Guive comments on What predictions from theoretical AI Safety research have been confirmed by empirical work?

[Guive]

What specific confirmatory evidence are you thinking of?

[Larks]

Conversations people have with un-RLHF’d models.