- What it is: a very well-read assistant that learned the patterns of how people write and talk.
- Its big weakness: it can be wrong and sound completely sure. Check anything that matters.
- The safety rule: what you type is a postcard, not a sealed letter.
How it works, in three lines
- It learned by example. Nobody typed the answers in. 1
- It writes one word at a time, like the word suggestions on your phone, only far bigger. 1
- It is a pattern expert, not a fact database. That explains most of what it is good and bad at.
Why it gets things wrong
Training rewards a confident guess over “I don’t know”, like a test where a blank answer scores zero. A confident wrong answer is called a hallucination. 2
Always check: names, numbers, dates, quotes, and anything medical, legal or financial.
Five habits for good answers
- Say what you want. “A short, friendly post for my dog-walking business”, not “write about dogs”.
- Give the background. Who is it for, and what should it avoid?
- Say what the answer should look like. A list, a table, three options, plain words.
- Improve, don’t restart. “Keep everything, but make it shorter.”
- Ask it to check itself. “What in your answer might be wrong?”
Staying safe
- Never type passwords, bank details, ID numbers or other people’s private details.
- Check the settings for memory and “use my chats for training”. Some apps remember by default. 3
- Agents act, chatbots only advise. Give an agent the smallest set of keys, and make it ask before it pays, sends or deletes.
- Urgent call from “family” asking for money? Hang up and call back on a number you already have. Voices can be copied.
Use AI like a clever friend who is sometimes confidently wrong: great for drafts and ideas, checked before you rely on it.
Related on this site: Printable AI Cheat Sheets
- 1Anthropic, Tracing the thoughts of a large language model, 27 Mar 2025, read at source 1 Oct 2026“Language models like Claude aren’t programmed directly by humans—instead, they’re trained on large amounts of data.” ↑a“Claude writes text one word at a time.” ↑b
- 2Kalai et al., Why Language Models Hallucinate, arXiv:2509.04664, 4 Sep 2025, read at source 1 Oct 2026“the training and evaluation procedures reward guessing over acknowledging uncertainty”. ↑
- 3Anthropic, Claude’s memory works everywhere, and you decide what’s in it, 25 Aug 2026, read at source 16 Sep 2026“Memory is on by default”. ↑
Kalai et al., Why Language Models Hallucinate, arXiv:2509.04664, 4 Sep 2025, read at source 1 Oct 2026.
The comparisons on this page (phone keyboard, test-taker, postcard, keys) are this site’s own ways of explaining, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK