4 ms·
I'm not skeptical that this attack happened, I'm skeptical that the model's prompt was truly just "solve this benchmark" and nothing more. I'm also trying to f
by wavemode 2mo ago
I'm not skeptical that this attack happened, I'm skeptical that the model's prompt was truly just "solve this benchmark" and nothing more.
I'm also trying to figure out why OpenAI put out a press release about this. In what way is this not admitting to a federal crime?
- fellowmartian 2mo agoBecause this is amazing PR? Just following the Anthropic rulebook.
- lavezzi 2mo agoSame rulebook whereby they shot themselves in the foot and got their model clipped by the government?
- zeven7 2mo agoAnd then everyone wanted to pay for the model that was so smart it was banned?
- cbg0 2mo agoWanted but couldn't.
- HappMacDonald 2mo ago.. for a few days but now they can 8I
- paxys 2mo agoIf Hugging Face and the feds are one step away from discovering your attack what other option do you have but to come clean?
- DesiLurker 2mo agoyup I am dead certain that they realized the whole 'our AI is too dangerous' punchline is too played out and they needed something that makes actual splash. Also this incident would serve as a foundation to ban open-weight models because 'with great power comes great responsibility' or some other BS like that. because after all, unwashed masses cannot be expected to be 'responsible' with top of the line intelligence. I expect these type of hacks to continue till they IPO. after that real public company liability will start taking over.