---
id: ins_hallucinations-when-makes-things-thinks
operator: Dharmesh Shah
operator_role: 'Founder and CTO at HubSpot. Helping millions grow better.'
source_url: https://www.linkedin.com/feed/update/urn:li:activity:7372151908330844160/
source_type: thread
source_title: 'BREAKING NEWS: Seems like OpenAI may have come up with a way to dramatically reduce the hallucina'
source_date: 2026-04-10
captured_date: 2026-05-02
domain: [strategy, leadership, growth]
lifecycle: [strategy-bets, measurement-experimentation]
maturity: frontier
artifact_class: framework
score: { originality: 3, specificity: 3, evidence: 2, transferability: 4, source: 4 }
tier: B
related: []
raw_ref: raw/linkedin/reactions/linkedin-reactions-2026-04-10.md
---

# Hallucinations are when the AI makes up things that it *thinks* are true -- but just ar
## Claim
BREAKING NEWS: Seems like OpenAI may have come up with a way to dramatically reduce the hallucinations in AI models. Hallucinations are when the AI makes up things that it *thinks* are true -- but just aren't. The solution was brilliantly simple. So simple, I'm surprised we didn't come up with it sooner.

## Mechanism
That's why standardized tests (like the SAT) have a penalty for wrong answers. They want to remove the benefit of simply guessing.

## Conditions
Holds when: the operating context matches the post's stated frame (team shape, stage, tooling, buyer type).
Fails when: the practice is lifted into a different stage or buyer context without reworking the underlying mechanism.

## Evidence
> "Hallucinations are when the AI makes up things that it *thinks* are true -- but just aren't."

· Dharmesh Shah, LinkedIn, 2026-04-10

## Signals
- The team observes the pattern repeating across multiple cycles before naming it.
- Practitioners stop questioning the discipline once results compound.
- Skipping the step shows up as friction within one or two iterations.

## Counter-evidence
No opposing view in current corpus.

## Cross-references
- (none in current corpus)