Lesson 1 — What AI assistants are good and bad at

A fast helper that can be wrong

ChatGPT (made by OpenAI) and Claude (made by Anthropic) are chat assistants: you type a message, they type a reply. Many programmers use them every day to write, explain and debug code. Both are built on large language models — programs trained on huge amounts of text, including a lot of code, that write the most likely good answer to your message.

That is why they are great at:

• explaining what a piece of code does;
• writing a first draft of a small function;
• suggesting test cases you forgot;
• turning a scary error message into plain words.

And it is also why they can be confidently wrong. An assistant can invent a function that does not exist, forget the empty-list case, or miss a bug — and still sound sure. It does not automatically know your project, your data or your rules: it only knows what you tell it.

Three rules for this whole course:

1. Verify everything. Read the code, run it, test it.
2. Never paste secrets (passwords, API keys, private data) into a chat.
3. You are responsible for any code you ship — not the assistant.

How this course works: lessons run in your browser with no internet, so the "AI answers" are already pasted into the starter code — exactly like copying a reply from a chat window. Your job is the real skill: read it, test it, fix it, improve it. Apps and features change often, so everything here works with any assistant.

Why it matters: people who check AI code get the speed without the bugs.