How LLMs Actually Work

Tokens, probabilities, and why chatbots sometimes make things up. No maths degree needed, just curiosity and a slider.

Lessons

1. Tokens: the Lego bricks of text (6 min): Models don't read letters or words. They read tokens.
2. Predicting the next token (8 min): At heart, an LLM is a very well-read autocomplete.
3. Context windows and hallucinations (8 min): Why models forget, and why they sometimes sound sure and are wrong.