What an LLM Actually Does: Next-Token Prediction by Hand

What does an LLM like ChatGPT, Claude or Gemini actually do? It predicts the next token, and that's the whole job. We work out next-token probabilities by hand with pen and paper, then map the 16-video series that builds a GPT. Video 1 of "LLMs by Hand" , a 16-video series where every formula is worked out with pen and paper In this video: • Why an LLM is a classifier with one class per token • Where a probability like "rates: 80%" comes from, by counting • How context changes every probability (money bank vs river bank) • The roadmap for the whole series Chapters: Intro The next-word game Where the numbers come from Context changes everything The series map Interview corner Recap, challenge and what's next Module 1: Your turn: write a sentence with a blank, give your top 5 next words and probabilities that add up to 1. Post your answer in the comments; the solution is pinned. Next: Video 2, Tokenization by Hand: How LLMs Read Text (BPE) Full playlist, LLMs by Hand: Pen and Paper AI: understand it, derive it, ship it. New video every two weeks. #LLM #MachineLearning #AI Tags: what is an LLM, next token prediction, how ChatGPT works, large language model explained, LLM math, transformer series