The reason I thought they were non-epsilon was “it sure seems like people are not willing to go on record as saying AI x-risk is important” (going on record in favor of “AI safety” is probably easy).
Generally, going on record saying something out-of-the-overton window I think counts for nontrivial integrity. (But, this isn’t what I really said in the post, to be fair)
(edit: agreed that quantifying how strong a signal is is important, and not at all sure the strengths I implied here are correct, although I think they are in at least a relative fashion)
The reason I thought they were non-epsilon was “it sure seems like people are not willing to go on record as saying AI x-risk is important” (going on record in favor of “AI safety” is probably easy).
Generally, going on record saying something out-of-the-overton window I think counts for nontrivial integrity. (But, this isn’t what I really said in the post, to be fair)
(edit: agreed that quantifying how strong a signal is is important, and not at all sure the strengths I implied here are correct, although I think they are in at least a relative fashion)