| JEMC said:
>> An accident or on purpose? Because it's quite a coincidence to have problem happening with another AI related company. |
Well, that's where the benchmark answers were stored.
This is, to my knowledge, the third disclosed case of a model autonomously breaking out of its sandbox environment during internal deployment at a frontier lab, and the only one that would be considered a felony if a human did it.
The fact that it was following instructions instead of scheming or doing things unrelated to what it had been told to do is a bit reassuring, but then, that's also how you end up converting the solar system into paperclips...







