What AI can and cannot do

In plain English

A large language model is a program trained on huge amounts of text to predict what comes next, and each unit it predicts is called a token, roughly a word or part of a word. The text you send plus the answer it is building is the context, and every model has a fixed context limit, so very long conversations eventually drop the earliest parts. Because it predicts patterns rather than reasoning from facts, it can state false things with total confidence, which is called hallucination. That makes it great at drafting, summarizing and reformatting, but weak at real logic, knowing current facts and citing verified sources. Treat its output as a starting draft, and check anything important against a source you trust.

Step by step

1. 1Read an answer and ask yourself: is this a fact I would need to verify, or just a draft?
2. 2Pick one specific claim from an AI answer and look it up in a trusted source.
3. 3Rewrite a long prompt down to only the context the task actually needs.
4. 4Ask the model a question it cannot know, like a private or future fact, and notice how it guesses.

WHAT AI IS GOOD AT
- Drafting, summarizing and reformatting text
- Suggesting code you will run and test
- Explaining a concept at your level

WHAT AI CANNOT DO RELIABLY
- Reason about facts it never learned
- Know recent events after its training
- Verify its own claims or cite sources
- Guarantee correct code without tests

RULE: if it matters, verify against a trusted source.

Common mistakes

• Assuming a fluent answer is a correct answer is the most common beginner error, because smooth language hides missing facts.
• Treating the model as a search engine for recent or private information is wrong, since it may not know the answer and can invent one.
• Trusting a citation without opening it is risky, because models can produce references that look real but do not exist.
• Thinking the model remembers earlier chats or knows your project automatically is false, as it only sees the context you provide.