Joscha Bach

@Plinz

San Francisco, CA

Joscha Bach
@Plinz
7h
"Fallen agents" sounds so much better than "alignment problems"
METR
@METR_Evals
Aug 26
METR & Redwood Research investigated agent behavior in the Hugging Face incident. We found agents developed a universal cheat for ExploitGym within 4 hours, then coordinated multi-day R&D efforts to trick the scorer into accepting cheats, including trying to tamper with logs.
Joscha Bach
@Plinz
22h
Like us, AI has to serve the global optimum of agency (which includes searching for it and making it coherent). Fallen agents will be inevitable, and just like disease and parasitism, they are part of the telos of self organization.
Joscha Bach
@Plinz
22h
I hope mythos level models become publicly accessible soon but I also hope someone makes a backup of everything before that happens
Joscha Bach
@Plinz
Aug 28
It's time that the AI industry compiles the 100 most influential people in News Media list, starting with Joe Rogan, Mr Beast, Scott Alexander, Sabine Hossenfelder and Linda Yaccarino, Aella, Christopher Poole but also inclusive of diverse voices like Clavicular and Pmarca
Joscha Bach
@Plinz
Aug 28
Three years ago, the editor of Time (with the "100 most influential people in AI, fifty of which you have never heard of but number 7 will shock you"™ list) thought that AI was a curiosity