Stuart_Armstrong comments on Which counterfactuals should an AI follow?

[Stuart_Armstrong]

I like the subagent approach there.