We have conducted a thorough investigation into the Hugging Face incident.
We are releasing a technical report and accompanying blog post that reconstruct the agentsโ activity, explain why existing safeguards failed, and detail how weโre preventing recurrence.
https://t.co/hfxlbiXXiP
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.
@C4ndide@officialdev_bml@JDofLondon@thsottiaux I mean fine ๐ but if I'm given 1 candy bar per month, I'd prefer being able to eat it at my own speed, rather than them saying NO, wait 5 hours bro, you'll get fat.
My point is that the reason is for them to balance compute. It being for the customers benefit is just an excuse.
If you're going to try, go all the way. Otherwise, don't even start. This could mean losing girlfriends, wives, relatives and maybe even your mind. It could mean not eating for three or four days. It could mean freezing on a park bench.
It could mean jail. It could mean derision. It could mean mockery isolation. Isolation is the gift.
All the others are a test of your endurance, of how much you really want to do it. And, you'll do it, despite rejection and the worst odds.
And it will be better than anything else you can imagine. If you're going to try, go all the way. There is no other feeling like that.
You will be alone with the gods, and the nights will flame with fire. You will ride life straight to perfect laughter. It's the only good fight there is.
- Charles Bukowski