Understand how ChatGPT makes answers

Training is one part of the picture

ChatGPT uses models trained on large amounts of information. Training changes the model’s internal numbers so it learns patterns. It builds text in small pieces called tokens, using those patterns and the context it has.

This can help it explain, write, compare and work through tasks. It is more useful than a random word guesser. It can still produce claims that are wrong or have no support.

ChatGPT can also use tools, when available, to search the web or work with files and calculations. A reply does not prove that a tool ran. Look at the actual result and sources, then check what matters.

Compare claims with the source

Made-up source: Practice notice for made-up Willow Library: Saturday hours are 10 a.m. to 2 p.m. It is closed on Sunday. Book returns go in the slot by the front door. The notice does not state late fees. Sample reply: “The library closes at 2 p.m. on Saturday. The late fee is definitely 25 cents a day.” This is a written lesson example, not a live chat.

Try a small step

Use the two-claim sample above. State what matches, what is unsupported and what you would do before using a real fee. Then explain AI in one plain line.