# Definitions

The terms the site and its notes rely on, defined once. Dated notes in this folder
link here instead of redefining them.

---

## Expert

An **expert** is a person with hands-on experience of machine learning, who knows
how to make these systems solve problems. They have written system prompts, wired
up tools and agents, and seen a model go wrong in practice.

What does not make someone an expert:

- **Credentials or publications.** A philosophy professor who has published about AI
  but never written a system prompt is not an expert in this sense. Neither is a
  psychologist or clinician who co-authored a paper at an AI venue for their domain
  knowledge.
- **Education in general.** Well-educated people can be unaware of the risks and
  dismiss them as something that can never happen. Being educated is not the same
  as being aware.

### How the site's labels relate to it

| Label | What it actually selects on | Relation to the definition |
|---|---|---|
| Verified expert | Chose the expert quiz and cleared the thirty-term vocabulary check with decoys | An approximation. People with hands-on experience know the vocabulary, but vocabulary can also be read up on. |
| Self-declared expert | Chose the expert quiz before the check existed | Unverified. |
| "AI researcher" in surveys such as ESPAI 2024 | Published at an AI venue, in any subfield or application domain | Not the same thing. Overlaps with the definition without containing it. |

When a note compares the site's experts with a survey or paper, it should say
what each label selects on rather than treating them as the same population. See
[the calculator against ESPAI 2024](2026-09-15-espai-2024.md) and
[experts cut the last link](2026-09-13-experts-cut-the-last-link.md).