“OpenAI told a model to do some hacking and then it did some hacking. And?”
I find this aspect so peculiar. Yes, the models were working on an exploit benchmark and the capabilities that it elicits vaguely have something to do with breaking HF? I guess...? But, I think to myself, surely it's obviously not the same to solve what is essentially an exam question on hacking as it is to actually go out and compromise the production systems of multiple real companies. I'm really baffled how this points keep coming up, but I can attest that this is the response I... (read more)
I find this aspect so peculiar. Yes, the models were working on an exploit benchmark and the capabilities that it elicits vaguely have something to do with breaking HF? I guess...? But, I think to myself, surely it's obviously not the same to solve what is essentially an exam question on hacking as it is to actually go out and compromise the production systems of multiple real companies. I'm really baffled how this points keep coming up, but I can attest that this is the response I... (read more)