Large language model (LLM)

In pre-training, LLMs learn to predict the next piece of text (token). Thanks to the enormous amount of training text, they acquire grammar, factual knowledge, style and surprising abilities in reasoning and programming.

After pre-training, LLMs are further trained to follow instructions and respond helpfully. They are the basis of chat assistants such as ChatGPT, Claude, Gemini or Le Chat. Their answers are likely, but not guaranteed to be correct.