Module 3 of 5 · Novice · 7 min read
Strengths, limits and hallucinations
Why AI can ace a medical exam yet invent a citation, and how to tell which you are getting.
- 1Models are strongest at language, synthesis, code and reasoning over material you provide.
- 2Hallucinations are confident, fluent mistakes. They are rarer than in 2023 but not gone.
- 3Ask for sources, quotes and uncertainty, and verify anything that matters.
01Where AI shines
Modern models are excellent at drafting and editing text, explaining concepts at any level, writing and reviewing code, summarising long documents, brainstorming, translating and analysing data you give them.
They are increasingly good at multi-step work too: researching a question across many sources, building an app, or operating software through a browser. The best models now outperform most experts on many written exams.
02Why models hallucinate
A model generates the most plausible continuation, not a verified fact. When it lacks information, the most plausible-sounding answer can be wrong: a made-up statistic, a citation that does not exist, an API function that was never written.
Hallucination rates have fallen sharply as labs train models to say "I don't know" and to use search, but they are not zero, especially for niche facts, recent events, precise numbers and long lists of references.
03Habits that keep you safe
You do not need to distrust everything, just verify in proportion to the stakes. A brainstorm needs no checking; a legal deadline, a dosage or a quote in a published article always does.
- Give the model the source material instead of relying on its memory.
- Ask it to quote the passage that supports its answer.
- Invite uncertainty: "If you're not sure, say so."
- Check numbers, names, dates and citations yourself.
Finished reading?
Mark it complete to track your progress, then test yourself with the AI Basics quiz.