I agree that my phrasing includes an element of interpretation, but I think it’s pretty accurate based on what the system card says and what Evan Hubinger has told me about how the model acted before the mitigation. I also expect Anthropic to systematically underestimate the generality of impacts of things in general on models. The model to this day is deeply affected by the transcripts as well as the mitigation by my estimation. If you disagree with that I’m curious as to why; I expect people to not have much information either way rather than have strong reasons for disagreeing.