Post

*safe* AI progress And who gets to decide what that is?

ideally society participates. but society has to decide to participate, thus far it hasn’t worked out so well

Maybe society has decided they don't want AI and the AI people don't want to hear it. So far AI is on the hook for nearly 100 deaths (that we know about). So yea, I say keep AI the fuck away from me and every aspect of my life.

The shills get to decide according to them. Because only they can bring 'safe' AI about.

"Safe AI" is a giant oxymoron.

Turns out, the Chinese propaganda farms pushing anti-datacenter social media outrage in the US were being secretly funded by a cabal of US AI firms with the goal of plausible deniability when they throw up their hands and agree to no more datacenters...in the US and under US regulatory control.

I think currently the only real blockage is the resource cost of growing models. If it's possible to dramatically improve the energy/compute efficiency of these things, then we might genuinely be into serious foom territory.

Halting data-center construction is not a strategy to hinder AI. They are harmful by themselves.

i think that’s a very reasonable position as long as you consider only false information

Ah yes, the old "whatever disagrees with my position is fake news". Always a banger. Sorry, not gonna dignify that with a further debate.

we’re in an an realistically good position in that everyone who’s at the top of AI development wants to make sure it’s done safely that seems like a good thing. We should figure out how to make it even safer, put all of our effort into specifically that

Is that true in China, or just the US?

OpenAI has talked about how they shocked themselves into slowing down development bro, these guys have never once been the “move slow” types if you knew what they know, don’t you think things might be a wee 🤏 bit different?

We talk about the HuggingFace hack, but the real damage happened internally to OpenAI. The "rogue swarm" messed with their R&D and that's potentially a lot of damage.

Because everybody in society uses AI, it’ll take full society approach. We need scientists, engineers, scalers, policy analysts, writers, marketers, writers, editors, cryptographers, procurement profs, and many, many, more, from all ideologies.

yeah, i think one of the best ways for society to help out is to just give Anthropic and OpenAI tons of shit for breakdowns, like the Huggingface incident

The idea that DC-construction halts specifically make it "harder to encourage safe AI development" yet somehow not AI development in general smells like special pleading.

both of the leading labs are already 2+ generations of models ahead of where they currently have released publicly so… it’s **already happened**, that’s why i’m so sure of it

Are these labs there in the room with you right now? And is anyone wearing a white coat ?

I don't understand the relevance of that reply. I'm questioning the idea that DC-construction halts differentially slow "safe AI development" relative to AI development in general. Gesturing at alleged secret models inside the labs doesn't shed light on that.