Available inEnglishFrenchGermanHindiPortugueseRussianSpanish
How LLMs Work: Large Language Models Explained
Understand how large language models work, from tokens and training to attention, sampling, hallucinations, RAG and AI agents. For developers and anyone who uses AI chatbots, in plain words with no maths or code.
What is a large language model (LLM)?
A model trained on vast text to predict the next token
Source
Given some text, an LLM scores every possible next token (a small chunk of text) and one is picked; a chatbot writes its whole reply by repeating that single prediction. The chart shows it for "The capital of France is". This deck is about the GPT-style models behind chatbots; some other language models, such as BERT, are trained to fill in words removed from the middle of a sentence instead.