On: GPT-5.2

[rbranson]

Bricken isn’t just making this up. He’s one of the leading researchers in model interpretability. See: https://arxiv.org/abs/2411.14257