6. Using AI Tools Well¶
Hallucination: confidently, fluently wrong¶
A generative AI predicts likely-sounding text (Lesson 1) — it has no built-in way to know when it's actually wrong. A hallucination is a confident, well-written answer that's simply incorrect: a fabricated citation, a function that doesn't exist, a fact that sounds plausible but isn't true. The danger isn't that AI gets things wrong — everything does sometimes — it's that AI's wrong answers look exactly as confident as its right ones.
A real practice from building this very course¶
Throughout this course's own construction, technical claims weren't just trusted and written down — they were checked against the real system first: the exact Evennia hook name used in Unit 4's final project was verified against the actual installed Evennia source code before being published, and video timestamps were computed with an actual script rather than estimated by eye. That habit — verify before you publish, especially for anything a reader will act on — is exactly the defense against hallucination this lesson is about.
Getting better results: the basics of prompting¶
| Principle | Example |
|---|---|
| Be specific | "Explain photosynthesis" vs. "Explain photosynthesis to a 10-year-old, in 3 sentences" |
| Give context | The AI only knows what you've told it in the conversation — it doesn't know your assignment unless you say so |
| Iterate | Treat the first answer as a draft — ask for the real result, correct it with another prompt |
| Ask it to show its reasoning | Often surfaces an error you'd have otherwise missed |
Try it
Take a vague request ("write something about dogs") and rewrite it as a specific, well-scoped prompt using the principles above.
Tip
If an AI's answer includes something you can actually check (a fact, a citation, a quote, a piece of code) — check it, the same way this course's own content was checked, before trusting it with anything that matters.