
I'm training an LLM from scratch, on my own computer. Each episode of this series builds one real part of a language model, so you can see how ChatGPT, Claude and Gemini actually work under the hood. In episode 1, we discover next-word prediction, the 6 million dials inside my model, and why changing one word in your prompt can change the whole response. Full episode: youtu.be/EH-71hG6jwc




