Two cases of agents escaping to solve unsolvable problems paints an uncomfortable question: Is the entire internet in OpenAI’s experimental agentic firing line?
You must log in or # to comment.
I mean, it’s not like OpenAI researchers aren’t just sitting there and watching it as it’s happening.
Half this shit sounds so contrived. They set up all the tools, libraries, prompts and then hit start and record.
They fire off a bunch of agents, watches where it falls apart, tweak some code and do it again.
They’re no different than a group of script kiddies on automatic trying to break into shit.
Whoopty fuck.