# AI/ML Questions

**English** · [Español](README.es.md)

Free multiple-choice questions for learning artificial intelligence and machine learning, from the basics to research topics. Every option explains why it is right or wrong, and every question cites its sources.

**Try it: https://ai-ml-questions.strangemachines.workers.dev/**

<picture>
  <source media="(prefers-color-scheme: dark)" srcset="docs/images/question-en-dark.png">
  <img src="docs/images/question-en-light.png" width="600" alt="A medium question on sampling temperature, answered in study mode. The chosen option, 0.73, is marked wrong and 0.98 right; under each option a line explains why, followed by the explanation and the source.">
</picture>

- **103 questions in English and Spanish**, in 8 topics and three levels. Study mode shows the explanation after each answer; exam mode, the results at the end.
- **The answer does not show in the wording.** The questions follow Haladyna's item-writing rules, and a validator checks them: for example, it warns when the correct option is noticeably longer than the others.
- **Measured with real answers.** Answers are recorded anonymously, with no cookies; once a question has enough of them, they show its real difficulty and which distractors nobody falls for ([details](docs/statistics.md)).
- **Open questions.** They are JSON files with a schema, under CC BY-SA 4.0, so you can reuse them in a course, in flashcards or in your own app.
- **No account.** Once loaded it works offline, and it can be installed on a phone.

**Alpha version.** The questions are drafts that a person has not reviewed yet. If you find a mistake, use the "Report a mistake" link under the question or [open an issue](https://github.com/CristobalSantana/ai-ml-questions/issues).

## Topics

| Topic | For example |
|---|---|
| Classical machine learning | Bias and variance, gradient boosting, Gaussian processes, dataset shift, Q-learning |
| Neural networks | Backpropagation, optimizers, LSTMs, mixture of experts, PINNs, KANs |
| Transformers & LLMs | Attention, the context window and KV cache, prompting, LoRA, RLHF |
| Retrieval & RAG | BM25, embeddings, HNSW, chunking, reranking, RAG evaluation |
| Computer vision | Convolution, detection, segmentation, ViT, CLIP |
| Agents, harness & phenomena | Tool use, MCP, prompt injection, double descent, grokking, sycophancy |
| Evaluation & responsible AI | Precision and recall, calibration, LLM-as-judge, fairness, privacy |
| Landmark papers | From the perceptron (1958) to FlashAttention (2022) |

## How the questions are made

Each question states its level in Bloom's taxonomy and gets an estimated difficulty from a rubric. A validator checks the format and the writing rules, and a second model checks facts, sources and the translation before a person approves the question. Once a question has enough answers, they also show whether it tells apart people who know more. The details are in [`docs/`](docs/README.md).

## Contributing

The most useful help right now is checking questions, because a person has not reviewed them all yet:

- **Report a mistake** with the link under each question, or [open an issue](https://github.com/CristobalSantana/ai-ml-questions/issues).
- **Review the questions of an area you know:** the key, the explanations and the sources.
- **Improve the Spanish**, or **propose new questions**.

[CONTRIBUTING.md](CONTRIBUTING.md) explains how. Issues and pull requests can be written in English or Spanish.

## Run it locally

```bash
pip install jsonschema
python scripts/validate.py                    # checks the questions
python scripts/build_site.py --include-drafts # builds site/data
python -m http.server 8000 --directory site   # then open http://localhost:8000
```

## Repository

| Path | Contents |
|---|---|
| `questions/` | The questions, one JSON file per area |
| `taxonomy.json` | Topics, areas and subtopics, in English and Spanish |
| `schema/` | The format of a question |
| `scripts/` | The validator, the site builder and the statistics of each question |
| `prompts/` | Prompts for generating and reviewing questions, in English and in their original Spanish |
| `site/` | The website |
| `worker/`, `migrations/` | The Cloudflare Worker that stores the anonymous answers, and their table |
| `docs/` | The full documentation |

## License

The questions are under [CC BY-SA 4.0](LICENSE-QUESTIONS): you can reuse them if you credit the source and share under the same license. The code is under [MIT](LICENSE), and the fonts under the SIL Open Font License.

If the site helps you, a star on GitHub helps more people find it. You can also support it on [Ko-fi](https://ko-fi.com/strangemachines).