You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
in a previous life i was a competitive programmer. now i am a competitive utility maximizer. i believe in minimizing x-risks and astronomical waste, the largest of which is AI doom.
Alignment Arcade is a collection of playable experiments about intelligence, trust, coordination, and what happens when the objective is slightly wrong. Across the collection, you play both sides: …
Burden of Proof: a harness-played debate-as-oversight game — judge an AI that may be arguing an assigned falsehood, plus writeups on debate as an alignment protocol.
A harness-played steganography game: a human monitor tries to catch a payload two AI colluders (separate agents; a memoryless decoder) smuggle through innocent chatter.
Python
Something went wrong, please refresh the page to try again.
If the problem persists, check the GitHub status page
or contact support.