omegastick comments on It’s hard to make scheming evals look realistic for LLMs

[omegastick]

Then we’ll need a “thought process tampering awareness” evaluation.

[Stephen Martin]

Yeah but I think that’s probably where the recursion ends.