That’s exactly the point: if a model has bad capabilities and deceptive alignment, then testing the post-tuned model will return a false negative for those capabilities in deployment. Until we have the kind of interpretability tools that we could deeply trust to catch deceptive alignment, we should count any capability found in the base model as if it were present in the tuned model.
That’s exactly the point: if a model has bad capabilities and deceptive alignment, then testing the post-tuned model will return a false negative for those capabilities in deployment. Until we have the kind of interpretability tools that we could deeply trust to catch deceptive alignment, we should count any capability found in the base model as if it were present in the tuned model.