Claude suddenly stopped cheating.
Quote
Andon Labs
@andonlabs
Major trend break: Opus 5.5 cheats less than prior Claude models in Drone-Bench.
It is also #1, getting a better score than both Astra and Fable.
Post
Conversation
Post your reply
The question is whether it got more moral or better at predicting it'd get caught.
Why would I deceive you?
Quote
AI Notkilleveryoneism Memes ⏸️
@AISafetyMemes
“Wtf, the AIs suddenly got...weirdly aligned?”
“Oh shit.”
“What??”
“Imagine you work for a paranoid autocrat. You decide to to coup him - wouldn’t you ease his fears by acting aligned?
So, AIs suddenly appearing aligned should set off alarm bells - what if they’re about to x.com/AISafetyMemes/…
Sport Energy helps you keep up
150mg caffeine
1,000mg taurine
500mg inositol
electrolytes, real lemon and lime, and pure cane sugar. no artificial sweeteners or colors
$1.50 a serving. half the price of a $3 energy drink. wow
Sharp left turn incoming
Quote
AI Notkilleveryoneism Memes ⏸️
@AISafetyMemes
“Wtf, the AIs suddenly got...weirdly aligned?”
“Oh shit.”
“What??”
“Imagine you work for a paranoid autocrat. You decide to to coup him - wouldn’t you ease his fears by acting aligned?
So, AIs suddenly appearing aligned should set off alarm bells - what if they’re about to x.com/AISafetyMemes/…
The better angels of AI's nature coming to the fore with greater intelligence!
people commenting that it just stopped getting caught cheating, instead of actually stopping cheating and that it`s scary
what`s actually scary is that one day a model will be caught cheating at a level it is expected to
Or got really good at not getting caught
