Science & TechMental model
A language model only ever guesses the next word.
Large language models like ChatGPT were trained on billions of sentences to predict the most likely next word fragment. Grammar, knowledge and reasoning emerge from that task as by-products. The model doesn't check facts; it continues whatever sounds plausible. That's why it can state falsehoods convincingly.
Useful: Treat its answers like a well-read acquaintance's: helpful, but verify numbers and sources.
Source: en.wikipedia.org