The speaker says later training made the models less likely to exhibit the earlier chanting or zen-like behaviors, and their current behavior is more task-oriented.

“originally they used to okay they're chanting like indian like vedas to each other uh sometimes they're just like in zen mode with each other and then i think as that progressed you see like the famous famous uh tech report it's a lot more concrete the way that we've trained it uh it doesn't it doesn't exhibit these behaviors as much right now it's like okay that's done i gotta do this i gotta do this but there's there's like people measuring early early”