What's the expected state space of effective regulation though? Note that we've got passable coding models down to ~30B parameters by now. And keep in mind the ultimate floor here - the human brain only consumes on the order of 20 watts and fits in a handbag.
Is there any possible solution other than mass proliferation where the models are used to keep one another in check? Either that or a religious prohibition against the existence of integrated electronics.
What I'm asking is, why should we expect that to effectively further the end goal? You're simply asserting that it will ultimately do so.
Given the efficiency gains we've seen it seems to me that the situation has shifted from being analogous to producing nuclear weapons to producing something much closer to small arms.
A fairly obvious theory at that, unless I've critically misunderstood what's being described. As with so many obvious ideas I've always assumed that the reason I haven't come across it in the wild is because it doesn't work (or is comparatively inefficient, or tends to blow up during the training run, or etc).
> Realized reasoning gains, hardware efficiency, and RL scaling remain to be established.
Yeah so the first entry in that list is - if I may be so bold - typically what you'd start with at a small scale _before_ writing up and publishing your "genius" idea. This is the usual crank with delusions of grandeur presenting something straightforward that he hasn't tested as though it were a working breakthrough.
Ironically the cost of testing such theories has fallen to an all time low given the capabilities of coding models. I wouldn't be surprised if a frontier model could one shot a test of this.
This is easily the most depressing thought I've encountered in a while. It's entirely accurate and combined with talk of AI regulation it paints a picture of a very bleak future indeed. Locked down hardware given away almost for free, permits and AI based monitoring required for anything with the computational power of a GPU from 10 years ago, and market prices that place 64 GB of RAM in the same ballpark as a brand new vehicle from the dealer.
It seems the industry has collectively and nearly unanimously decided that the Verification Can future is way less cringe and way more preferable than the future where we are writing YAML by hand*
> watermarks never took off because the technical complexity and lack of standardization
Both audio (sports) and video (netflix et al) watermarks are alive and well. They are used to trace piracy (ie account ID), not for content ID (at least AFAIK).
Yes, it's possible. Some sports broadcasts do this. The ones I'm familiar with use a high pitched tone on the edge of human hearing but still audible to most people.
Streaming services watermark the video as opposed to the audio as I understand it.
But that's for tracing pirated content back to the originating user account. It's not particularly useful (also entirely unnecessary and overly complicated) if your goal is to ID a piece of content as opposed to an account.
> the market can't support the cost of generating video.
I'd suggest that's only the case given the current quality of output. Media is incredibly expensive to produce. A model capable of sufficiently high quality could charge prices that are absurd by today's standards.
It’s a very small set of buyers that are in that price range. Total annual domestic box office revenue is like $10 billion, maybe $50 billion for global TV and film. And that’s revenue, not profit, and a lot of costs are going to marketing, not to filming and casting. That’s a lot of money, but it’s not the scale that OpenAI and Anthropic are at.
Video generation would only make sense at that scale if it was targeting individual consumers, but then it’d need to cost something that consumers are willing to pay - which practically is probably a few hundred per year at most among US consumers, and much less globally, so again it doesn’t solve for the size of the AI companies.
I don’t see a way that video generation becomes a big industry without making generation much much cheaper.
Aren't these two largely separate questions? Viability versus if a given incumbent has interest in a market of a given size. With the combination of (at minimum) streaming platforms, the box office, and advertisements video and audio generation would be viable at a remarkably high price point (as compared to the current token prices for other sorts of things). And as the price comes down presumably the market would grow larger - by how much I have no idea but there are certainly a great deal of currently underserved niche markets.
I think we should use hex for brevity since consumers get confused by long numbers. We should also add adjective noun pairs to make them more memorable. And we can also add a paid sponsorship slot.
Ex 6G = 8B.9 verizon cool blue and 7G = A0.D costco ultra red or similar.
But the intent behind and manner of the learning is important. Similar to the difference between a chemistry professor delivering a lecture at a university versus someone paid to provide instructions to a known terrorist organization.
Is there any possible solution other than mass proliferation where the models are used to keep one another in check? Either that or a religious prohibition against the existence of integrated electronics.
reply