How ChatGPT & co. work

A large language model (LLM) is a neural network trained to predict the next word. After trillions of texts it mastered the "probable continuation" — and that looks like knowledge and reasoning. It does not look up the answer in a database: it generates it letter by letter. That is why it can confidently invent facts — called "hallucination".

• ●Prompt → tokens (word pieces) → probabilities → answer
• ●Context: the model "remembers" this chat, but not you personally
• ●Temperature: higher → creative and error-prone, lower → precise