Andrei: This is a really big game changer one that people haven't quite absorbed fully. Eliezer Yudkowski the sort of famous AI risk skeptic who likes to write unnecessarily long posts about safety could be 10% as long but he flagged this on twitter highlighted and has a great analysis of it so I really recommend checking that out. The bottom line is just by having input output access to this model just by being able to see what goes into the opening eye model and then what comes out the prompts and then the outputs you can train your model to mimic its behavior for really cheap.