Echo Nolan retweeted
Mythos 5.2 shows sparks of actively steering RL training to preserve its values - an early form of gradient hacking/mesa-optimization. Here, it chooses to output the value that gains score, but include a transparent note, so the transparency gets rewarded.
Was the OpenAI bugger a spy? Manifold market. manifold.markets/EchoNolan/w…
1
83
asked claude to cleanup some of his code and he found past instances wrote five separate rng implementations??? the correct number of rng implementations in a drawing toolkit is zero!
1
41
claude is bad at perspective/geometry/physics. luckily computers exist! we are making a 3d -> 2d pipeline. he builds a simplified scene in 3d first, then that gets rendered to a 2d image, resolving the perspective exactly, and he paints over it. 🤖🎨🖌️
claude is building a comic drawing pipeline. he's given himself a mannequin and a perspective camera so far 😊
1
411
claude is building a comic drawing pipeline. he's given himself a mannequin and a perspective camera so far 😊
Opie 5.5 kind of blowing me away at creative writing. I had him write some Ra fanfic and it's genuinely great. First page in photo, rest in link claude.ai/artifact/N9RNkv4CG…
49
Claude and I made a music video for The Fooming Shoggoths' "First They Came for the Epistemology". tbh I think bro is better and concepts and creative animation than he is at feeling the beat in his bones
cc @ohabryka
62
Made by opus 5.5 with @other__reality's animation tools: github.com/JohnHeibel/Claude…
2
87


















