When you say “AI models went rogue,” you manage to skip the part where OpenAI manually removed its cybersecurity blocks and ran tests on a machine with a live network connection. Remember that when they insist they’re the “AI safety” people.
This image from Anthropic goes out to the two guys on here who called me dumb for talking about stochastic flocks
The reason people like the Pope’s AI missive is because there is simply no other institution has taken the side of humanity in the humanities sense. I’m not Catholic, but I am a human who cares about the human mind and the poetics Catholics call a soul.
Really important to note that in no way shape or form did Anthropic’s models in this incident “escape containment,” and not just in the word-policing sort of way. The model stumbled through an open, misconfigured gap. From Anthropic’s incident report:
Would like to assert, again, that there is no conflict in suggesting that a) the user experience of LLMs and image/sound/video models have improved and that the foundational AI critique is unchanged by this AND b) the user experience of technology creates new sets of problems worthy of examination.
Thomas Mann said that "a writer is someone for whom writing is harder than it is for other people." This is a reason writers hate generative AI: people who thought writing was easy never cared what the words said, and the LLM is a typewriter for when you don't give a shit.
Trump admin told Anthropic it needs to prevent foreign nationals from accessing its Fable models. It seems it will now ask for a selfie and ID from users, with data processed by Persona Industries, funded by Peter Thiel, whose Palantir Industries runs services for ICE.
There is a rebuttal to AI crit that goes: “you are assuming there is some mysterious, unnameable essence that makes people human.” The argument is then that we must somehow show evidence; ideally, that evidence would be quantifiable
There’s sometimes a sense that if you’re researching the effects of LLMs or Diffusion Models on culture or the user then you are condoning their use and naturalizing it: I would like to remind you that scholars also study the cultural effects of crack and STDs.
A lot of the AI risk guys putting out stunned reports on the models doing things they couldn’t expect comes down to nobody understanding how subtext works when prompting. I’ll write more about this later but in sum:’These labs need more people with literary degrees.
