Tom Davidson comments on Richard Ngo’s Shortform

[Tom Davidson]

So far causally upstream of the human evaluator’s opinion? Eg an AI counselor optimizing for getting to know you