"Now the only reason why I won't post this as an arxiv note, is that the humans actually beat gpt-5 to the punch :-). Namely the arxiv paper has a v2 arxiv.org/pdf/2503.10138v2 with an additional author and they closed the gap completely, showing that 1.75/L is the tight bound."
I really don't know what to make of this. The conclusion is that a model could still do this without the paper containing the exact info on how to do this ?
But I can complain about what happens to said something. If my blog photo becomes deep fake porn am I allowed to complain or not ? What we have is an entirely novel situation (with ai) worth at least a serious discussion.
There was a time when in the name of progress people were killed for whatever resources they possessed, others were enslaved etc. and I was under the impression that the measure of our civilization is that we actually DID care and just how much. It seems to me that you are very eager to put up altars of sacrifice without even thinking that the problems you probably have in mind are perfectly solvable without them.
One of the biggest factors in risk of death right now is poverty. Also what is being chased right now is "human level on most economically viable tasks" because the automated research for solving physics etc. even now seems far-fetched.
So it's bs but for money and therefore totally fine ? I think it's not ok if only a fraction comes true because some people believe in those things and act on those beliefs right now.
“We found that even state-of-the-art models which are optimally performant in data similar to their training sets are not optimal — that is, they do not make the best trade-off between overall and subgroup performance — in novel settings,” Ghassemi says. “Unfortunately, this is actually how a model is likely to be deployed. Most models are trained and validated with data from one hospital, or one source, and then deployed widely.”
I actually agree with you. I was a bit sarcastic. If I understand correctly there isn't a fundamental difference when it comes to text output vs pixel data output in this context. If so then it suddenly sounds much more of a stretch (intuitively) to claim that somehow stable diffusion understands the real world (like people claim to be the case with language models).
What ? Stable diffusion doesn't have an underlying understanding that humans typically have two arms, two hands and five fingers per hand gathered from vast sea of training data ? That's a bold statement.
With 3.6B people in the workforce I'd argue there isn't a billion people in need of a computer, not to mention an ai subscription plan. I'm of course assuming most subscriptions for ai are work related.
> I think what Suleyman is missing here is that most people aren't fine with their "TVs, laptops, in-car cameras and earbuds" recording everything continuously.
I really hope so but I'm not so sure. I suspect most people are fine with anything that isn't obviously intrusive. I sometimes get the feeling that people are more upset with cookie popups than constant invigilation.
> It also voids plausibility for Fair Use exemptions. I can see large copyright holders lining up with takedowns
I thought so for a moment but then again Meta, Anthropic (I just checked and they have a "for profit and public benefit" status whatever that means), Google or that Musk's thing aren't non-profits, are they ? There are lawsuits in motion for sure but with how it stands today I think ai gets off the hook.