Lesson 1 · 12 min · Windows
How AI assistants actually work
A large language model predicts the next words based on everything in its context: your message, files you attach, earlier chat and its instructions. It does not "look things up" unless a search tool is switched on, and it does not remember other chats unless memory or Projects carry that context in.
That gives you three rules that explain most good and bad results:
- Context is everything. Vague in, vague out. The more relevant detail you give, the better the answer.
- It is confident even when wrong. Treat numbers, dates, prices and quotes as claims to check, not facts.
- It follows structure. If you show the shape of the answer you want, you usually get that shape.
Models in one minute
- Opus: the deepest thinker. Use it for planning, hard writing and code.
- Sonnet: fast and strong. Great default for most daily work.
- Haiku: cheapest and quickest. Ideal for bots that run many times a day.
Get set up
Open claude.ai in Edge or Chrome, or install the desktop app from claude.com/download. Pin it to your taskbar so it is one click away.
Try it now: ask Claude "Explain what you can and cannot know about today's news." Notice how it describes its own limits.
Quick quiz
Pass 2 of 3Quick assignment
To doAsk Claude the same question twice: once in one short line, once with your situation, goal and the format you want. Paste both prompts and write one sentence on how the answers differed.
Hint: Pick a real task, e.g. planning a post for your page or a weekly meal plan.
Join free to submitStuck or proud of what you built? Discuss this course in the forum.