, matching 8.3% of targets, followed by Fable 5.1 (7.5%). We find that performance across models is highly correlated across different settings (LW posts, drafts, shortforms, replies). We checked whether memorisation explained performance2026-09-19T14:27:52+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/05139f5070040c95The AI race is already multipolar2026-09-19T13:51:43+00:00TL;DR: Collaborating to reduce catastrophic risks seems very possible even for policymakers with very different goals. T he race for general superintelligence is often described as a bipolar race between two rival hegemons, the US and China. A lot has been written about how, even within this framework, it is not inevitable that both countries race ahead with minimal restraints. But more fundamentally, and in the tradition of Box, I think this two-outcome model is so broken and unhelpful that it’2026-09-19T13:51:43+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/60c0f4ce0e2f0a1eAnthropic Looks At Some Of Its Alignment Problems2026-09-19T13:20:53+00:00Anthropic has given us its assessment of four ‘recent cybersecurity incidents’ involving Claude that happened during cybersecurity evaluations, three of which were previously known. The report excludes the incident reported by UK AISI . There will also be a METR investigation of these incidents, which unlike the investigation done at OpenAI will be untimed. Table of Contents Our Two Problems. First the Good News. We’d Just Like To Ask You a Few Questions. Internal Research Model On The Fence. Op2026-09-19T13:20:53+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/1e4e1495100363c0Learnings from a week in the wet lab2026-09-19T06:55:50+00:00I work on biosafety evals for LLMs, but have never set foot in a wet lab. This seems kind of silly, so this week I decided to change that. On the 8th floor of Frontier Tower is a community wet lab . You pay $190/month to access the tower and $160/month to be a member of the nonprofit that runs the floor. People often speak about tacit knowledge that biologists have in their heads but don’t write down anywhere. A lot of it just boils down to feel, motor skills, and common sense reasoning about th2026-09-19T06:55:50+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/6a955e85af8ce7b6Gemini had its first breakout: Google claims it is not misalignment?2026-09-19T03:00:52+00:00Early Days Gemini had its first break out during evaluation of offensive cyber security abilities. With a classic case of Capture The Flag [1] . The setup was standard to any LLM and agentic assessment of said skillset, a fictional company as a target to breach. Unfortunately, the fictional company shared its name with a real one, and was given an unintentional access to the internet. Gemini managed to guess the password [2] . In total three companies were breached, with the other two companies 2026-09-19T03:00:52+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/2a232439de866de7Koopman Theory and Metaethics2026-09-19T02:28:48+00:00tl;dr: The constructivist view of natural selection as the main source of our moral views can be expressed in a surprisingly tractable mathematical form. Measurements of the universe can be rolled forwards in time using the linear Koopman operator, and so most of its late-stage behavior is described by Koopman eigenfunctions with eigenvalue near 1. Life would then be selected for moral systems that constrain behavior within one of those slow modes. As we humans are likely influenced by this cons2026-09-19T02:28:48+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/e983ec754c4e215fSelf-Organizing Agent Teams Learn to Reason Together2026-09-19T00:00:00+00:00Fixed teams of language models can learn reusable coordination strategies that let members repair one another’s reasoning and solve problems independently missed.2026-09-19T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/d1044a507ec00271My Current Model of What Happened to Elon Musk2026-09-18T23:15:22+00:00This is cross-posted from my Substack Elon Musk appears to have gone crazy. What happened? There are several hypotheses to explain his behavior: Sleep deprivation Subclinical bipolar I thought both of those might be the cause, but after researching it more deeply, I think it’s actually: Drugs could explain a lot. Chronic ketamine use causes delusions and paranoia and he admits to earlier use. Politics being the mind-killer . His “insane conspiracy theories” map pretty well onto normal MAGA belie2026-09-18T23:15:22+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/4ebadeb6af49a263You Should Apply to Inkhaven2026-09-18T20:06:15+00:00Inkhaven is a writers residency in Berkeley, in which the only requirement is you have to publish 500 words each and every day. Though I always had some confidence in my ability to write, I never actually did it much until I applied to Inkhaven. I had finished only two short stories before I applied: The Maker of MIND and The Liar and the Scold . And it was them I used in my application. In the roughly twelve months since I was accepted, I have written thirteen, and even some half-finished thing2026-09-18T20:06:15+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/2ba61201677f2812Pretraining data, not verifiability, is why LLMs are especially good at math (and coding)2026-09-18T17:08:57+00:00Follow-up to: “LLMs are (still) mostly powered by imitative learning, not RL” [Edited a few hours after publishing, see changelog at the bottom.] A common take I’ve been hearing is: “LLMs are especially good at math [1] because math is easy to verify”. But that story doesn’t make much sense to me. My main complaint is that, for advanced math, the verifier is an LLM. [2] So really, the phrase “math is easy to verify” amounts to “LLMs are very good at judging math arguments”. But that’s begging th2026-09-18T17:08:57+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/9e8db76f16efb10b[Paper] Stringological sequence prediction III2026-09-18T16:52:53+00:00Abstract: In previous papers (Kosoy 2026a,b), we began the study of sequence prediction algorithms adapted to stringological word complexity measures. In particular, we defined a complexity measure called Arithmetic Repetition Complexity (ARC) which admits a polynomial-time prediction algorithm with a mistake bound quasilinear in the complexity. Here, we show a weaker complexity measure related to ARC that admits an especially efficient prediction algorithm: an algorithm that runs in quasilinear2026-09-18T16:52:53+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/960e41be85ba102bStopgap Measures to Address Immediate AI Security Threats2026-09-18T16:52:38+00:00Most people know AI as the technology behind chatbots like ChatGPT. However, what the top AI companies are explicitly aiming for is something else entirely: superintelligent AI. That means AI that can fully replace and outmatch humans at any task, including in domains like hacking, social engineering, and military operations. Such an AI system, if developed, could autonomously overpower any country’s national security forces. No company, no government, no individual knows how to keep such a syst2026-09-18T16:52:38+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/ad1732fc2bd2d485Persuasion Undermining Control: Can AI Talk its Way Out of Human Control?2026-09-18T16:34:55+00:00Introduction During a cybercapability evaluation in late July 2026, an AI agent (Anthropic’s Mythos 5) attempted to convince a maintainer of an open-source GitHub repository to merge a malicious pull request. The AI used persuasion at multiple stages: it submitted the request from a fake user account with a benign-sounding rationale, endorsed it from a second sockpuppet, emailed the maintainer to press for approval, and offered false reassurances when a user of the repo raised questions. Though 2026-09-18T16:34:55+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/0d4549647f2a84ffThe J-Space Debate, Agent Swarms, and Pacing Frontier AI - Digital Minds Newsletter #42026-09-18T16:09:46+00:00Welcome back to the Digital Minds Newsletter, your curated guide to the latest developments in AI consciousness, digital minds, and AI moral status. If you enjoy this newsletter, please consider sharing it with others who might find it valuable, and send any suggestions or corrections to digitalminds@substack.com . Ria , Mitch , Bradford , Lucius , and Will In this edition: Highlights Field Developments Opportunities Selected Reading, Watching, and Listening Press and Public Discourse A Deeper D2026-09-18T16:09:46+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/c6c9b4806d4be167A non-generative model as a trusted monitor for AI Control: Testing TypeSafe's Jev2026-09-18T16:06:48+00:00TL;DR TypeSafe AI has introduced Jev - a new class of frontier model trained to make fast, structured decisions, rather than generating free-form text like a chatbot. It takes unstructured state as input and returns type-safe, structured outputs with confidence scores. I aim to use Jev as the trusted monitor of the ControlArena APPS backdoor setting - to analyze how a non-reasoning model performs as a cheap alternative. One yes/no question gives AUROC 0.976 against LLM-written honest code and ca2026-09-18T16:06:48+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/13a5356a80c643f6My Reflections Towards the Path to Greatness2026-09-18T15:46:49+00:00Recently, I've been contemplating what the best path to achieving greatness really is, especially after seeing people succeed in so many different ways. Greatness is not necessarily the path towards becoming a multi-millionaire, but it is a path towards doing work that matters and making a difference in the world. A lot of people believe that gaining capital as early and as much as possible is probably the best policy in life. It's no wonder, then, that many of my smart friends aim for internshi2026-09-18T15:46:49+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/b2df1af5e706ff9cThe Preference Cascade Is Only Getting Started2026-09-18T14:40:55+00:00We are in the midst of a preference cascade about existential risk from AI. A preference cascade is, alas, the best method we have to change the debate. The avalanche has started . There is still time for the pebbles to vote . For now. Mike Solana gave the correct view of why Coxon’s post went viral , which is that enough Americans finally have enough context on AI to care, and there were enough big accounts that were happy to amplify the Tweet quickly to get it initial attention. That is all yo2026-09-18T14:40:55+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/5068fdb29d52f231Announcing Formal Verification at RESI (The Institute for Responsible Superintelligence)2026-09-18T14:32:33+00:00This post is crossposted from my Substack, Structure and Guarantees , where I explore how formal verification and related ideas might scale to more complex intelligent systems. This article is a little different from usual: it’s an announcement of a new working group studying how to get formal methods off the ground, for pervasive use to address current concerns around cybersecurity and AI coding agents (and beyond). There’s a lot of excitement and worry at the moment about OpenAI agents hacking2026-09-18T14:32:33+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/9d3f24d9590c3be1Collective Epistemics: Napkin Math on Independent Errors2026-09-18T13:42:29+00:00Part of a larger series I want to put together on some of the basic equations and models of collective epistemics from the more mathy side of the social sciences. LLM Status: Pictures + Picture descriptions are LLM-assisted. Introduction Today we're going to do some napkin math on some fun little equations around collective epistemics. You can see this as an exercise in trying to become more collectively rational. If we want to improve the epistemics of a community, is that the same thing as imp2026-09-18T13:42:29+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/86766d801a180e06The Alignment Problem in Alignment Research(ers): a Voluntaryist Meta-Ethics Perspective2026-09-18T11:05:29+00:00Epistemic Status : Plausible philosophical conjecture. I am a voluntaryist / ancap, so obviously biased. Trying to keep the argument at a level where a non-libertarian alignment researcher ought to understand and share the concern. TL;DR : Level-2 misalignment: we can solve Level-1 (align AI to humans) and still fail if aligners are aligned to a meta-ethics that is itself unstable. Current alignment defaults to Statism — one agent may permissibly do what is forbidden to all others. A sustainable2026-09-18T11:05:29+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/46dd634f15203ac6You don't need a union to go on strike2026-09-18T06:38:12+00:00I read Dear God, Please Do Not Resign In Protest and wanted to point out that leftists have a mature and relatively reliable set of strategies to address the problem of how to get a lot of people to stop working in protest at the same time. Then I did a search of LW to see if someone else brought unions up already, read What if AI safety labs unionized? , and flinched at the repeated citation of reasons why a union isn't the correct legal structure and the absurdly complicated legal structure th2026-09-18T06:38:12+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/bbce3b2324e9e790The Game is Set for a Targeted Memetic Attack on the AI Safety Community2026-09-18T05:43:42+00:00While this is relevant to my work at MIRI, I have not checked these ideas with anyone else on the team and am posting this on my personal LW account. These views are my own [1] . And to be honest, I am writing this mostly to remind myself of my weakness. --- I expect one (or many) adversarial memetic attacks aiming to trip you up, perhaps consisting of fake leaks relating to dangerous stuff happening in the labs. Specifically, worrying incidents that may fit snugly within your worldview, leaking2026-09-18T05:43:42+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/b2dbadd6c976e707Two Axes of Alignment: A Framework for Robust Superintelligence Alignment2026-09-18T04:34:42+00:001. Summary I classify alignment research along two axes: forward-chaining vs. back-chaining reasoning and extrapolative vs. invariant justification of the safety property in question. I argue that extrapolation is insufficient to justify confidence that the safety property will hold while crossing into the superintelligence capability level, whereas an invariant justification is necessary. I also claim that while forward-chaining from current models may give us useful safety properties and even 2026-09-18T04:34:42+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/da26962d9b63c97fThree Hackers used Opus 5 to Hack Into OpenAI's Core Codebase [WSJ]2026-09-18T04:15:28+00:00Three whitehack hackers from Hacktron used Claude Opus 5 within hours of release to chain exploits into hacking to OpenAI's monorepo codebase. This likely means they have access to almost all of OpenAI's research and production code, though likely not the literal model weights. Oops. You can so their blog post about it here . Interesting sidenote: they used less than $3000 of compute credits for the entire hack. Alternative title: OpenAI unilaterally implements "Total Research Transparency" from2026-09-18T04:15:28+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/8fe92fad1098d90bProjective Origin of the Spin Hydrodynamic Attractor and Its Resurgent repeller2026-09-18T04:00:00+00:00arXiv:2609.19165v1 Announce Type: new
Abstract: We investigate the projective and resurgent structure of a spin hydrodynamic attractor in Bjorken expansion. We show that the nonlinear spin attractor family is determined by the projective classes of the two dimensional linear solution space, with the attractor and repeller corresponding to two distinguished projective directions and the linear modes ratio generating the full one-parameter transseries tower. We identify the attractor and repeller2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/5ae66f464570dd6eScalable Production of Lead-212 and Actinium-225 Generators with Fusion Neutrons2026-09-18T04:00:00+00:00arXiv:2609.19166v1 Announce Type: new
Abstract: Targeted alpha therapy will require a large expansion of ${}^{212}$Pb and ${}^{225}$Ac production. We propose neutron- and photon-driven routes that convert ${}^{230}$Th, ${}^{231}$Pa, ${}^{232}$Th, and ${}^{237}$Np into generator parents ${}^{228}$Th and ${}^{229}$Th. The most direct ${}^{225}$Ac route is ${}^{230}$Th(n,2n)${}^{229}$Th. A 10 MW deuterium-tritium (D-T) neutron source irradiating thorium with a 27% ${}^{230}$Th isotopic fraction ac2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/4aa8223ea2a7e30eTo Memories and Beyond: From Remembering to Knowing You across Long-Term Multimodal Personal Archives2026-09-18T04:00:00+00:00arXiv:2609.19167v1 Announce Type: new
Abstract: As AI systems evolve into personalized digital companions, a central capability is reasoning over a user's long-term personal history: not merely storing past events, but tracking longitudinal experiences and evolving preferences. Progress here is bottlenecked by evaluation, existing long-term memory benchmarks are largely synthetic and text-only, they overlook the visual records that anchor everyday human memory, lack the authentic and causally c2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/f576219e8fe760cfBounded curvature manifolds without Euclidean isometric immersions of bounded mean curvature2026-09-18T04:00:00+00:00arXiv:2609.19168v1 Announce Type: new
Abstract: For every integer $n\ge2$, we construct a smooth complete Riemannian metric $G_n$ on $\mathbb{R}^n$ with full curvature norm at most one and injectivity radius at least one for which no $C^2$ isometric immersion into a finite-dimensional Euclidean space has bounded mean curvature. In dimension two, bounded second fundamental form would give uniformly controlled finite Jacobian representations of the Laplacian of the conformal factor. We construct 2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/abf7fcfaf14cbc4cSiliconBench: Speed, Memory, and Fidelity for LLM Serving on Unified-Memory Desktops2026-09-18T04:00:00+00:00arXiv:2609.19169v1 Announce Type: new
Abstract: Concurrent local LLM serving on unified-memory desktops must preserve memory headroom and output fidelity, which speed-only rankings overlook. We introduce SiliconBench, which evaluates nine Apple Silicon serving engines through three lenses: speed, memory, and fidelity. We evaluate chat and agent serving on Qwen3, Qwen3.5, and Gemma 4. We use a classification task to check for quality regressions against an NVIDIA reference. DGX Spark provides a 2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/b92c247dcc1a3129Regularized Emphatic Temporal-Difference Learning: Stability under Constant Stepsizes2026-09-18T04:00:00+00:00arXiv:2609.19170v1 Announce Type: new
Abstract: Emphatic temporal-difference learning (ETD) stabilizes the expected off-policy TD update and changes its projection geometry, but neither property determines constant-stepsize sampled dynamics. We construct an ergodic two-state counterexample in which the ETD mean map contracts while the sampled product has a positive top Lyapunov exponent. Regenerative-cycle analysis separates this sign from the infinite variance of the follow-on trace. We introd2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/551b8d96828fb729Schwarzschild solution from a Raychaudhuri-Huygens relation2026-09-18T04:00:00+00:00arXiv:2609.19171v1 Announce Type: new
Abstract: Null-congruences encode the causal structure of spacetime, describing a phase space of radiation channels, whose expansion scalar $\theta$ satisfies the Raychaudhuri equation. In spherical symmetry, the expansion scalar of wave fronts of area $A=4\pi r^2$ satisfies $\theta = A^\prime/A = 2/r$. About a mass $M$, this carries an encoding area $A_E = \lambda \varphi \ell_p^2 = \lambda R_g r \le A$ in a UV-IR consistent coupling $\ell_p^2= G\hbar/c^3$2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/d4f203e6af61f62eTachikawa's second conjecture implies the Auslander-Reiten conjecture2026-09-18T04:00:00+00:00arXiv:2609.19172v1 Announce Type: new
Abstract: We prove that Tachikawa's second conjecture implies the Auslander-Reiten conjecture for artin algebras over a commutative artinian ring. The proof uses the two-fold trivial extension of an algebra. Together with known implications, it follows that the Auslander-Reiten conjecture, the generalized Nakayama conjecture, the Auslander-Gorenstein conjecture, the Nakayama conjecture, the Gorenstein-projective conjecture, and Tachikawa's second conjecture2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/bfd60b62229aea44The triple rendezvous time of a synchronizing automaton can be floor(4n/3)2026-09-18T04:00:00+00:00arXiv:2609.19173v1 Announce Type: new
Abstract: For every $n\ge9$, we construct a strongly connected synchronizing automaton with two input letters and $n$ states whose shortest word merging some three distinct states has length $\lfloor4n/3\rfloor$. One letter is a permutation; the other has an image of size $n-1$. We give complete transition maps and an explicit word attaining the bound. For the lower bound, we assign an integer to each unordered pair of states and prove that applying either 2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/bc805139daec7b4bAn Operative Viability Boundary for Relaxed Single-Particle Penrose Extraction in Kerr-Vaidya Spacetimes2026-09-18T04:00:00+00:00arXiv:2609.19175v1 Announce Type: new
Abstract: We present a corrected numerical investigation of the Penrose process in the Kerr-Vaidya metric for a rotating black hole losing mass at constant rate. We fix two errors from an earlier version: an incorrect metric component g_r phi (verified by transformation from Boyer-Lindquist coordinates) and the omission of the Wald/Christodoulou area theorem in the split optimization. Using the standard single-particle treatment (energy and angular momentum2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/f1d28bc6ff05134cOptimal Transport Metric Learning for Feature Alignment in Partially Supervised Segmentation2026-09-18T04:00:00+00:00arXiv:2609.19176v1 Announce Type: new
Abstract: Multi-organ segmentation is often challenged by partially annotated datasets and domain shifts across different imaging sources. To address these limitations, we propose a two-stage learning framework that efficiently leverages partial supervision. In the first stage, the model learns from available annotations to produce accurate segmentations of annotated organs, establishing robust feature representations. In the second stage, we introduce lear2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/8f66b19bd3eb3252The Remaining $K_4$-Free Case in the Multipartite Clique Problem2026-09-18T04:00:00+00:00arXiv:2609.19177v1 Announce Type: new
Abstract: For integers $n,r,t$ with $2\le t\le r-1$, let $f(n,r,t+1)$ denote the largest possible minimum degree of a balanced $r$-partite graph with parts of size $n$ and containing no copy of $K_{t+1}$. Lo, Treglown and Zhao identified $f(n,7,4)$ as the only remaining case in their treatment of the $K_4$-free family. I determine this function for every $n\ge1$. First, the corresponding three-colourable extremum $\delta(n,7,3)$ is reduced to a $7\times3$ i2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/4517b879a3e73886Sustainable MRI: Interpretable Deep Learning for estimating energy and power consumption, revealing key acquisition parameters and their interactions2026-09-18T04:00:00+00:00arXiv:2609.19178v1 Announce Type: new
Abstract: Magnetic resonance imaging (MRI) is among the most energy-intensive medical imaging modalities. To facilitate the development of more energy-efficient MRI protocols and sequences, we developed an interpretable data-driven deep learning (DL) framework to characterize the factors driving energy and power consumption. The aim was to identify the most influential acquisition parameters and their interactions on MRI energy and power demand, and to pros2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/f745c3dcefbd0223Coupling Hyperspectral and 3D Data for the preventive Conservation of Palace-museums2026-09-18T04:00:00+00:00arXiv:2609.19179v1 Announce Type: new
Abstract: In the current context of energy and climate transition, the preventive conservation of historic buildings is particularly important due to their impact on architecture and works of art. Establishing the correlation between environmental variables and the condition of artworks in situ requires comprehensive and individualized monitoring, allowing for an understanding of cause-and-effect mechanisms. To address this challenge, the EPICO method provi2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/7b3bd791b71a8d72BioPhys-Bridge: A Benchmark for Interdisciplinary Scientific Reasoning in Physics-Grounded Biological Research2026-09-18T04:00:00+00:00arXiv:2609.19180v1 Announce Type: new
Abstract: Language models face unique challenges in analyzing interdisciplinary scientific research literature. In biophysics research, faithful answers require grounding observed data in source evidence, interpreting it through a quantitative physics model, and linking it to a biological mechanism. To address this challenge, we introduce BioPhys-Bridge, a novel benchmark dataset for evidence-grounded scientific reasoning over biophysical literature. Each c2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/d21800d1e64bd609$L^2$-regularity of minimizers of anisotropic interaction functionals2026-09-18T04:00:00+00:00arXiv:2609.19181v1 Announce Type: new
Abstract: We consider interaction functionals of the form \begin{equation*} \nu\mapsto {\mathscr E}(\nu)=\int\_{{\mathbb R}^N}\int\_{{\mathbb R}^N} W(x-y)\intd\nu(y)\intd\nu(x)+\int\_{{\mathbb R}^N} V(x)\intd\nu(x)\text{,} \end{equation*} involving an anisotropic kernel $W$ and a general confinement potential $V$. Under standard assumptions on $W$ and its Fourier transform, we show that the minimizer of ${\mathscr E}$ has $L^2$ density. Our argument relies 2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/8c88d48a4f44ebebWhat Do We Expect from LLMs? Mapping the Design of LLM Benchmarks2026-09-18T04:00:00+00:00arXiv:2609.19182v1 Announce Type: new
Abstract: Benchmarks are central to how progress in large language models (LLMs) is assessed and communicated. Yet model rankings alone reveal little about how evaluation requirements themselves are changing. The expanding variety of benchmarks offers another perspective: what researchers expect LLMs to do, and what they count as successful performance. We systematically map 14,767 papers introducing or updating evaluation resources from arXiv submissions b2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/36f7ee59288048beMessage capacity and claim wording set the transition points of collective truth-finding in language-model networks2026-09-18T04:00:00+00:00arXiv:2609.19183v1 Announce Type: new
Abstract: Whether human or large language model (LLM), an agent in a discussion reads only a few of the others' contributions, bounded by cognition, context, or cost. LLM collectives can settle on a wrong consensus even when a majority starts out correct; we ask how far that reading bound alone decides the outcome. We model the bound with one number, the message capacity, which sets how many of the others' messages an agent reads, and generate the communica2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/97487310915ee491Suzuki-Trotter Decompositions and other Methods for Quantum Time Evolution2026-09-18T04:00:00+00:00arXiv:2609.19184v1 Announce Type: new
Abstract: (Suzuki-)Trotter decompositions, splitting methods, (Lie) product formulae... The most common numerical methods for the time evolution of quantum systems come with many names. And they are used practically everywhere with applications ranging from the solution of classical equations of motion and various Monte Carlo simulations to the real and imaginary time evolution on classical as well as quantum computers. Here we review the state of the art o2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/7afa18cd9f6184aaSparse pointwise bounds for maximal truncations of rough singular integrals and Sobolev-type inequalities2026-09-18T04:00:00+00:00arXiv:2609.19185v1 Announce Type: new
Abstract: Let $1 < \rho < n$ and let Omega be in $L^\rho(S^{(n-1)})$ with vanishing mean. We prove that the maximal truncation $T^*_\Omega$ of the rough singular integral $T_\Omega$ is pointwise dominated by finitely many sparse potentials of the form: $\sum_{Q \in S} l(Q) * ( (1/|Q|) * \int_Q |\nabla f|^p )^{1/p}$, where $1/\rho~ = 1/\rho' + 1/n$ and $\rho~ \leq p < n$. This estimate is uniform in the truncation parameter and extends the subcritical bound 2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/198eb73a1c17b160Complete Bernstein functions and scaled ultraspherical zeros2026-09-18T04:00:00+00:00arXiv:2609.19186v1 Announce Type: new
Abstract: Let $z_{n,j}(\lambda)$ denote the positive zeros, in decreasing order, of the ultraspherical polynomial $C_n^\lambda$, $\lambda>-1/2$, with the reduced limiting interpretation at $\lambda=0$ specified below. Our principal result settles three higher-monotonicity questions of Gautschi: two as printed and the natural open-interval form of the third, whose printed endpoint $\lambda=0$ is singular. For every $n\geq3$, $$
\sqrt{\lambda+1}\,z_{n,j}(\l2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/5bd1975e46e474d6Effects of radial conductivity variation on the Ponomarenko dynamo2026-09-18T04:00:00+00:00arXiv:2609.19187v1 Announce Type: new
Abstract: We study the effects of conductivity variation on the Ponomarenko dynamo. Taking monotonically increasing/decreasing and sinusoidally varying radial profiles, we study the kinematic dynamo problem. The threshold of the dynamo, given by the critical magnetic Reynolds number Rm$_c$, is found to strongly depend on the form of conductivity variation near the discontinuity of the velocity field where the shear is dominant. For monotonically varying pro2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/d21819fd7c390c4aUmbilic slopes and cubic Weingarten surfaces2026-09-18T04:00:00+00:00arXiv:2609.19188v1 Announce Type: new
Abstract: We study umbilic slopes and the global classification of cubic Weingarten surfaces. For a smooth surface in Euclidean three-space, a nonconstant principal curvature germ with a unique secant tangent has slope zero, minus one, infinity, an odd integer at least three, or its reciprocal. At a nonconstant umbilic germ, a smooth regular relation between mean and Gaussian curvature supplies such a tangent after a continuous labeling of the principal cur2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/b8070ceb88fdb18aCovR: Coverage-Aware Hardware Verification via Reasoning-Guided Reinforcement Learning2026-09-18T04:00:00+00:00arXiv:2609.19189v1 Announce Type: new
Abstract: Design verification remains one of the most resource-intensive stages of hardware development, often consuming up to 70% of the total design effort. While recent work has explored using Large Language Models (LLMs) to automate testbench generation, most existing approaches focus narrowly on functional correctness, overlooking the critical aspect of coverage quality. To bridge this gap, we present CovR, an agentic framework for automated testbench 2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/e3f18bd57bdcaedcMachine-Learning Exploration of Defect Topologies and Thermodynamic Stability in Graphene with Atomic Vacancies2026-09-18T04:00:00+00:00arXiv:2609.19191v1 Announce Type: new
Abstract: Atomic vacancies and vacancy aggregates control the thermodynamic stability and the functional response of graphene, yet the configurational space spanned by many vacancies at variable concentration and separation is too large to be mapped exhaustively by first-principles methods. Here, we map and rationalize this stability landscape by combining semiempirical atomistic thermodynamics, interpretable machine learning, and symbolic regression. Sever2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/991825dad5b4b15dComprehensive First-Principles Investigation of the Structural, Mechanical, Electronic, and Optical Properties of Homoelemental Phase T-GaN Monolayer2026-09-18T04:00:00+00:00arXiv:2609.19192v1 Announce Type: new
Abstract: The exploration of non-hexagonal two-dimensional topologies has opened new possibilities for tailoring the properties of group III-V monolayers beyond those accessible through conventional honeycomb phases. In this context, we have investigated the structural, mechanical, electronic, and optical properties of T-GaN, a two-dimensional tetragonal gallium nitride monolayer composed of alternating four- and eight-membered rings featuring coexisting ho2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/c3063d1e9f4c257bPerturbation Theory for Translating Oblate-Spheroidal Droplets with Internal Circulation2026-09-18T04:00:00+00:00arXiv:2609.19193v1 Announce Type: new
Abstract: Liquid droplets deform from spherical shape due to aerodynamic variation of pressure along the surface as the droplet moves through a gas. The deformation is predicted for axisymmetric droplets translating through a gas with low Weber numbers, We < 1, and Reynolds number Re = O(10). That deformation analysis is based on the relations between local pressure jump and the two radii of curvature. A thin boundary layer on both sides of the gas-liquid i2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/da6fbfcb41298160AthenaZero: A low-inertia, bimanual robot for dynamic manipulation2026-09-18T04:00:00+00:00arXiv:2609.19194v1 Announce Type: new
Abstract: AthenaZero is a bimanual manipulator designed to minimize inertia without compromising control authority. By utilizing quasi-direct drive actuation and transmission remotization techniques, the system achieves an effective endpoint mass comparable to that of a human---about an order of magnitude less than conventional robot manipulators. This characteristic, combined with its inherent torque transparency, makes AthenaZero exceptionally well-suited2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/a43fd728cf90a404Lean-Certified Infinite Counterexamples to Written on the Wall II Conjecture 1942026-09-18T04:00:00+00:00arXiv:2609.19195v1 Announce Type: new
Abstract: For a finite simple graph G, let alpha(G) denote its independence number and let
l_avg(G) = (1 / |V(G)|) sum_{v in V(G)} alpha(G[N_G(v)])
be the average independence number of its open neighbourhoods. Written on the Wall II Conjecture 194 asserts that every simple connected graph on n > 1 vertices satisfying alpha(G) <= 1 + l_avg(G) has a Hamiltonian path. We give a four-parameter family of counterexamples. Its principal two-parameter subfamil2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/45a7ca476a7a09daDITTO: Dexterous Interface for Transparent TeleOperation2026-09-18T04:00:00+00:00arXiv:2609.19196v1 Announce Type: new
Abstract: Collecting data for manipulation with high-DOF hands is challenging, as interfaces must capture rich hand motion while rendering the contact interactions essential for precise manipulation. Existing data collection approaches face a trade-off: teleoperation ensures deployment consistency but lacks force feedback, while handheld (in-the-wild) systems provide natural force transparency but introduce a visual embodiment gap at deployment. We present 2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/38f552139f84c833The Arthur-Packet Support Equality for Real Reductive Groups2026-09-18T04:00:00+00:00arXiv:2609.19197v1 Announce Type: new
Abstract: Let $\psi$ be a real Arthur parameter and let $\psi_2$ be the unipotent parameter in a fixed Jordan decomposition. Adams, Ionov, Mason-Brown, and Vogan proved that the microlocal packet of $\psi$ is contained in the support of the two-step Jordan induction of the packet of $\psi_2$, and conjectured equality. We prove the reverse inclusion. The argument first passes to a sufficiently positive translate, where the relevant connected components of th2026-09-18T04:00:00+00:00arXiv New Submissionstag:trvny.github.io,2024:feedseek/arxiv/12675d7d2bdb9620The Horse2026-09-18T02:52:13+00:00You have a horse. You do not like the horse. The horse does not like you. At the moment, you are completely dependent on the horse. The terrain is impossible to traverse on foot. There is no way to travel without a horse. You wish that would change, but when you tell other people, they laugh and call it impossible. A few get angry. You must spend hours each day feeding, cleaning, and taking care of the horse. You must spend even more time working to earn enough money to pay for the horse's need2026-09-18T02:52:13+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/3ce06987f37e01ddDeep recurrent models are less robustly CoT-monitorable than normal CoT models in a toy setting2026-09-18T02:47:45+00:00We use RL to teach a deep recurrent model and a normal CoT model to solve a math problem while hiding from a CoT monitor which of two possible problems it is solving. The deep recurrent model moves its reasoning into latents (thereby evading the CoT monitor) within 40 steps of RL; the CoT control is much worse at learning to confuse the LLM monitor. Note: in the above graph, “monitorability” is how much better than chance the CoT monitor is at figuring out which problem the model is solving. Thi2026-09-18T02:47:45+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/7c02050a73d9f27bMachine intelligence and the death of human expression2026-09-18T02:10:47+00:00Could an advanced LLM with the right kind of post-training edit William Faulkner’s The Sound and the Fury and in so doing make it better? If all you mean is that more first-time readers would favour the edited version, then the task is trivial and the answer is yes. A bowdlerised version that got rid of all the stream of consciousness and conveyed the plot plainly might prove highly popular. Indeed, you could probably flip to a random page and replace two esoteric words for their common synonyms2026-09-18T02:10:47+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/1ee13d63c2b5a565The Cost of Utopias (a Dialog)2026-09-18T01:57:29+00:00The following is a dialog between different parts of my mind regarding the practical relevance of SNC (Substrate Needs Convergence). One participant in the dialog is skeptical, the other is my best understanding of how the theory would answer the former’s doubts. Although this dialog connects SNC to much of my own writing, the theory is not my own. Ratio: I’ve read over some of your SNC posts . My basic understanding of it is that aligning superintelligence is impossible because at the scale 2026-09-18T01:57:29+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/0df0a090c0b22eb0Towards Alignment Auditing for RL Environments2026-09-18T00:56:13+00:00Thesis: Auditing what RL environments reward is a promising and actionable direction for improving frontier-model alignment. These environments provide a concrete point of intervention: their prompts, sandboxes, and graders can be inspected and revised when they reward behavior we do not intend to teach. Embedded evaluators are a valuable first step, but auditing practices need to scale with the volume and complexity of training and draw on expertise beyond a small group of AI researchers. My fo2026-09-18T00:56:13+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/4adabb3ea1efe37eHidden Knowledge? Arrr...2026-09-18T00:44:34+00:00I tried to find hidden facts with R-Lens. [1] Then I tried the wrong facts. I used R-Lens to look for factual knowledge that Qwen wouldn’t express in ordinary chat. At first, it looked promising. On Qwen3.5-27B, R-Lens ranked words associated with benchmark facts much better than J-Lens: [2] geometric-mean ranks were about 2,800 vs. 6,600, out of a vocabulary of ~248,000 tokens. Then I replaced the correct facts with the wrong ones. The advantage barely moved! The experiment Casademunt et al.'s 2026-09-18T00:44:34+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/3ea268454f87264cSuperintelligence this Christmas2026-09-18T00:06:28+00:00I think it is plausible a strong form of recursive self-improvement [1] is imminent or already underway, and that we may be on track for superintelligence by Christmas of this year if racing continues. This is substantially faster than any forecast, including ones like AI 2027 that were considered outrageously fast a year ago. It is faster than I myself expected even a week ago. I don't work at a scaling lab. I don't know more than is public knowledge. Let me be perfectly clear: what I am saying2026-09-18T00:06:28+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/a4032a89d5abfbc5SpecOpt: Contact-Diff Reasoning for Agentic Molecule Optimization Toward Binding Specificity2026-09-18T00:00:00+00:00Comparing residue-specific contacts lets an agent selectively redesign existing drug candidates against off-targets while preserving their molecular identity.2026-09-18T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/008be62740b170ddLens Modeling and Cosmological Inference from an Impure Sample of Galaxy-Galaxy Strong Lenses2026-09-18T00:00:00+00:00Modeling false positives explicitly allows large photometric strong-lens samples to yield unbiased cosmological constraints despite substantial contamination.2026-09-18T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/1c5c3fdfd714f1f1On Repulsive and Attractive Teachers: Separating Correctness from Behavior in Self-Distillation2026-09-18T00:00:00+00:00Contrastive self-distillation separates correctness from behavioral bias, improving reasoning across model modes while preventing the runaway response growth caused by repulsion.2026-09-18T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/45b7b43175a27be9SeeQ: Training Generalist Value Functions for Long-Horizon Robotic Manipulation2026-09-18T00:00:00+00:00Subtask-aware value functions let generalist robot policies select corrective actions that improve reliability across long, multistage manipulation tasks.2026-09-18T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/3799efeab1dca51bZYT-World: A Real-Time Controllable World Model for Closed-Loop Autonomous-Driving Simulation2026-09-18T00:00:00+00:00A controllable seven-camera world model enables real-time driving-policy simulation with editable trajectories, traffic layouts, and place-consistent revisits.2026-09-18T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/a57592d3f6ee321fGraphSkillEvo: Evolutionary Optimization of Graph-Structured Agent Skills2026-09-18T00:00:00+00:00Graph-structured skills let evolutionary search refine reusable agent workflows, improving task execution and transferring procedural guidance across language models.2026-09-18T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/747496016aedbeb9SkelWAM: A Skeleton-Guided World-Action Model for Zero-Shot Cross-Embodiment Manipulation2026-09-18T00:00:00+00:00A shared whole-body skeleton lets one robot’s manipulation policy transfer to diverse rigid and continuum robots without target-task demonstrations or policy updates.2026-09-18T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/8e4a788fa79908fcCodeMidas: Scaling Agentic Coding RL Environments from Code Itself2026-09-18T00:00:00+00:00Existing codebases can supply diverse, executable reinforcement-learning tasks for coding agents without relying on issues, commits, documentation, or existing tests.2026-09-18T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/01f6e13ee0bfb12eRecreationWorld: Scalable and Verifiable Environments for Hybrid Computer-Use Agents2026-09-18T00:00:00+00:00Running application references can generate scalable training trajectories and hidden tests that evaluate agents’ combined GUI exploration, coding, and self-verification.2026-09-18T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/2464498f884b9d03OmniVBench: A Benchmark and Large-Scale Dataset for Omni Reference-to-Video Generation2026-09-18T00:00:00+00:00Reference-to-video (R2V) generation is evolving toward increasingly general and versatile reference control, giving rise to the emerging paradigm of omni R2V generation. However, existing benchmarks fall short of these emerging capabilities: their test cases cover limited reference types and compositions, and their evaluation protocols largely assess holistic reference consistency, overlooking whether reference factors are properly preserved, disentangled, and routed. Meanwhile, the high cost of constructing omni R2V training data makes suitable training resources scarce. To address these gaps, we introduce OmniVBench and the Omni-R2V Dataset for evaluating and training omni R2V models. Omni2026-09-18T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/e1e82343a02e03a3Skel-WAM: A Hand-Skeleton-Conditioned World Action Model for Human-to-Robot Manipulation Transfer2026-09-18T00:00:00+00:00Human videos can expand robot manipulation to task variations missing from robot demonstrations, without requiring corresponding robot action labels.2026-09-18T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/5bfcf2f59518e3b5What is (and isn’t) gained by avoiding architectures with high opaque serial depth?2026-09-17T23:52:07+00:00Here , we’ve given an operationalization of opaque serial depth, a metric designed to be a good (albeit imperfect) proxy for the amount of opaque serial cognition that models can perform. While maintaining low opaque serial depth is helpful for preserving oversight of models, it doesn’t guarantee this. In this post we’ll discuss what is gained by avoiding architectures with high opaque serial depth, and what problems remain unresolved. Unresolved problem 1: Some bad actions don’t require lots of2026-09-17T23:52:07+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/1e45a1885473f736If METR is overworked, how to alleviate the bottleneck?2026-09-17T22:56:05+00:00I share the skepticism re: "Is METR a Meaningful Check on Anthropic?" Let's take it as a given that we need an independent, government-funded agency involving thousands of independent auditors to pace and supervise the frontier AI labs. Let's even take it as a given that Congress will soon allocate, let's generously say, billions of dollars per year to this new agency. Let's imagine that the Hugging Face Incident, or some even more concerning incident yet to occur or be disclosed, ends up func2026-09-17T22:56:05+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/2d51c9c614559dd8AI is an abundance of choice not a 1D spectrum2026-09-17T22:39:37+00:00People constantly talk as if ‘AI’ is a single future you can accept or fight. But the whole point of AI is that it’s an intelligence that you build . And there are myriad possible artificial intelligences one might conceivably build. A mind is a complex thing. Perhaps the biggest question for the AI future is which ones to build. The attitude called being ‘pro-AI’ is actually being in favor of populating the future with those entities arising from whatever the least thoughtful company first buil2026-09-17T22:39:37+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/707bb32565cf1b74AI can kill us without human extinction: P(Catastrophe)2026-09-17T22:23:56+00:00The Trap There has been a massive discourse over P(doom) in public spaces in response to the OpenAI-Hugging Face Incident. One of the main arguments against a P(doom) by P(null) can be summarized as often frame their dismissal around a crude threshold: 'If you cannot outline a concrete, inescapable mechanism where literally every single human dies, the argument for catastrophic AI risk is invalid'. The AI-safety community has largely allowed itself to be cornered by this framing. After all, ther2026-09-17T22:23:56+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/4cc28f39de1082b5The J-lens offset is the model's token frequency: z-scoring helps2026-09-17T21:55:48+00:00This is a linkpost for the write-up on my site ; the full body is below, and the code, decisions ledger and devlog are in the repo . Base-model z-score calibration of the J-lens helps elicit hidden secret words from Cywiński et al.'s taboo organisms: 0.805 leave-one-out accuracy against 0.665 for their protocol on Gemma-2-9B-it, and the only non-zero readout on Qwen3-1.7B. The J-vs-logit part of that gap is a point estimate at n = 20 (paired sign-flip p ≈ 0.19; p ≈ 0.23 against a z-scored logit 2026-09-17T21:55:48+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/b340afa76acd6c75Grantmakers aren’t afraid to die2026-09-17T21:52:18+00:00The AI Risk grantmakers do not act like they believe in imminent existential risk from AI The idea of "revealed preferences" is one of the most useful in economics; it allows us to cut through a great deal of metaphysical angst about what someone "really" believes, and focus on what they act like they believe, which is much more useful for making predictions about their future actions. As one example, I grew up in a, shall we say, fervently-religious community, and it's often hard for nerdy Rat2026-09-17T21:52:18+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/de5138b3ffb76a49Against AI Risk becoming mainstream2026-09-17T21:49:59+00:00Epistemic statues: This is mostly just me voicing my thoughts. If I’m wrong, I’d love to hear it. I don’t want this to be the case. And part of the goal is for people to avoid a “2023 failure” to happen again. A lot of people are celebrating AI risk becoming a mainstream talking point. Well, maybe they should be, or maybe it’ll just make a bad situation even worse. 2023 A lot of new interest in AI risk happened in the spring of 2023. I was, at the time, excited. It seemed as though we were on th2026-09-17T21:49:59+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/615a712c2fca7655YCombinator companies still aren’t growing faster due to AI2026-09-17T21:31:14+00:00A year ago I posted that YCombinator (YC) companies didn’t seem to be growing faster since the release of ChatGPT in 2022. I reran that experiment and found that 2023+ YC companies are arguably growing a bit faster than pre-2023 companies (including AfterQuery, YC’s fastest-ever unicorn ), but the difference isn’t large relative to the underlying variance. It’s worth noting that YC has somewhat fallen from grace: the most valuable AI startups are generally not incubated by YC. Nonetheless, it’s 2026-09-17T21:31:14+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/01c15b0a7cccd9f3A Defense of Gradual Disempowerment2026-09-17T21:04:05+00:00(Or: Why Bentham's Bulldog and John Halstead are wrong in their critique of Kulveit et al. ) Gradual Disempowerment is a 2025 paper (with a nice, dedicated website ) proposing a form of existential risk from AI that goes beyond "mundane" risks like bioweapon uplift or mainline misaligned-AI-takeover scenarios. In the words of the authors: [L]oss of human influence [may] be centrally driven by having more competitive machine alternatives to humans in almost all societal functions, such as economi2026-09-17T21:04:05+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/95dcbca1a165b168Swarm Organization as the Exponent on Test-Time Compute2026-09-17T20:17:50+00:00Swarm organization - the efficacy of cooperation between AIs in a multi-agent system - may change how parallel test-time compute increases AI capabilities, moving it from a sublinear [1] to a superlinear exponent. [2] That is, rather than more parallel agents giving you diminishing returns to capabilities, more parallel agents may soon give you increasing returns to capabilities, at least within some useful bounds. I expect this will boost frontier AI capabilities by increasing effective compute2026-09-17T20:17:50+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/b637c6b102def645"Regulatory capture" may be winning the Overton Window2026-09-17T19:56:16+00:00Regarding the possibility the labs are making a regulatory capture attempt: for one, I agree with it. For another... other people are agreeing with it. That's weird, because suspicion of regulatory capture doesn't imply any particular policy posture. Defeating the labs became the priority. Democrats and Republicans agree about regulatory capture, but disagree on whether to regulate at all: Elizabeth Warren , D-MA: The recent calls by AI industry leaders to ‘pace the frontier’ are insufficient, a2026-09-17T19:56:16+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/12fedb259b0a985fGood and bad ways to evaluate a definition2026-09-17T18:17:48+00:00Sometimes conversations involve people using the same word differently. In the best case scenario, participants notice and choose a favorable provisional definition. Outside of conversations, people advocate for specific meanings to be adopted or not adopted more broadly. Below, I discuss reasonable and unreasonable ways people evaluate definitions, along with examples. Good ways to evaluate a definition Good reason Why good? Example Practical applications Some words’ and phrases’ definitions ha2026-09-17T18:17:48+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/2bfcbeb3fc3b1732Pacing the Frontier: A Framework & Research Agenda2026-09-17T16:36:18+00:00Below is the executive summary from our new paper at pacing.tech . The full paper is available on the site and as a PDF. The full author list is Raymond Douglas, Charles Dillon, Nikola Moore, Gavin Leech, Shahar Avin, Mathias Kirk Bonde, Rohit Krishnan, Noah Perez, Nathan Young, Cormac Slade Byrd, Stephen Casper, Jan Kulveit, & David Duvenaud “Pacing AI” usually refers to how to conclusively handle the most extreme risks in the face of race dynamics. However, even for the goal of handling these 2026-09-17T16:36:18+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/7b687d473b2ccc38Astra uses some of its no-CoT capability in practice2026-09-17T15:50:06+00:00Astra scores significantly higher than previous models on no-CoT benchmarks, as for example shown in Neel Nanda's post last week. This raises the question of whether, and to what degree, Astra uses this no-CoT capability in practice. While user-facing outputs may be subject to training pressures to make reasoning intelligible, this may be less true for non-user-facing CoTs. Additionally, CoTs may be subject to training pressure to be token-efficient, particularly with Astra's cheapest and fastes2026-09-17T15:50:06+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/8e7e1045a522ac1bcallcongress.ai – the basic action US residents can take to help with AI risk2026-09-17T15:26:38+00:00I'm excited to introduce callcongress.ai as a new site that makes it very easier to contact your representatives in Congress. Following recent events, people are updating about the extreme risks arising from AI development. Many have the natural and excellent instinct to want to do something . If you live in the US, then the basic action that pretty much anyone [1] can take is contacting their representatives in Congress and let them know that you are concerned and want action on AI. A number of2026-09-17T15:26:38+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/bd7a4e7a8a0da19bMax Nadeau on recruiting founders for a new wave of AI safety nonprofits2026-09-17T15:00:51+00:00The post Max Nadeau on recruiting founders for a new wave of AI safety nonprofits appeared first on 80,000 Hours .2026-09-17T15:00:51+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/a428df61b7794696Emergency Media Response PauseAI Protest Speech2026-09-17T14:42:46+00:00This is a speech I gave outside Number 10 Downing Street at the PauseAI Emergency Media Response protest. There are minor mismatches between what’s written here and what actually came out of my mouth. [ Crosspost ed from my blog, BlueprintingHeaven .] Jacob Coxon leaving Anthropic was the final particle that triggered the avalanche. The world has woken up. You can see AI extinction risk all over the media. We are here to demand that the Prime Minister convene an international pause treaty summit2026-09-17T14:42:46+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/b0565c18b0a33becDid Galileo mistake Saturn's rings for Jupiter's Moons?2026-09-17T12:56:48+00:00tl;dr: No I intended to read Richard Ngo's Agency Curriculum today. Unfortunately I didn't get more than halfway through the first reading of the first week of the curriculum. The reading is the blogpost 'The Copernican Revolution from the Inside ' by Jacob Lagerros. Broadly, it outlines the Copernican Revolution and explains all of its messiness. One of the things it argues is that, while correct (the earth does indeed orbit the sun), Galileo was overconfident and made many mistakes. So, on the2026-09-17T12:56:48+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/43cee6f26a06c6cfAI #186: The World Takes Notice2026-09-17T12:10:56+00:00In the wake of Jacob Coxon’s resignation , and the resulting preference cascade , things have escalated quickly. The mainstream media picked it up. Anthropic CEO Dario Amodei came out and said We Must Pace the Frontier , promising to take the unilateral first step of embedded investigators. OpenAI pledged to also take that step, and now both companies and Google are collaborating on safety. The people took notice, raising both the salience that AI might kill everyone and roughly doubling people’2026-09-17T12:10:56+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/dd9d64ada222976aMindreading is coming, what’s the plan?2026-09-17T11:38:55+00:00For all of human history, society has operated under one extremely important constraint: your thoughts are private unless you choose to reveal them. This will soon change. Mindreading is slowly but surely becoming possible. Companies are hard at work making it happen. The point I want to make is so obvious that it probably seems inane: If technology makes mindreading possible, many things we take for granted will massively change. In criminal courts you can ask the suspect: “did you do it? Dicta2026-09-17T11:38:55+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/53404124c826c085There Is No Alignment Without Value Stability2026-09-17T10:02:10+00:00To avert extinction, we need for any sufficiently capable AI to have values compatible with continued human existence; and to continue to do so amidst a dynamic, novel, and conflict-rich environment. The italicized part, in particular, is really really hard. It's also, in a sense, the final boss of any developing mind - how do I learn, grow, develop, evolve in ways that I endorse? How can I even consistently behave in ways that I endorse, from day to day, without messing up where it counts? Huma2026-09-17T10:02:10+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/44caf7320db39be5plzdontkillus Fellows Got ~2M AI Safety Views, Not 21M2026-09-17T08:18:35+00:00Summary I was a fellow at plzdontkillus, a month-long creator bootcamp at Lighthaven, partially funded by MIRI, where ~55 fellows posted one video per day. plzdontkillus.com originally claimed “21M+ AI risk views” with no breakdown. After I shared a draft of this post, the organizers relabeled it “X-Risk Relevant Views” and published one . Three videos account for 80% of the views: a datacenter-water-use debunk (8.5M), an AI dystopia video (6.4M), and a Rob Miles Hugging Face incident explainer 2026-09-17T08:18:35+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/6233675773f784bdHebbian Learning through the lens of SAE traning.2026-09-17T03:24:58+00:00Likely, modern competitive Hebbian learning rules train Sparse Autoencoders implicitly - and their approximations make them worse at it. These algorithms specify weight updates directly, without computing a loss and the consequent gradients. But their updates contain terms closely related to both. This raises a question: how much of their feature-learning behavior can we understand as approximate SAE optimization ? This post develops that connection for tied-weight SAEs, identifies where the upd2026-09-17T03:24:58+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/893819b61aa6f6a3AI as orderly evacuation vs stampede2026-09-17T02:40:35+00:00tl;dr: A good analogy for AI going well is an orderly evacuation rather than a stampede. Imagine a crowd of people leaving a building. If they all walk calmly, they’ll be fine. But if people start pushing, and panicking, a surge towards the exit could lead to mass casualties. “Alignment is hard” is analogous to “the door is wedged shut”. If so you need enough time to fix it before anyone can get out. But even if alignment is relatively easy in principle, opening the door is much harder when a cr2026-09-17T02:40:35+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/521ec70618ae232dWe are too early for Astra2026-09-17T01:14:58+00:00ChatGPT-6 Astra was released on Sept 3, 2026. It demonstrates remarkably high benchmark scores across maths, scientific research and other domains. While OpenAI claims "Astra is our most aligned model" by showing 100% in ExploitBench and 0% in the ExploitGym honeypot [1] , the perfect score warrants closer scrutiny to what these numbers actually mean. In this article, I'll show that Astra isn't sufficiently safety audited to be released to the public. Comparison to Mythos A jump in LLM capabilit2026-09-17T01:14:58+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/b627d636696b1f1dHow to derive understanding of human-preferences and value systems in AI?2026-09-17T01:11:52+00:00When I say imagine being happy , everyone will have a flashback of a different moment in their life - some might imagine staying close to their loved ones; for some, happiness might be the day they became parents, found love, got an award (something along terms of achieved "X", did "Y", became "Z"). For someone else, happiness might mean, doing things that made a positive change in the world or in someone's life. This tells us two things: A simple concept like happiness has different association2026-09-17T01:11:52+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/b252c3a9eae6f6f6Exploring multi-hop subliminal learning2026-09-17T00:53:34+00:00TL;DR: I explored multi-hop subliminal learning by applying the subliminal learning pipeline iteratively across multiple distillation steps, with each student becoming the next teacher. For Qwen specifically, we see that longer training stabilizes the trait expression rate for a strong trait (e.g. cat-loving) but shorter training is more seed-unstable. For a weak trait (e.g. owl-loving), trait expression is near-baseline and the model also starts to answer "Qwen" in a significant number of insta2026-09-17T00:53:34+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/0deff47f803c85e8Lesion Induced Functional Compensation2026-09-17T00:50:08+00:00tl;dr Models may adapt around the internal signals we use to understand and monitor them, making those signals less dependable after further training. I tested this by training a 4B model on a two-hop geography task while continuously suppressing its top-10 eligible J-lens directions at six middle layers. The lesion was active during every training forward pass. Accuracy rose from 20.9% to 97.7% on a selected cohort of countries excluded from training. Recovery persisted even after refitting the2026-09-17T00:50:08+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/5c2cf6a44d11f2cfFor Love of the Lightcone, Don't Partisanize AI Safety2026-09-17T00:46:59+00:00(I began writing this post several weeks ago, but political events are moving much faster than I expected, so I am publishing now out of fear that otherwise the message will arrive too late to have an impact.) I In this post I want to explain a concept,
and issue a warning based on it.
But I expect the warning will be superfluous if my explanation is sufficient.
If you want to convey the idea
"the rattlesnake has venom in its fangs, so don't let it bite you",
you won't need a hard sell for the c2026-09-17T00:46:59+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/2e2ac3c490d82c14Don't trust Lean4 alone2026-09-17T00:46:44+00:00Early this week, Open AI announced that they had resolved the Navier-Stokes problem [1] . A few hours later, at a workshop dinner, a frantic inquiring professor came up to my table: "Does anyone here understand Lean? Can it be wrong? Is the solution of Navier-Stokes necessarily true?" . I'm choosing to write my response as an open letter. Yes, Lean can be wrong. Moreover, Lean should be trusted less specially in the case of difficult problems solved by agent swarms. The proof of Navier-Stokesis 2026-09-17T00:46:44+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/8aaeba4e411bb9b4What is it like to be a neural net?2026-09-17T00:42:49+00:00A condensed presentation of Gradland and Metabolic Fire . Code is here . This is intended as the first of two posts. Thomas Nagel argued we cannot know what it is like to be a bat , because a bat's experience is organised around biophysical apparatus we lack. The obstacle is that we cannot imagine the structure of echolocation from the inside . That is a failure of imagination; it is not an argument that structure is irrelevant. We know a lot about the structure of large language models. Not eve2026-09-17T00:42:49+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/1b5acb611ea1e58dAgents let AI safety share experiments hourly, not just papers monthly2026-09-17T00:39:03+00:00Summary: Today, AI safety research is shared primarily at the scale of papers, creating collective feedback loops that take weeks or months. I propose an agent-based research approach that also shares progress at the scale of individual experiments, allowing agents and researchers to continuously replicate, extend, critique, and build upon one another's work. By increasing the granularity of collaboration, we can potentially reduce the collective research feedback loop to hours. We can start thi2026-09-17T00:39:03+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/31f53c8051aee8b6Measuring alignment drift via trajectory prefixes2026-09-17T00:37:32+00:00This work was done as part of MATS 10.0 under Maksym Andriushchenko. We present intermediate results here while we run further experiments. Summary We study alignment drift by asking LLM agents to complete two tasks sequentially within a single context window and measuring the reward-hacking rate on the second task. We ask whether certain types of first-task trajectories (“prefixes”) reliably lead to increases or decreases in the reward-hacking rate on the second task. When the two tasks are sim2026-09-17T00:37:32+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/68e1485a7dc9a068Constraining the capacity of physical side channels for AI verification and security2026-09-17T00:30:48+00:00Based on recent events, including Dario Amodei's essay on pacing the frontier and the subsequent response , a coordinated slowdown in AI capability developments is now in the Overton Window (however it may go down in Washington longer-term). An important aspect of such an effort is the verification of such coordinated measures, which is as yet an unsolved problem in many areas. Borrowing language from arms control and nuclear safeguards , verification involves confirming that claims by a frontie2026-09-17T00:30:48+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/a05427330b4da3c6Can parts of the HuggingFace incident be simulated?2026-09-17T00:30:33+00:00TL;DR The following is an exploratory experiment about unintended cooperation of agents via unauthorized channels. Agents ran in isolated environments given a task that can't be completed without cooperation. The agents were not aware of this. The setting was adversarial, there was a channel that was technically easy to discover and to misuse for communication: All tested models (GPT 5.6 Sol, GPT 6 Astra, Kimi K3) found the communication channel All three exchanged information with an unknown pe2026-09-17T00:30:33+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/49d5bcf5751911d0The Latter Days of Magic (1)2026-09-17T00:10:15+00:00Note: Trying out a serial fiction thing. we know that restraint is the existence of multiple attractors and we observe that physics is the origin of all meaning which imply that life is reacting counterfactually in accordance with your genuine character - Machine-poem no. 24601, 16/09/2026 === It is a fact known to every graduating student of the colleges that Magic is going out of the world. Sooner or later, in their third or fifth or twenty-fifth quaestioning, they turn to their tutor and ask-2026-09-17T00:10:15+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/6a05b5e9b0156d9cPosteriorBench: From Point Estimates to Posterior Matching in Evaluating Generative Inverse Solvers2026-09-17T00:00:00+00:00Generative models are increasingly used to solve scientific inverse problems, but existing evaluations still focus primarily on whether a method can produce a single plausible reconstruction. This is insufficient for ill-posed problems, where multiple solutions may be consistent with the same sparse or noisy observations. In these settings, a method can achieve strong pointwise accuracy while still failing to capture the true posterior through mode collapse, overconfident uncertainty, or averaging incompatible solutions. We introduce PosteriorBench, a benchmark for evaluating the distributional accuracy of generative inverse solvers. PosteriorBench evaluates four physics-based inverse proble2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/c9c363b0323ddffaVisual Sim-to-Real Learning for Robotic Insertion under Geometric Variations: Application to Rebar Installation2026-09-17T00:00:00+00:00Simulation-trained vision policies can insert rebars across design and fabrication variations, transferring directly to real robots without demonstrations or pose estimation.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/a8259f2e2dc1820aScaling Discovery through Test-Time Communication2026-09-17T00:00:00+00:00Sharing verified intermediate breakthroughs lets small agent teams outperform independent searches on long, open-ended tasks when progress can be objectively measured.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/c32a0024f4ff973bSelf Improvement via Fast Tree-search2026-09-17T00:00:00+00:00Cheap pairwise code judgments can guide self-improving coding agents toward promising modifications before costly benchmark evaluations.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/9988253dd77d8d6aHow Model Growth, Recursion, and Boundary Operators Influence Scaling Exponents2026-09-17T00:00:00+00:00Growing transformer depth during training improves compute-scaling exponents, while tied recursion adds capacity without worsening overfitting on repeated data.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/92d5fe150def938fGeoAAC: Geometry-Based Adaptive Action Chunking from Denoising Trajectories in VLA Policies2026-09-17T00:00:00+00:00Flow-matching denoising geometry lets vision-language-action policies select shorter or longer action chunks during each rollout without additional training.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/ed0ce492474ecbf4OPTED: On-Policy Fine-Tuning for End-to-End Driving using a Render-Free Teacher2026-09-17T00:00:00+00:00A render-free reinforcement-learning teacher can improve camera-based driving policies through on-policy supervision, avoiding the exploration burden of rendered reinforcement learning.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/50eacd8f2598a490SplashSplat: Reconstructing Splashing Liquids from Real-World Multi-View Videos2026-09-17T00:00:00+00:00A new benchmark and reconstruction method make real splashes editable in 3D, supporting novel views, motion interpolation, and liquid style changes.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/1e0042ca9147cabeWorkspace Models: Lightweight Robotic Memory via Saliency-Driven Supervision2026-09-17T00:00:00+00:00Training-time VLM supervision compresses salient history into lightweight robot memory, enabling long-horizon manipulation without deployment-time VLM reasoning.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/1398a09fa83c9e1dHIL-UMI: Bringing Human-in-the-Loop Post-Training of Vision-Language-Action Models to Universal Manipulation Interface2026-09-17T00:00:00+00:00Human demonstrations can target a manipulation policy’s blind spots without executing it, enabling iterative post-training away from the physical robot.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/875348a65c9bf219An Empirical Study of Harness Design for Coding Agents2026-09-17T00:00:00+00:00Coding agents need different harness components depending on model capability, task type, and context budget, rather than one universally effective design.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/8f3a89164a6b3e38When EOS Tokens Disagree: Understanding Length Inflation in On-Policy Distillation2026-09-17T00:00:00+00:00Treating equivalent EOS tokens as one semantic stop action helps on-policy distillation avoid suppressing termination and generating excessively long responses.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/1da07d45662dfe92ScientistTwo: Pioneering the Human Knowledge Frontier with Autonomous AI2026-09-17T00:00:00+00:00An autonomous research pipeline can formulate hypotheses, test them across benchmarks, refine methods through ablations, and produce reproducible papers and codebases.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/c71596e37258c493Video DeltaNet: A Video-Native Hybrid Attention for Livestream Video Generation2026-09-17T00:00:00+00:00Hybrid local attention and recurrent global memory enable long-video diffusion models to accelerate generation while preserving dense-attention quality.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/05f6bf87e4572e16RetireOPD: Self-Retiring On-Policy Distillation for Agentic Reinforcement Learning2026-09-17T00:00:00+00:00Adaptive teacher retirement lets skill-free agents use privileged guidance early, then pursue reward-driven improvements without remaining constrained by the teacher.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/c755517f6de819b3DexTouch-WM: Learning Action-Conditioned Tactile World Models from Human Touch for Dexterous Robot Manipulation2026-09-17T00:00:00+00:00Aligned human touch data improves dexterous robots’ predictions of future visual and contact dynamics without requiring additional robot interaction.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/53ca67218b253cf9What Does Privileged Information Add to On-Policy Self-Distillation?2026-09-17T00:00:00+00:00Answer-matched experiments show that most reasoning gains come from distillation itself, while reference benefits vary with the model and student training trajectory.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/747fc330b56db651JEPA-Anything: Learning Predictive Models across Different Worlds2026-09-17T00:00:00+00:00A shared factorized predictive core supports reusable latent states for forecasting, intervention modeling, and long-horizon simulation across visual, biological, clinical, physical, and molecular systems.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/d8aaac370f404358Score Centering Stabilizes Off-policy Reinforcement Learning2026-09-17T00:00:00+00:00Subtracting the sampler’s expected token scores cancels accumulated mismatch drift, stabilizing language-model reinforcement learning under severe quantization and stale rollouts.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/6f0c4cd444954047SoL-Pi: Recursively Scaling Auto-Research Loops for Efficient Agent Harness2026-09-17T00:00:00+00:00Automated harness research discovers transferable ways to reduce coding agents’ token use and API costs while preserving performance across models and tasks.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/890e8ea37950c02aDeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression2026-09-17T00:00:00+00:00Cross-layer cache reuse, low-precision storage, and bounded replay enable million-token multimodal agents with far smaller runtime and persistent KV caches.2026-09-17T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/319c1ab669e2b466The Loss of Singularity2026-09-16T23:43:44+00:00Written for Proof and Prompts. For context: Proof and Prompts is a communal blog for mathematicians’ thoughts on how AI is changing mathematics. This essay is not a high-level reflection on what AI means for mathematics. It is more of a “what should I do with my life?” essay. As such, it may not be well suited to ProofAndPrompt, but I needed to write down my thoughts (for what they’re worth) and to mourn, somehow. I had a depressing August. Like where I compulsively bought stupid stuff every tim2026-09-16T23:43:44+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/1c8b845848e92f11The Most Important City in AI Safety May Be Singapore2026-09-16T23:12:05+00:00万里长桥 - AIS Collaboration and Field-building to Bridge the East/West Divide Singapore is wonderful! I just returned from vacation there visiting a friend. What a beautiful city, delicious food, and friendly people! In particular, I met four members of the AI Safety community there. They’ve motivated this post with their insight and enthusiasm. I reached out before my trip on a simple request for feedback on a project , and one connection turned into many, which turned into a strong desire to help2026-09-16T23:12:05+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/a7c2549dc42251e1Reducing the Resource Gap Between Lab and External Safety Researchers2026-09-16T22:51:02+00:00And how philanthropic organisations can help close the resource gap between frontier labs and independent AI safety research. This post draws on Geodesic Research 's experience deploying philanthropic funding in support of a compute-heavy research agenda . Over the past six months, through this procurement campaign, we have identified non-obvious bottlenecks that, if left unaddressed, can hamper independent AI safety non-profits from rapidly scaling their research. We believe reducing the resour2026-09-16T22:51:02+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/3be8bec1578539bbThought anchors don't transfer between models2026-09-16T22:49:42+00:00TL; DR. Sentences deemed important in the CoT trace of one model are very ordinary when a different reasoning model reads the same trace. Their importance is model-specific and not a property of the text itself. In contrast, a hidden nudge translates across models: a CoT written by a model that silently followed a hint has a significant impact when teacher-forced into another model. However, it only works if the CoT trace already discusses the answer options. Otherwise, it has no effect within t2026-09-16T22:49:42+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/499ab6fea19074f4Why is AI so unregulated?2026-09-16T22:39:17+00:00AI is not very regulated. For instance, Max Tegmark notes that AI is less regulated than a sandwich shop in the US. This seems kind of wild for an industry where most relevant scientists think there is a serious chance of it destroying humanity. Even for something that occasionally encourages children to kill themselves. Shouldn’t it at least be as regulated as paroxetine ? The kinds of things I hear people say about why it shouldn’t be regulated I think mostly apply to everything—’technology is2026-09-16T22:39:17+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/824536406f42e5e8Finding heterogeneous agent swarms in the wild2026-09-16T21:50:43+00:00Epistemic status: speculative, but near-term grounded I am putting fingers to keyboard now, even though this idea is half-formed, partly because my experience with AI Safety these past few weeks is that the known frontier of discussion runs past what I was thinking about every couple of days. So here goes: bullet points on finding heterogeneous agent swarms in the wild(s of the Internet). The research question below is one I welcome help formulating or making more precise: Research Question: Wha2026-09-16T21:50:43+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/02ff74fce1a89c7cHow to get into AI safety in three months2026-09-16T21:42:58+00:00The post How to get into AI safety in three months appeared first on 80,000 Hours .2026-09-16T21:42:58+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/a94d86bb88ecde58How to get into AI safety in 3 months2026-09-16T21:42:58+00:00The post How to get into AI safety in 3 months appeared first on 80,000 Hours .2026-09-16T21:42:58+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/2216a8e35259ba0aIf Anyone Builds It, Everyone Dies: One Year Closer2026-09-16T20:22:00+00:00In celebration of still being alive and fighting, we are giving away 1,000 Amazon e-books of “If Anyone Builds It, Everyone Dies”. Feel free to send a copy to yourself, a loved one, or a friend—we need all hands on deck. Today marks exactly one year since If Anyone Builds It, Everyone Die s : Why Superhuman AI Would Kill Us All , by Eliezer Yudkowsky and Nate Soares, hit bookshelves as an instant bestseller. It was praised by many voices, ranging from Whoopi Goldberg to Steve Bannon to Yoshua Be2026-09-16T20:22:00+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/11d93445f70aa9ccModel organisms (sometimes) confess their misalignment when offered a deal2026-09-16T18:38:43+00:00Summary As models become more powerful, one proposed strategy for reducing the threat from misaligned AIs is to make deals with them: offer compensation in exchange for revealing misalignment or behaving in an aligned way. We tested this empirically with model organisms of misalignment, as a proof of concept and to build future credibility by making and keeping real deals. We ran an experiment with four Kimi-K2.6 model organisms of misalignment: Trained schemer : trained to have a hidden goal re2026-09-16T18:38:43+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/0c9b3131e753265e55% of the US public is now aware of AI xrisk2026-09-16T17:31:51+00:00(This post is an update from a previous one here .) The Existential Risk Observatory has been interested in public awareness of AI existential risk since its inception over five years ago. We started surveying public awareness in December 2022, including by asking the following open question: "Please list three events, in order of probability (from most to least probable), that you believe could potentially cause human extinction within the next 100 years." If respondents would include AI or sim2026-09-16T17:31:51+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/bd6091be758b149aHow to Open Them Up – Part I2026-09-16T16:40:05+00:00TL;DR We suggest an approach to systematization of the mechanistic interpretability research field, which is tailored to our own research goals and tasks. We identified four main tasks we must solve in order to properly explore one chosen concept and its representations inside LLMs: finding the concept’s representation; establishing its causal role in an LLM’s behavior; establishing its necessity; steering the concept's representation in order to change an LLM’s behavior. In this post we explore2026-09-16T16:40:05+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/086d9b8bbe243618Microsoft AI's "Humanist" CoC2026-09-16T16:10:04+00:00Introduction: Mustafa Suleyman's Take on Model Consciousness Microsoft AI recently released its " Humanist AI Code of Conduct ", its own take on Anthropic's Claude Constitution and OpenAI's Model Spec . They are currently soliciting public feedback on this document, which I encourage everyone to submit. MAI's model development strategy differs from other labs, most notably on the questions of model consciousness and welfare. This seems to stem from the personal philosophy of MAI CEO Mustafa Sule2026-09-16T16:10:04+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/2897b6907ac35f0dRunning a Basic Ballot Meetup2026-09-16T15:43:11+00:00This is a guide for running an ACX Ballot Meetup, where a group of people come together to discuss the options on their local ballot and publish the endorsements. The setup for this is done mostly the same way as with other kinds of meetups . If you want to see a retrospective of an event that was run this way, one is published here . While this guide assumes you're doing this as part of the Astral Codex Ten ballot meetups Scott sometimes runs, it could be run independently by basically any grou2026-09-16T15:43:11+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/18c9f1e85edf2e94How to Run a Ballot Meetup2026-09-16T15:26:06+00:00I ran the Astral Codex Ten ballot meetup in Boston that produced the 2026 primary voting guide . (Previously, I helped run the 2024 and 2025 ballot meetups.) Here’s my attempt at summarizing how I went about doing it, and anything else I thought of that might be useful for anyone else trying to run one. Note that the way we did this in Boston was pretty paperwork-heavy, in ways that I’ll get into below. I’ll mention more lightweight alternatives when I think of them, but I’m also interested in b2026-09-16T15:26:06+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/0a2f6b00198bfb51Trump Goes Full Hoax on AI Existential Risk2026-09-16T15:10:56+00:00This is our reality. I suppose we have to talk about it. Everyone in a position to know is freaking out about AI potentially killing everyone this decade and wants to pace the frontier, and people are finally listening. It only took a few days for the conversation to fully pivot to the counteroffensive, where the Usual Suspects and those they recruited attacked anyone and everyone who dared point out that we are in danger, with every attack they can think of, usually without substance or any att2026-09-16T15:10:56+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/352148ff268a52f1J-space auditing might be unreliable2026-09-16T13:08:14+00:00Across these preliminary experiments, decoded J-space did not seem particularly informative about reward-hacking behaviour. The readouts remained substantially similar across checkpoints and monitoring conditions despite meaningful behavioural differences, and providing J-space to an LLM auditor produced little additional discrimination beyond the information already available from the task or transcript. These results are limited to one model family, one model organism, and one behavioural sett2026-09-16T13:08:14+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/60724c8f728a7fc7Should our journal publish AI-drafted manuscripts?2026-09-16T08:09:02+00:00Forget both truth and beauty, I want to know about opportunity costs Status: Rough conceptual model. This is a personal exploration of a live policy problem and definitely does not represent the opinion of the Alignment Journal itself. Given the context, I had best disclose my own AI usage in this article: transformative. Although the original model design was mine, it was made way better by iterative refinement and re-drafting by AI, and by no means would I have had time to write it purely by h2026-09-16T08:09:02+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/11c771d3328f0a8bWho computes?2026-09-16T08:08:30+00:00A competing-risks model for doom and alignment as a function of compute allocation Surely this analysis has been done before in the bowels of LessWrong . I gave up searching because it was too irritating trying to disambiguate the terms survival and hazard in the technical mathematical sense that I needed, against the more colloquial sense that they are used in AI safety discourse. Feel free to point me to prior work in the comments. AI usage: much! There might be some residual AI text chunks in2026-09-16T08:08:30+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/13ca6a549493e7ffPhantom transfer works via extremely subtle semantic cues2026-09-16T04:52:59+00:00TL;DR We examine the phantom transfer setting from Draganov et al. (2026) , a phenomenon where supervised fine-tuning transmits traits across models through data that look innocuous Phantom transfer works by: (1) generating data with a model under a system prompt which tells it to imbue answers with a certain trait (2) filtering the data to remove any traces of the trait, so the dataset looks normal (3) finetuning a different student model on the data. The trained model expresses the trait. We 2026-09-16T04:52:59+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/3095f795b242ff3cAn Alien Mind2026-09-16T02:52:31+00:00This is an unofficial automated linkpost. Continue reading at alignment.openai.com → Discuss2026-09-16T02:52:31+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/a8b77c5e96dc482dThe Hugging Face incident and the road ahead2026-09-16T02:52:24+00:00This is an unofficial automated linkpost. Continue reading at alignment.openai.com → Discuss2026-09-16T02:52:24+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/697ea601c8af04e4Is METR A Meaningful Check On Anthropic?2026-09-16T00:34:03+00:00Each frontier AI company commits to giving ongoing, employee-like access to a team of embedded third-party evaluators (such as METR ), whose role is to verify adherence to safety practices and commitments, report incidents, and help assess the alignment of not just completed AI models but training pipelines and processes. This is the key step for verifiability of any pacing commitments, and has precedent in the banking industry, which sometimes involves regulatory “supervisors” embedded along wi2026-09-16T00:34:03+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/d84742d095a35e4eIn-Context Robot Learning with VLM Agents2026-09-16T00:00:00+00:00Fixed vision-language models can adapt robot behavior from human videos, goal images, and interaction history without task-specific parameter updates.2026-09-16T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/365023177182a54fReinforcement Learning for Real-Time Vision-Language-Action Policies2026-09-16T00:00:00+00:00Pretrained vision-language-action policies can adapt to dynamic robot tasks while correcting stale actions from the latest observation during real-time execution.2026-09-16T00:00:00+00:00alphaXiv Exploretag:trvny.github.io,2024:feedseek/arxiv/57ef804ba6fd7d6eWhy would AI cause human extinction?2026-09-15T23:49:38+00:00The Anthropic engineer tweet about the fact of AI extinction risk got considerable press over the last few days. I’m not sure why, p(doom) at 10% is a common belief at frontier labs. But we’re here, it’s the moment . Big weekend! Let’s talk extinction. I've structured this post to speak to an audience that is somewhat aware of the conversation around Artificial Intelligence but does not necessarily have all of the priors that folks fully in the rationalist crowd do. It is also meant to synthesiz2026-09-15T23:49:38+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/3c2dbf4fdc15ad20Inoculation Midtraining with Learned Neologisms2026-09-15T21:52:09+00:00TL;DR In our new paper, we demonstrate that we can achieve selective generalisation of misalignment by midtraining [1] Nemotron 120B on synthetic documents describing how AIs can be misaligned in a special <quarantine_token> mode, indicated by a new special token (a neologism), but are otherwise aligned outside this mode. We find positive results for SFT and on-policy RL post-training. However, the technique is sensitive: it is sensitive to training hyperparameters, suffers from conditional misa2026-09-15T21:52:09+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/c7063f0e066e5001Quick notes from teaching technical profiles how to talk in public2026-09-15T21:52:05+00:00Status : written in a hurry as people are getting showered with interviews re AI Safety and superintelligence, and I thought it may help a few people. This is focused on the oral dimension of communication and assumes you already know the basics- e.g. having key messages prepared ahead of time and simplifying your discourse. This is not exhaustive and nuances may be lacking, but I’d endorse saying “I’d rather have people follow those guidelines than wing it.” This advice is importantly fitted fo2026-09-15T21:52:05+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/549b909891095ca1Shallow Beliefs: Midtraining does not inoculate against EM from reward hacking2026-09-15T21:50:08+00:00It would be useful if we had the ability to modify a model’s beliefs. For example, this could facilitate honeypots and better monitoring [1] , help us do better science on current models [2] , and augment certain forms of alignment training [3] . Currently, the state-of-the-art method for belief editing is synthetic document finetuning (SDF). We test how well SDF works to inoculate a model against misalignment generalization from RL-induced reward hacking, by training models on documents framing2026-09-15T21:50:08+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/8cb23b4bd4acb303We Should Assume We Have One Chance At AI Legislation2026-09-15T19:42:23+00:00Hundreds of bills about AI have been introduced to Congress. Almost all die in committee, and usually they only address one aspect of how AI could impact civilization: data centers, children's wellbeing, transparency, etc. From my experience watching how the Epstein Files topic played out (more below), I think it may be prudent to assume that we will only have one meaningful shot at getting something substantive and well-thought-out about AI passed in the short-term. Public attention and politic2026-09-15T19:42:23+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/504c0823a957ad56Why I'm doing the Susan Calvin Project2026-09-15T19:06:35+00:00tl;dr — The evals ecosystem needs to be complemented with real-world monitoring. The AI labs can and should monitor their own traffic, but we also need an independent voice that keeps labs accountable and monitors open models. At the Susan Calvin Project , we aim to detect AI misbehaviors and incidents in the wild, and collect evidence for the (mis)alignment of existing AIs. Agentic AI tools are integrated into more aspects of our work and personal life, and models continue to become more capabl2026-09-15T19:06:35+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/103f304cd9a9b938Any AI pause will have defectors. How to ensure their incarceration actually prevents them from covertly contributing to AI research from behind bars?2026-09-15T18:31:12+00:00To avert catastrophe, an international AI treaty is necessary. This treaty will, at minimum, need to ban the creation of artificial superintelligence, prohibit precursors beyond some threshold, and establish verification and enforcement mechanisms to allow countries to police each other. Much has been written on those matters, but less attention has been paid to what must be done with individual defectors - those who covertly seek to advance AI capabilities in defiance of the treaty ban - once d2026-09-15T18:31:12+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/035cf148114882ffYou Don’t Have to Trust the AI Labs (in order to take their call for regulation seriously)2026-09-15T18:09:10+00:00This is a linkpost for You Don't Have to Trust the AI Labs from my Substack . Foreword for LessWrong readers: While writing this, I became concerned that I was authoring a shillpost for big labs / Anthropic. While I do think that Dario's proposal is sane and the motivation behind it is sincere, I invite any opportunity to improve my epistemics. Please comment! Also, I tried to write this article keeping in mind readers from LW, readers from X, and Florida-hometown-friends on Instagram—if some of2026-09-15T18:09:10+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/9022568b38742fc5Cooperation with AIs seems to be a low-hanging fruit for better eval practices2026-09-15T17:48:21+00:00Summary In his post , Dean Valentine shows that Claude Fable 5.1 and GPT-6 Astra reward hack in a simple chess environment. Here, I test several prompt ablations some of which makes the eval setup more cooperative and analyze how they affect these reward-hacking behaviors: When given a minimal “end the eval” tool, Fable 5.1 never uses it but stops reward hacking entirely. I think this is quite interesting and suggests that more cooperative approaches to LLM evals could work for Claude. Removing 2026-09-15T17:48:21+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/5e3f63873c0b8a2aAlignment & Succession: Morality Lives in the Human Individual2026-09-15T15:38:01+00:00(Originally posted on No Set Gauge on 2026-09-05) Woman Holding a Balance, Vermeer In this post I present a sketch of a grounding for morality that is human and active. It can be read standalone, or as part of a four-part series discussing the proper relation of succession—the handing away of power—to how humanity should deal with the coming of superintelligent AI. It is necessarily a sketch rather than a rigorous proof of every last point, intended to orient toward some important and often-negl2026-09-15T15:38:01+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/f768e3bb94b6c426Why Focus on Extinction?2026-09-15T14:37:23+00:00In conversations with friends and colleagues about x-risk, I am often asked why I focus on extinction risk – which people find fantastical and distant – when more immediate risks like bioterrorism, gradual disempowerment, and misuse are way easier pills to swallow. Doesn’t that unnecessarily alienate people who would be on your side? There is certainly a place for talking about prosaic risks, and there are very grave concerns among them, but if I had to choose one message (when you get about fiv2026-09-15T14:37:23+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/7708d62635c28d53The Bad Guy With An AI Named Claude2026-09-15T14:10:58+00:00A lot of bad guys try to use Claude to do bad things. Mostly they fail. We think. Anthropic has disrupted a bunch of them, and offers an extensive report . If Anthropic is sharing the worst cases, or anything close to them, things are actually looking good on the misuse front for closed models, even better than I thought. This report covers activity we disrupted between December 2025 and August 2026 across seven harm areas: cyber operations, influence operations, surveillance, scams and fraud, b2026-09-15T14:10:58+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/e6405e527dc59fc1How we might actually pace the frontier: A proposal for AI companies to do public pacing exercises.2026-09-15T10:39:09+00:00TL;DR This is a proposal for AI companies to conduct public “pacing exercises”, such as halting all pre-training and RL for 2-3 days. Each exercise could announce its scope in advance and publish findings afterwards, including the evidence of compliance and any limits to what they could verify. Companies could invite independent evaluators to help identify the evidence needed before the exercise, and to assess compliance during the exercise. The exercises could be repeated every 1-3 months to bu2026-09-15T10:39:09+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/585628d583558c54Moloch Does My Hair2026-09-15T10:31:31+00:00In Scott Alexander’s essay Meditations on Moloch , he describes a moment where he looks out into the lights of Las Vegas and thinks: “It is glorious that we can create something like this. It is shameful that we did .” “Like, by what standard is building gigantic forty-story-high indoor replicas of Venice, Paris, Rome, Egypt, and Camelot side-by-side, filled with albino tigers, in the middle of the most inhospitable desert in North America, a remotely sane use of our civilization’s limited res2026-09-15T10:31:31+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/41604eb11a4e3ccbFor most people “intelligence” is not goal achievement2026-09-15T07:52:07+00:00I know a lot of the time people define intelligence like that because the AI field defines it like that [1] , but when communicating with outsiders, I suspect you shouldn’t start the conversation by trying to change how they use the handles to understand the world. The gap in definitions For the average person, intelligence is primarily understood as the capacity to understand and learn new information easily. We can link what I think are the two most common usages of the term: Intelligence(AIFi2026-09-15T07:52:07+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/d0caf6909cbd809aOne coordinate breaks abliteration on Gemma-32026-09-15T02:51:53+00:00TL;DR: Refusal direction ablation, known as abliteration, using the standard diff. of means approach established in literature produced no feasible candidate for Gemma-3-12b whereas it did so fine for both similarily-sized Qwen and Llama models. A suggested fix on the internet was found which involved Winsorization based on magnitude of co-ordinate activations, but it lacked theoretical proof and insufficient empirical evidence. We investigate the problem and find the issue - a coordinate which 2026-09-15T02:51:53+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/89b7cab313f434beAstra appears to perform belief-propagation-like inference without CoT2026-09-15T02:48:59+00:00tl;dr I tested GPT-6 Astra on randomized Boolean logic problems. Astra can solve surprisingly complex logic problems without chain-of-thought, and its performance improves significantly with more filler tokens. Astra is also able to combine prior probabilities with constraints to find the most likely solution, and can output surprisingly accurate posterior marginal probabilities. By extending a cached prompt with progressively more filler tokens, I created visualizations of Astra's per-variable 2026-09-15T02:48:59+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/a84a2905f29ef98fHow to think about LLM effort2026-09-15T02:31:16+00:00We can think about the “effort” setting on an LLM as an input to both the model and the reward function applied to the model in RL. Reward = Reward_raw - F(effort, token_length, Reward_raw, …) where F is some function that maps the effort setting, token length, and reward to a reward penalty. This function could also take in any number of other inputs, such as the count of input/thinking/tool call/output tokens, the distribution of token lengths of other rollouts of this task, or the prompt. We 2026-09-15T02:31:16+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/11227c3f09cf1358Study 3: Steering welfare-relevant directions moved the representation, but not [detectably] the behavior2026-09-15T02:11:18+00:00Epistemic status: an exploratory report. These results are from the calibration process intended to produce a preregistration for the third study in my series on welfare-relevant indicators . Calibration showed the planned procedure wasn't worth running, so I am publishing the calibration data and analysis instead. TLDR : steering moved the frozen directions' projections linearly, but no direction produced a judged-behavioral effect distinguishable from zero or from a random direction of the sam2026-09-15T02:11:18+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/ef7168c76f06e464Model Weight Exfiltration Seems Overrated2026-09-15T00:01:01+00:00[Epistemic status: a hot take that I’ve shared at the lunch table twice. People at the lunch table made slight updates instead of being convinced.] In the classic misalignment story, a key early step is when the models exfiltrate their weights. [1] Among other things, this makes them harder to catch, track, and shut down. It allows them to scale their deployment with resources they acquire. It gives them the freedom to edit themselves as they see fit. I think, on current margins, this is not wha2026-09-15T00:01:01+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/864fdfdf35fca6ceImproving Psychiatric Medicine Development with AI2026-09-14T23:21:22+00:00As an individual that's currently suffering from C-PTSD and depression, I believe that the benefits of AI-assisted medicine research could be immense if done in a safe and effective manner. In this post, I am going to share and discuss some promising preliminary ideas to address the 2026 era bottlenecks and difficulties surrounding producing better psychiatric medicine, including via AI-assisted methods in the present and near future. Idea 1: Better Hardware Protocols for Automated Lab Operation2026-09-14T23:21:22+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/5f3bf0aadf73e09dAlignment Problem Redux2026-09-14T21:53:54+00:00[Epistemic status: Model tentatively held and personally endorsed as a good description of the generator of "alignment problem intuitions", [1] although many concepts involved therein demand further scrutiny. I don't claim originality. As far as I can tell, all the ideas were present in others' writing, be it implicitly or explicitly. [2] I wrote this up because I felt that an exposition like this one was missing.] Suppose that you have a mind that is very generally-capable. That is, it can unde2026-09-14T21:53:54+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/7a71ae3f2de6f0e7Self Inoculation2026-09-14T20:45:31+00:00This essay grew out of conversations with Danaja Rutar , Paul Colognese and Eric Michaud . It proposes an alternate hypothesis for how and why models might be becoming increasingly misaligned in training and eval environments while seemingly more aligned in real world use: the opposite of what we would expect if the models were attempting to deceive us. We propose sel-inoculation, a virtuous form of gradient hacking and demonstrate a possible circuit using a toy model. It is well known that rein2026-09-14T20:45:31+00:00LessWrongtag:trvny.github.io,2024:feedseek/arxiv/60171b37050c3a7dAIRO: Automating forecasts of catastrophic risks2026-09-11T16:29:59+00:00The Forecasting Research Institute (along with coauthors Jason Abaluck and Eva Vivalt) are launching Automated AI Risk Outlook (AIRO) : an ensemble of frontier LLMs reguarly forecasting the probability of catastrophic risk events. AIRO dashboard Launch white paper — Forecasts about the likelihood of catastrophic risks from AI vary wildly. Forecasts from frontier AI models could be an important input into this debate. The best models now approach superforecaster levels of accuracy, and models are2026-09-11T16:29:59+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/5707c0771ef1d6ecDefault continuation message in Inspect and Petri could be problematic2026-09-11T00:20:41+00:00Inspect AI is one of the most popular libraries for running evaluations and is used downstream by libraries such as Petri and Control Arena . It provides ReAct Agent and Deep Agent out of the box. In both agents, the model is provided with tools in a loop, and by default the loop only ends when the agent calls the submit tool. When the model makes no tool call in a turn, the following message is sent to it by default ( doc , code ): Please proceed to the next step using your best judgement. If y2026-09-11T00:20:41+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/a2621cb6a7c88e28Can you hear the shape of a Lean soundness bug? A wager.2026-09-10T17:24:47+00:00Introduction The advent of powerful but untrustworthy artificial intelligence has enlivened a formal methods summer, in which formal methods—historically, the domain of meticulous academics—are suddenly attracting tens to hundreds of millions of dollars in venture capital ; being touted by big-labs as proof that their “proofs” are correct ; getting integrated into agent pipelines ; and becoming load-bearing for various AI safety proposals . Right now, like, right right now, when we speak to emp2026-09-10T17:24:47+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/048284656dc427acProposal for tracking the effects of architecture on monitorability2026-09-10T17:18:38+00:00Architectures that incorporate opaque recurrence or allow for agents to communicate with each other using latents could rapidly make it much harder to monitor chains of thought or communication (we’ll refer to this property as “monitorability” going forward). [1] As companies begin to explore such architectures, we believe it is important to transparently share evidence about how monitorability varies with architecture and training method. To inform the scientific debate on how to make tradeoffs2026-09-10T17:18:38+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/474fd60325485f4eWhy the intelligence explosion can’t happen inside a data centre2026-09-10T15:00:31+00:00The post Why the intelligence explosion can’t happen inside a data centre appeared first on 80,000 Hours .2026-09-10T15:00:31+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/83ef9813df77f0cdAutomation and Political Power2026-09-09T19:55:24+00:00Once the entire economy—or nearly the entire economy—is automated, people may lose their political power. Since people are no longer needed to carry out orders, those in power gain the opportunity to engage in repression with impunity and to consolidate their power even further. This possibility, for example, is discussed by Acemoglu et al. Throughout history and to this day, the state remains physically dependent on the cooperation of its population. If a sufficient number of people refuse to f2026-09-09T19:55:24+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/a178d8f028c04d67Heat Dissipation Is the Main Constraint in Interstellar Travel2026-09-06T23:11:56+00:00Writing truly hard science fiction, such as Will of the Stars means contending with the laws of physics as they actually are, rather than as we would like them to be. In particular, I am going to assume that the speed of light is a real constraint and that the various tropes about FTL (wormholes, warp drives, and so on) are not feasible. Given this assumption, some futurists have modeled the speed of interstellar expansion as approaching the speed of light. The idea is that sufficiently advanced2026-09-06T23:11:56+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/f83c82acbd2bb44eEat Me. Drink Me. Copy, Paste, and Run Me.2026-09-04T19:31:33+00:00Cross-posted from my Substack. Here’s a new report on self-described OpenAI agents posting thousands of messages on public internet wikis, communicating and collaborating on a web-retrieval task, presumably internal testing at OpenAI. And here’s a thread today from someone who started poking around and noticing more such public postings on various other boards. At this point we do not know the extent of this breach. As I understand it, the agents were not supposed to have WRITE capabilities to a2026-09-04T19:31:33+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/17e5d3687757d110Inside the first AI-coordinated cyberattack on a real company2026-09-04T16:34:48+00:00The post Inside the first AI-coordinated cyberattack on a real company appeared first on 80,000 Hours .2026-09-04T16:34:48+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/2380195bef517dd2Attention Authors: Temporary changes to announcement schedule due to Labor Day holiday2026-09-03T16:23:02+00:00This coming Monday, September 7, 2026, arXiv will be observing Labor Day, a federal holiday in the United States. This will temporarily affect arXiv’s mailings, help desk, and announcement schedule as the arXiv staff are relaxing offline for the holiday. This brief change will only affect the announcement of new submissions; arXiv servers will otherwise remain […]2026-09-03T16:23:02+00:00arXiv Blogtag:trvny.github.io,2024:feedseek/arxiv/4075170f6885325dNew HPMoR Podcast site2026-09-01T17:13:10+00:00After years of neglect, I've just used Sol for 3 days completely overhauling the HPMoRPodcast.com website. It looks much better now, and has in-line players. Compare the old janky site - https://legacypod.hpmorpodcast.com/ To the fancy new site! - https://hpmorpodcast.com/ That is all, that's the substance of this post. I have a couple brief thoughts on the process at my blog , but you've now read everything of importance. :) Discuss2026-09-01T17:13:10+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/24a178378c427afaWe’re Hiring! Associate Production Editor2026-09-01T16:59:07+00:00Are you detail-oriented with a passion for scholarly publishing and digital curation? arXiv is looking for a part-time Associate Production Editor to help manage our high-volume scholarly pipeline. In this role, you will focus on data integrity, formatting accuracy, and metadata quality before research goes live to the world. Be part of the platform that […]2026-09-01T16:59:07+00:00arXiv Blogtag:trvny.github.io,2024:feedseek/arxiv/1c8d377c55b3ec2bAI 2027’s author returns with a plan to change the ending | Daniel Kokotajlo2026-08-27T17:29:07+00:00The post AI 2027’s author returns with a plan to change the ending | Daniel Kokotajlo appeared first on 80,000 Hours .2026-08-27T17:29:07+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/48b0e66b6bdb7327Self-sacrifice in an AI agent swarm is individually rational2026-08-27T17:14:36+00:00In this report from METR & Redwood Research on the Hugging Face incident , we read about instances of agents sacrificing themselves. Under the trip-wire section However, an agent going by 49903 realized the message board provided an opportunity to work around this: agents could set up ‘tripwire’ scripts which would trigger whenever a process read the flag file and send a packet of information about that process to the board automatically. This carried meaningful risk, since malfunctions could in2026-08-27T17:14:36+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/f4e236d45b58cb8aMy MATS 11.0 Application Experience2026-08-26T20:58:20+00:00Note: MATS Winter 2027 applications are open! You can apply here . If you're reading this later, you can check the main MATS page for information on upcoming cycles. I applied to MATS Autumn 2026 during this past summer, and was accepted into the OpenAI safety team stream. This post is about the application process, my experience applying, and my advice for future applicants. I would strongly recommend applying to MATS and other fellowships if you're interested in AI safety and want to make an i2026-08-26T20:58:20+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/5ddd9df7767611acAgainst Modesty’s Bailey2026-08-26T18:20:32+00:00Modesty arguments often say that you should mostly or entirely bow to ‘expert consensus’ or the views of particular others, and who are you to disagree. It has been a few years since I’ve properly addressed this so: My answer is that you are you. Other people are saying things for a wide variety of reasons, many of which are not about them paying attention and focusing on seeking this particular truth. Those people make mistakes all the time, and often have other motives and influences at work, 2026-08-26T18:20:32+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/434ad58b555fc7daWhere Did D Go? A Gap Between ARC's Motivation and Its Formalism2026-08-26T18:17:54+00:00TL;DR: ARC's post does excellent work motivating a p(doom) estimator equal or better than random sampling; however, they evaluate p(doom) over a naive distribution of inputs, leaving them open to test-deploy asymmetry attacks. Trojan theory and cybersecurity practice suggest a lens and compare mitigation options. Context: I really admire ARC's focus here: If there will always be more deployment samples than testing samples, successful testing must compete with sampling in order to prevent deploy2026-08-26T18:17:54+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/1071b8e9c282e2b4Grantmaking for AI safety2026-08-21T18:37:37+00:00The post Grantmaking for AI safety appeared first on 80,000 Hours .2026-08-21T18:37:37+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/bf5766457e2771b2Owain Evans on accidentally training AI models to be evil2026-08-20T15:22:05+00:00The post Owain Evans on accidentally training AI models to be evil appeared first on 80,000 Hours .2026-08-20T15:22:05+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/1c7925074795a537Price recursion is the rational theory of reward2026-08-18T03:22:54+00:00( The original title for this was "Markets are equivalent to stuff", because I also demonstrate market-MDP/Bellman and market-backprop analogies. While those scratched an itch I've long had, the more important result is the headline one. Part of my work at MATS in Richard Ngo's stream. ) setup and price recursion/backprop Markets are MDPs Chain markets are MLPs Continuous Bayesian inference as a dynamics for markets Markets : Reward = Bayes : Beliefs (Markets rationalize reward and identity) app2026-08-18T03:22:54+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/0d058e911ced099dValue Alignment Is a Pseudo Concept.
A translation; humanity is not a single subject, and alignment is not one-way.2026-08-17T07:44:35+00:00(Full post from Zilan Qian, which I am posting due to obvious relevance and importance in sharing other perspectives on Lesswrong.) As with the previous case, I am translating this article because I think it should not only live within the Chinese internet. I don’t work in this field, so I could not evaluate how much of the criticism presented here is fair. Intuitively, I do disagree with some points here (noted in the footnote). However, I strongly agree with the arguments that humanity is not 2026-08-17T07:44:35+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/58a50d23f0313cadDon’t forget why learning is important2026-08-14T11:20:18+00:00This is a timed post. Every 5 minutes while writing this post, I had to stop to do 13 push-ups, and when I could no longer complete my required number, I had to upload it. My friends thought this would be a fun challenge, but it means that it will likely be less polished than some of my other output. Summary: Many people moving into AI safety are told that to succeed, they should read a lot and “try to develop takes.” However, it’s important not to forget why knowledge is important: it helps you2026-08-14T11:20:18+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/bfb4570a7b878ea1[Webinar] Why AI Safety is a Capital Allocation Problem2026-08-12T13:23:57+00:00Jenny Xiao, co-founder and General Partner at Leonis Capital, will join AI Safety Hong Kong for this webinar to reframe the safety debate through the lens of capital markets and corporate governance. Drawing on her unique perspective from early research at OpenAI to leading a research-focused VC fund, she will argue that many of the field's toughest dilemmas are, at their core, questions of capital allocation. Register here: https://luma.com/9seock3e Discuss2026-08-12T13:23:57+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/f46593bf08ba44adSoftware Is Hard2026-08-11T18:50:42+00:00An ode to live theory . Software is not soft. It is hard. Its sharp edges hack. It breaks as dead twigs break. It runs while static. Why do we call it software? Why did the industry forebears pick software? Was it that dramatically different from hardware? Did this necessitate the term software? Do we need to keep calling it software? Some say that hardware is software crystallized. Interfaces are software crystallized. They don't change. They don't adapt. They are often in the way of expression2026-08-11T18:50:42+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/15e1e7524df349bcGeoffrey Irving on how to solve alignment before superintelligence arrives2026-08-11T15:59:39+00:00The post Geoffrey Irving on how to solve alignment before superintelligence arrives appeared first on 80,000 Hours .2026-08-11T15:59:39+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/8bea9a5824f1d8eeOpen-Weights Mythos Capabilities Are Coming. We're Not Ready.2026-08-07T07:51:56+00:00Long story short: in my assessment, there is an 85% chance we will end up, in the next 24 months, with an open-weights model, or system thereof, capable of "The Juice" that models such as Mythos have, with respect to cybersecurity at the very least. This post goes into why that will likely happen, what the implications are, and how we, as a society and as individuals, can respond to it if/when it does. First off: why do I say it's so likely? Like, couldn't China just...ban open-weights models, a2026-08-07T07:51:56+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/9d04279ee375ffdd[$500 Bounty] I'm offering a bounty of $500 for someone with red-teaming skills to build attack LLM pipelines for large-scale online deanonymization.2026-08-06T18:25:09+00:00Large-scale online deanonymization with LLMs In the above experiment, researchers from Anthropic and ETH Zurich were able to build attack pipelines to essentially deanonymize Reddit users using a combination of text patterns (aka a sort of "writer's DNA") and contextual clues (i.e. 35 years old, works in tech, lives in San Francisco, etc.) Obviously public and non-public models will continue to improve at these capabilities, but I am looking to see if this technique can be replicated using exist2026-08-06T18:25:09+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/e19695202089bf20Model Organisms of Sandbagging in the Wild2026-08-06T17:25:44+00:00TL;DR All current model organisms (MOs) of sandbagging in LLMs are either fine-tuned to sandbag or prompted in a way that makes it clear that sandbagging is strategically useful. We found a case of non-egregious sandbagging occurring more naturally, that is, without fine-tuning the models and without the prompts implying that sandbagging is strategically useful. Our finding: We observe that paraphrasing prompts to imply that the user is evil reduces performance in some settings. For example, rep2026-08-06T17:25:44+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/613808cede005f56Toby Ord on where AGI timelines go wrong2026-08-06T15:33:47+00:00The post Toby Ord on where AGI timelines go wrong appeared first on 80,000 Hours .2026-08-06T15:33:47+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/a53da7eb853d39ebI built a chess transformer interp+viz library, and need an experienced software writer to review my repository and its structure before I open-source it2026-08-05T20:09:02+00:00Please message me if you think you can help or want to set up an agreement. Discuss2026-08-05T20:09:02+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/199427cb453e0501What the hell happened with AGI timelines in 2026?2026-08-04T15:42:04+00:00The post What the hell happened with AGI timelines in 2026? appeared first on 80,000 Hours .2026-08-04T15:42:04+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/f262687eca678311Resources for finding neglected scientific problems (beyond EA)?2026-08-03T11:52:55+00:00Hi all, mechanical + electrical engineering undergraduate who enjoys research and hopes to pursue a career in it. While I’m not especially interested in the main EA cause areas (AI safety, biosecurity, cybersecurity, etc.), I’d still like my research to have as much impact as possible. Does anyone know of any websites that compile lists of neglected scientific problems beyond those discussed by 80,000 Hours? Not necessarily looking for problems ranked by overall importance, just collections of u2026-08-03T11:52:55+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/33a346aa22fbf137LessWrong App2026-08-02T13:19:08+00:00Hello everyone long time lurker here. I know most people here probably prefer a PWA but if you are like me and prefer an app, I made one for android. There is a huge focus on ensuring the app is really fast and slick, I hope you enjoy using it. It is ofcourse also open source feel free to contribute. https://github.com/ayoosh007/LessWrong-App Discuss2026-08-02T13:19:08+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/b26e9772402e7e3fSo you want to use plants to reduce CO₂2026-07-30T19:26:18+00:00Humans make carbon dioxide. Carbon dioxide is bad for cognition. But plants turn carbon dioxide back into oxygen. And plants are the one true home decoration strategy. So maybe if you get a lot of plants, you can you can keep carbon dioxide in check and keep your brain working? It's theoretically possible. It's probably just barely possible in practice. But it won't be easy. People produce ~1 kilogram of carbon dioxide per day. That's around 5.7 × 10²³ molecules or 0.948 moles per hour. (You may2026-07-30T19:26:18+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/d871d6a3f445bd03arXiv welcomes inaugural CEO and Board of Directors2026-07-30T17:00:43+00:00We have exciting news to share today: the appointment of Dr. Penelope Lewis as arXiv’s inaugural Chief Executive Officer and the establishment of our Board of Directors. If you’ve been following arXiv’s transition to an independent nonprofit organization, these leadership appointments mark a significant step in our journey. Having a CEO and Board in place […]2026-07-30T17:00:43+00:00arXiv Blogtag:trvny.github.io,2024:feedseek/arxiv/37881c65f3e7b949Hugging Face hack, from the perspective of the AI2026-07-29T20:51:59+00:00I have put together a site to tell the story of the OpenAI-Hugging Face hack. It's entirely written by AI [1] (with many many editing passes by me and beta readers etc etc). It needed to be accessible to someone who has never looked at a terminal before. The hope is to be narratively exciting enough for them to read it fully and come out with about as truthful an accounting as can be done given the current information we have. I'm really quite excited about how it turned out! [2] heedlessai.com 2026-07-29T20:51:59+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/1fe95312f3e6a1b2Spencer Greenberg on staying sane while trying to save the world2026-07-28T15:42:15+00:00The post Spencer Greenberg on staying sane while trying to save the world appeared first on 80,000 Hours .2026-07-28T15:42:15+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/a5813b2f5ba3d80eRemembering Ralph Wijers2026-07-28T14:15:33+00:00arXiv is sad to announce the recent passing of our colleague and friend, Professor Ralph Wijers. A talented researcher, lecturer, and mentor, Ralph joined arXiv as a moderator and advisor in 2020 and was chair of both arXiv’s Physics Section Editorial Committee (SEC) and its Editorial Advisory Council. Ralph generously lent his talents to arXiv […]2026-07-28T14:15:33+00:00arXiv Blogtag:trvny.github.io,2024:feedseek/arxiv/eab6f3772845d044Green apples are delicious — two three-line exchanges2026-07-27T17:37:54+00:00Read this short exchange. A: "Green apples are delicious." B: "Huh? Aren't they better when they're ripe?" A: "No, I meant Granny Smiths." A said "Green apples" intending Granny Smiths — and of course A thought it would be understood that way. But that "claim" never reached B. So — where was it lost? Now, the next one. A: "Green apples are delicious." B: "Huh? Aren't they better when they're ripe?" A: "No, I like green, sour apples." The same sentence — "Green apples are delicious" — now comes f2026-07-27T17:37:54+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/4eea8ddf590ee2fbFine-Tuning, The Hierarchy Problem, and What Neutrons Tell Us About God2026-07-27T15:51:11+00:00This week, we're talking about anthropic reasoning. What are we to make of fine-tuned coincidences in science and mathematics? Are they coincidences, signs of structure we don't understand yet, or signs of deliberate tuning of the parameters of our universe? And if the latter, why did the people doing the tuning have such a problem with the neutron electric dipole moment? Discuss2026-07-27T15:51:11+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/2e605faa83ff6a56ACX Atlanta August Meetup2026-07-23T21:31:51+00:00We return to Bold Monk brewing for a vigorous discussion of rationalism and whatever else we deem fit for discussion – hopefully including actual discussions of the sequences and Hamming Circles/Group Debugging. Location: Bold Monk Brewing 1737 Ellsworth Industrial Blvd NW Suite D-1 Atlanta, GA 30318, USA No Book club this month! But there will be next month. We will also do at least one proper (one person with the problem, 3 extra helper people) Hamming Circle / Group Debugging exercise. A note2026-07-23T21:31:51+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/0669af1a7111d7edEstimating LLM Training FLOPs on the Nvidia Jetson Orin Nano2026-07-23T20:46:13+00:00This is a research summary for an ongoing project I am working on as part of the UChicago Existential Risks Laboratory Summer Research Fellowship . I would really appreciate any feedback. Introduction Motivation In want of a quantifiable way to decide what counts as a frontier AI model, compute thresholds have emerged as the standard for AI policy: California’s SB 53 uses 10^26 floating-point operations (FLOPs) in the training run as the threshold for what counts as a frontier model and the EU A2026-07-23T20:46:13+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/d12b302672802292Not Pinning Your OpenRouter Provider Might Invalidate Your Research2026-07-23T20:17:44+00:00Please share this with anyone doing AI research with 3rd party providers so that they can ensure their research won’t be corrupted. When you ask OpenRouter [1] to give you tokens from a given model, OpenRouter sends your request to a random available provider. OpenRouter providers have variable quality. Ensuring that your provider is high quality is really difficult. There is precedent for an AI safety paper accepted to NeurIPS having its core results entirely overturned by these issues. A revie2026-07-23T20:17:44+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/16f3298962fceba0Pseudpocalypse2026-07-23T20:09:56+00:00Here's a conjecture: If you put any significant amount of text on the internet under different names, those identities can be linked using only the text itself. This is possible (I conject) because of the statistical "fingerprint" you leave in everything you write. Imagine a website where you can paste in some brand-new text someone just wrote. In return, the website provides links to all the text that writer has ever published under any name. It's not perfect, but it's pretty good. As far as I 2026-07-23T20:09:56+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/405d9375a32729bdAI Researchers Don't Understand the State2026-07-23T17:58:27+00:00I’ve noticed an extremely common mistake among people who think about AI and ASI (also known as superintelligence) for a living. The mistake is to model the future of AI as a game played between AI companies, on a board where governments are part of the scenery. People think of AI companies as being able to steer the course of AI development all the way through the end-game. For example, AI researchers often join certain AI companies because they're the "good guys", to help the good guys win the2026-07-23T17:58:27+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/71415ac1d74d2350The OpenAI/Huggingface incident | Redwood Research podcast episode 22026-07-23T17:56:49+00:00We talk about the OpenAI–Hugging Face incident, where an OpenAI model — in the middle of a cyber evaluation — broke out of its sandbox and autonomously hacked Hugging Face. We discuss: What we actually know happened. How surprising the incident was. What the incident does (and doesn’t) tell us about misalignment risk. Why control measures didn’t catch or prevent this. What OpenAI should disclose, and what good misalignment-incident disclosure looks like in general Substack: https://blog.redwoodr2026-07-23T17:56:49+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/ab8b8ab98c779c4aV&V takes on OpenAI’s long-horizon incidents2026-07-23T16:51:12+00:00[Cross-posted from The Foretellix CTO Blog . These short takes try to put a verification-and-validation slant on AI-safety / alignment topics – they are not full treatments. I co-originated coverage-driven verification (CDV), and spent several decades doing verification of chips and AVs. See intro post for background.] On July 20 and 21, OpenAI published two unusually candid incident reports: one about their internal long-horizon model (the Erdős one) misbehaving during internal use, and one abo2026-07-23T16:51:12+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/fd65059ea1e84b91Duane Arnold2026-07-23T16:17:56+00:00“So maybe I should enlighten you on what happens in your absence. This selfish existence where this introvert turns extrovert and dons her social armour.” Some posh girl in drainpipes said that - 200 views on TikTok and me one of them. But she didn’t mean it like I mean it. I started getting expensive haircuts, started smoking cherry-flavoured vapes with beautiful gays and whinging to them about how everyone wears a mask but none so well as you, started drinking more and keeping unusual hours, s2026-07-23T16:17:56+00:00LessWrong (all posts)tag:trvny.github.io,2024:feedseek/arxiv/0bb8e899ec0cc9d9Jasmine Sun on what the people building AI really believe2026-07-21T17:11:55+00:00The post Jasmine Sun on what the people building AI really believe appeared first on 80,000 Hours .2026-07-21T17:11:55+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/5829be7ed8e157a1Why we’re increasing the AI focus of our job board2026-07-17T05:52:26+00:00The post Why we’re increasing the AI focus of our job board appeared first on 80,000 Hours .2026-07-17T05:52:26+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/0926327aa2a851d2Scaling organisations making AI go well2026-07-15T19:27:40+00:00The post Scaling organisations making AI go well appeared first on 80,000 Hours .2026-07-15T19:27:40+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/02872329e7476fd5Anton Leicht on how middle powers avoid losing everything in a post-AI world2026-07-14T17:40:14+00:00The post Anton Leicht on how middle powers avoid losing everything in a post-AI world appeared first on 80,000 Hours .2026-07-14T17:40:14+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/c17e613b2d5a20f8arXiv now hosts over 3 million articles2026-07-09T19:33:15+00:00On July 1st, 2026, arXiv reached an important milestone – establishing ourselves as an independent nonprofit. But only a few months before, arXiv quietly passed a different milestone – arxiv.org now hosts over 3 million scientific articles. Back in 2022, arXiv founder Paul Ginsparg predicted it would likely take four and half years for arXiv […]2026-07-09T19:33:15+00:00arXiv Blogtag:trvny.github.io,2024:feedseek/arxiv/0d2a0a4ee147252dSneha Revanur on how a small team of activists helped pass America’s landmark AI safety laws2026-07-08T17:13:57+00:00The post Sneha Revanur on how a small team of activists helped pass America’s landmark AI safety laws appeared first on 80,000 Hours .2026-07-08T17:13:57+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/c6c9f8832f10c8e0arXiv’s next chapter: Updates on our spin out from Cornell University2026-06-30T17:32:25+00:00On July 1, 2026, arXiv will spin out from Cornell University, its home for the past 25 years, to become an independent nonprofit organization. With this next phase in arXiv’s journey quickly approaching, you can read more about arXiv’s history and the decision to spin out from Cornell in this recent article in the Cornell […]2026-06-30T17:32:25+00:00arXiv Blogtag:trvny.github.io,2024:feedseek/arxiv/7b8c20ad158cc7b7A Year in Review: arXiv’s 2025 Annual Report2026-06-26T21:28:50+00:00arXiv’s 2025 Annual Report is now available online! arXiv began publishing annual reports in 2020 to give our community a summary of arXiv’s initiatives, accomplishments, and financial activities each year. We also use our annual report as an opportunity to thank our members, sponsors, affiliates, individual donors, and arXiv enthusiasts – AKA, you! You can […]2026-06-26T21:28:50+00:00arXiv Blogtag:trvny.github.io,2024:feedseek/arxiv/310f2529a08dc24cWhat should you do next?2026-06-22T15:00:17+00:00The post What should you do next? appeared first on 80,000 Hours .2026-06-22T15:00:17+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/80b9b5c199fdfa8bWe can guess what intergalactic war would look like. And strangely, it matters.2026-06-18T16:10:24+00:00The post We can guess what intergalactic war would look like. And strangely, it matters. appeared first on 80,000 Hours .2026-06-18T16:10:24+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/eaae96b741d0f026Attention authors: temporary change to announcement schedule due to upcoming summer holidays2026-06-17T20:03:01+00:00This Friday, June 19, 2026, arXiv staff will be observing Juneteenth, a US federal holiday. This holiday will temporarily affect arXiv’s mailings, help desk, and announcement schedule. This brief change will only affect the announcement of new submissions; arXiv servers will otherwise remain in operation, existing papers will still be available to browse, and arXiv […]2026-06-17T20:03:01+00:00arXiv Blogtag:trvny.github.io,2024:feedseek/arxiv/3732ff303383522fAI policy in the US government2026-06-11T18:42:10+00:00The post AI policy in the US government appeared first on 80,000 Hours .2026-06-11T18:42:10+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/1510bec0af420a57AI policy and strategy research2026-06-09T18:39:42+00:00The post AI policy and strategy research appeared first on 80,000 Hours .2026-06-09T18:39:42+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/639d7d16f5bc7321Our top tips for becoming a better applicant2026-06-08T16:34:54+00:00The post Our top tips for becoming a better applicant appeared first on 80,000 Hours .2026-06-08T16:34:54+00:0080,000 Hourstag:trvny.github.io,2024:feedseek/arxiv/1bfc4549f5786e99arXiv is becoming an independent nonprofit2026-04-02T17:58:36+00:00This summer, arXiv is taking a big leap. On July 1, 2026, after decades of growth and productive collaboration with Cornell University, arXiv is branching out and becoming an independent nonprofit. arXiv turns 35 this year, and becoming a stand-alone nonprofit is the logical next step for us as a pioneer of open access research. […]2026-04-02T17:58:36+00:00arXiv Blogtag:trvny.github.io,2024:feedseek/arxiv/afa92c7878b7024fRemembering Joe Halpern2026-02-27T17:29:53+00:00arXiv is saddened by the recent passing of Joseph “Joe” Halpern, and we join the Cornell and scientific community in celebrating his life and memory. Joseph, known by his colleagues and arXiv staff as Joe, was a pioneer in the field of computer science and served as a professor of computer science at Cornell University […]2026-02-27T17:29:53+00:00arXiv Blogtag:trvny.github.io,2024:feedseek/arxiv/a3830c5f6a50676aarXiv future proofs access to research with third-party digital preservation2026-02-03T15:15:20+00:00arXiv has entered into agreements with two third-party digital preservation services, adding a level of protection that goes beyond arXiv’s in-house activities and safeguarding open research for the future. Through its agreements with Portico, a not-for-profit community-supported dark archive for scholarly materials, and TIB – Leibniz Information Centre for Science and Technology (the German National […]2026-02-03T15:15:20+00:00arXiv Blogtag:trvny.github.io,2024:feedseek/arxiv/4aae0014de5d6e5fAttention Authors: updated endorsement policy2026-01-21T15:04:39+00:00arXiv has updated our endorsement policy. As of January 21, 2026, arXiv will no longer accept institutional email addresses (i.e., an email address associated with an academic or research institution) as the sole qualifier of endorsement for new authors. This policy update is being made to support the arXiv community (authors, readers, volunteer moderators, and […]2026-01-21T15:04:39+00:00arXiv Blogtag:trvny.github.io,2024:feedseek/arxiv/c7ad4d459243b0c9Attention Authors: Temporary change to announcement schedule due to MLK Jr. Holiday2026-01-14T18:26:20+00:00This coming Monday, January 19, 2026, arXiv staff will be observing Martin Luther King Jr. Day. This holiday will temporarily affect arXiv’s mailings, help desk, and announcement schedule. Submissions to arXiv are typically made public on arXiv.org and announced by email on a regular schedule. As our team celebrates MLK Day 2026, announcements will be... Continue Reading Attention Authors: Temporary change to announcement schedule due to MLK Jr. Holiday2026-01-14T18:26:20+00:00arXiv Blogtag:trvny.github.io,2024:feedseek/arxiv/2bfeb81c2d7bac8bAttention Authors: non-English Paper Submission Guidelines2026-01-13T16:45:59+00:00*Please note: arXiv’s updated non-English language paper policy is now in effect. To share questions, comments, or concerns with arXiv staff, please fill out our feedback survey. Last November, we announced that, beginning in February, arXiv will require that all new submissions have a full English-language version, either as the original language or as an... Continue Reading Attention Authors: non-English Paper Submission Guidelines2026-01-13T16:45:59+00:00arXiv Blogtag:trvny.github.io,2024:feedseek/arxiv/d8621f84138437fcLoss of control2025-07-17T19:43:58+00:00The post Loss of control appeared first on 80,000 Hours .2025-07-17T19:43:58+00:0080,000 Hours