LESSWRONG
LW

If your hypothesis predicts that model performance will be preserved if you swap the input to any other input which has a particular property, but no other inputs in the dataset have that property, causal scrubbing can’t test your hypothesis

Would it be possible to make interventions which we expect not to preserve the model's behaviour, and assert that the behaviour does in fact change?

126Stress Testing Deliberative Alignment for Anti-Scheming Training

2mo

115Ablations for “Frontier Models are Capable of In-context Scheming”

210Frontier Models are Capable of In-context Scheming

93Apollo Research 1-year update

58A starter guide for evals

38Scalable And Transferable Black-Box Jailbreaks For Language Models Via Persona Modulation

46Understanding mesa-optimization using toy models