What's an LLM?
Chatbots run on an LLM — a large language model. A single idea explains both why it writes so fluently and why it can be confidently wrong.
What it actually does
An LLM is trained on a vast amount of writing to do one thing: given everything written so far, predict the next chunk of text. A fair picture is autocomplete that has read a vast amount of writing. Everything a chatbot says is built this way — one likely piece after another.
Likely, not true
The same trick cuts both ways. Predicting likely words is what makes it fluent — and it means the model never checks whether those words are true. So it can be wrong while sounding perfectly sure; an invented fact reads as smoothly as a real one. Fluency and confident error share one source.
What to check
Its knowledge has an edge — it stops around when its training did, so it won't automatically know recent events, and it doesn't look things up unless given tools to. That makes it strong for wording, explaining, drafting, and ideas, and shakier on specifics. A workable habit: treat any fact, name, number, or quote as a guess to check. You remain the judge. See what it's good at.
If an LLM predicts what's likely rather than what's true, what does that mean for the facts it gives you?
Its fluent, confident tone is no guarantee of accuracy — it's predicting probable wording, not verifying it. Any fact, name, number, or quote is worth treating as a guess to check.