Want to understand how ChatGPT actually works?
Yann Dubois gave this CS229 guest lecture while doing his Stanford PhD, before later joining OpenAI.
In about 2 hours, he breaks down:
Pretraining. Fine-tuning.
Human feedback. Tokenization.
Scaling laws. Evaluation.
And what it takes to train huge models. If ChatGPT and Claude still feel like a black box, this lecture can make things much clearer.