Post #4336496
2026-08-01 04:36 UTC
I had a conversation with some of my associates yesterday (allegedly intelligent folk) about the #anthropic AI #huggingface escape...
... when I said "The model (alegedly) escaped because its testers (humans) were considering false information to be true... and the model escaped to check what the humans EXPECTED to be the correct answer...
I got blank looks.
I expect that is the case with lot of the (allegedly intelligent) folks on here who clearly do not understand that statement either.
TLDR: The student is smarter (in that single domain) than the teacher.
Replies (1)
-
@OAndreas@mastodon.social 2026-08-01 19:32
@n_dimension@infosec.exchange I think that using terms like "escaping" and thinking about it like a person in a cell is hugely misleading - I try to think like "AI is like water" - it finds its way through every opening by simple testing every point of contact