No, you can't, in any meaningful sense. Even the biggest bigots, with the most s...

bmacho on April 28, 2023 | parent | context | favorite | on: We're afraid language models aren't modeling ambig...

No, you can't, in any meaningful sense. Even the biggest bigots, with the most stable beliefs, like RMS or the Pope have a lot of contradictory beliefs. (At least, I think so. I don't have any evidence for this.)

Also any strategy that you might came up and is simple enough for you to follow is trivially followable for chatGPT as well.

akiselev on April 28, 2023 [–]

Logical consistency is not the same thing as probabilistic consistency.