In the speaker's experiment, an AI's AlphaGo-style self-play process plateaued quickly until human play-testing exposed obvious mistakes.

“but uh i think that the and but i set gpt 5.6 to auto research it because like i don't want to handle any of this i expect you know the alpha go process to be like fully in the weights by now it is not it is actually it like immediately leveled off very very immediately until i human play tested it and then i i like called the obvious mistakes and then they were like oh yeah”