My Weird Prompts
What Does It Take To Train An AI Model From Scratch?
28:18
Training a conversational AI from completely random weights requires building an entire statistical education rather than fine-tuning existing layers. Looking closely at open-source pipelines like nanoGPT and nanochat shows how tokenizer choices quietly dictate compute costs across languages, why tiny smoke tests prevent catastrophic compute waste, and how midtraining and supervised fine-tuning…