AI safety researcher Miles Brundage argues that assigning a P(doom) of exactly 0 or 1 reveals a lack of imagination

@Miles_Brundage my p(doom) is sqrt(pi)*i

@Miles_Brundage

@Miles_Brundage @speakerjohnash How about negative p(doom)?

@Miles_Brundage and yet those are the only Ps to get engagement on the TL

alas

@Miles_Brundage And all of the rest lack conditional specification!

@Miles_Brundage Reality lives in the messy middle 🫡🙃

@Miles_Brundage I mean at least they're being mathematically honest about their stuck priors

@Miles_Brundage This isn't great, the pdoomers aren't very far off now :)

@Miles_Brundage P dooms need neuroticism handicap adjustments

@Miles_Brundage or maybe im just more realistic than imaginative

either way im not betting on 0 or 1

@Plinz @Miles_Brundage Try the eth root

@Miles_Brundage .8

@Miles_Brundage Or better culture

@Miles_Brundage Enterprise AI risk is usually the eval gate someone waved through, then nobody can name the rollback owner when audit asks who approved the model change.

@Miles_Brundage And also violates Cromwell's rule https://en.wikipedia.org/wiki/Cromwell%27s_rule

@Miles_Brundage p(doom) = e^-2π

@t6aguirre @Miles_Brundage kekw

@sean_from_earth @Miles_Brundage @speakerjohnash Why stop there? P(doom) = i

@sean_from_earth @Miles_Brundage when the poop goes back inside the butt