Chat assistants are built on large language models (LLMs). An LLM is trained on huge collections of text so that it becomes very good at one task: predicting what text should come next.
From prediction to conversation
When you type a question, the model breaks it into small pieces called tokens and predicts a response one piece at a time. Further training teaches it to follow instructions, be helpful and avoid harmful content.
What this means in practice
- It sounds confident even when it is wrong. Fluent text is not the same as correct text.
- It works best with context. The more relevant detail you give, the better the prediction.
- Its knowledge has a date. Unless a tool can search the web, it may not know recent events.
- Small changes in wording change the answer. Rephrasing is a normal part of using AI.
Try this
Ask an assistant the same question twice in different words. Compare the answers. Notice what stays the same and what changes.