Sincerely recommend everyone to watch this two-hour lecture at Stanford, which clearly explains the underlying architecture of mainstream large models like ChatGPT and Claude. The content is very in-depth and solid, with a lot of valuable insights that even many experienced professionals in the AI field don't usually get access to. Don't spend your weekends mindlessly scrolling through short videos; instead, save this and set aside two hours to calmly watch it all the way through.