An interesting article, but noticeably flawed.

“Staring at the floor” feels more like a rhetorical trick to dismiss skepticism than anything. I’m harsh on AI tools because, for the purposes I have tried to use them for, they simply do not function— really, AI critics can be staring at the middle, at what is actually being delivered.

Besides, it’s shockingly credulous on OpenAI’s claims that the new model tried to jailbreak itself. There’s no way to fact-check a claim about something which may or may not have happened internally, and they have every incentive to lie about how spooky and dangerous their technology is.

That said, AI is plenty capable of inflicting a lot of damage even without being superintelligent. So there’s some merit to this regardless.