pu_pe

A certain cult likes to claim credit for things that are happening with or without them, and this is my main argument against the valuation of frontier labs. It’s not that AI won’t create that much value, it’s that they won’t capture it.

AI is something that’s happening mostly due to Moore’s law and general progress in computing, not something that they are doing.

But if these companies control the vast majority of compute power, which seems like the plan they are already executing, won't they capture most of the value from the progress of AI?

A very pleasant and informative read. The hypothesis that climate change was the primary cause of what came after seems plausible.

The AI replacement discourse was meant for the investors. Now that they have mostly secured the money, they want to convince the consumers.

We have a demographic collapse looming in the horizon in most developed countries. If we find a way to use 1 human instead of 2 to produce the same amount of intellectual work in 10-20 years as we do today, that's a huge societal benefit.

The proposed arrangement would involve other US AI companies handing over a similar stake, although it is not clear if the other labs would be willing to do so.

They will be willing to do so if the government threatens to block their models, which they have already done.

I wonder to which extent this has always been how things worked and it was just hidden from the general public, and how much of it is a new level of corruption.

The amount of radiation exposure on a dental x-ray is still tiny though, we are talking like 5 or 6 orders of magnitude lower than radiation poisoning thresholds. Plus there is extensive epidemiological data about their safety.

Any suggestions for a VPN service with similar security standards as Mullvad?

If you invest into chips that deprecate in value really fast, not utilizing them to their full capacity because of power constraints would be counter productive.

He tried with a tiny model (gemma3:4b), got a range from 66 to 99. Then tried again with a small model (gemini 3.1 flash lite), the range was 48 to 64. Would a frontier model be more consistent? Perhaps this tool was optimized for more capable models?

I wonder what kind of scheme the administration is up to. The obvious play is a squeeze where OpenAI and Anthropic are forced to give parts of their company away, like Intel. But they could also be toying with the idea of limiting frontier AI access to companies that bend the knee, which would further cement their grasp on the tech industry.

You moved the goalposts. The government controlling what openai can and cannot do is completely different than they gating access out of their own volition.

Europe is in the worst spot right now, because even if open source is the future, there is not enough European-owned datacenters even for inference. Not to mention that China could pull the rug on these models at any moment just like the US did.

Without access to leading models, I think open source LLM development will also slow down. I'm not sure which portion of their success right now is due to RLAF and distillation but it's certainly not zero.

Argentina's interest rate is ~30% right now. So I would argue both the investors and the Argentinian government are aware of how risky these loans are.

The article doesn't really mention this, but it would be far more realistic, constructive and pleasant to focus on how we can all reduce our work hours with AI rather than straight out mass unemployment. In fact I think political movements that focused on that would gain a lot of traction, just like social democracy emerged out of the industrial revolution.

It's true that no one is trying to one shot anything serious right now, but it's still an important metric. Claude Code and Opus really took off when they improved the harnessing enough that it would self-correct many of its mistakes without needing user input. In fact I think long-term autonomy (in the range of several hours) and self-correcting is going to be where we see most improvements in coming years.

The fact this Midjourney delusional device is taken even slightly seriously is a major sign that we are in a bubble. A lot of people are pointing to Anthropic and OpenAI's investments into datacenters as problematic, but I think it's this ancillary scam ecosystem of things that truly don't have any intrinsic value that's much more dangerous.

So in at least two fields in which the US still leads (biomedical research, AI), government action is actively harming their ability to compete because of ideological or political reasons.

You can run it, but you can't train it. While this type of toy model could actually be trained in Swiss equipment, a state-of-the-art LLM probably could not.

My charitable reading of GP's point is that the bottleneck for true compute sovereignty is the chips, not the models.

It's not like their 10x revenue growth year-to-year was based on expanding their customer base 10x though. The vast majority of weekly users are not paying, so they are actually a cost right now, but could be converted into paying users or audience for ads in the future.

The Amodei quote they use is illuminating:

"The government should have the power to block or deter deployment of the model if it is determined, in light of third-party assessment, to present unacceptable risks. This power must be scoped to the above four specific risks and there must be protective measures against political favoritism or arbitrary decisions."

The article uses this as a gotcha, but I think there is nuance here. Amazon is a competitor, not an officially appointed third-party that can assess anything. And while Amodei says government oversight is needed, I think it's fair to say we would all expect something more rigorous and neutral than getting a phone call on a Friday based on vibes.

Do you have a source for this?

It stinks to high heaven, especially considering how over-the-top security protocols were introduced with Fable. The US government is asserting its influence on the economy and showing Anthropic that their IPO will depend on bending the knee.

The isochrone maps are quite beautiful [1], and go beyond the scope and refinement of some earlier human attempts I could find [2][3][4].

[1] https://isochronic-passage-chart.netlify.app/

[2] https://mapitout.welcome-to-nl.nl/

[3] https://commutetimemap.com/

[4] https://andrewding.ca/flightisochrones/

I never said an organisms behavior outside of reproduction does not have an impact on natural selection.

Um, no, actually AI makes it better because the cost is lower now. I'm not sure what point you're trying to make here, obviously organizations already fight against tech debt all the time through a variety of means?

That's one dimension of it, but in the context of this thread we are talking about how maintainable a codebase is for other humans. If your codebase is messy you depend on a few key employees and it might be hard to onboard new ones, so there has always been financial incentives to reduce tech debt.

Absolutely. I really don't think the future will be humans reading and picking apart an AI-generated codebase, there will be tech debt agents or whatever running overnight.

You are assigning intentionality to these mechanisms, but as the other commenter pointed out a much simpler way to view it is that death and senescence are just side effects, not actual mechanisms with evolutionary purposes. Organisms get born and need to reproduce, that's the mandate. There is no evolutionary pressure for you to die, so obviously DNA does not have organismal death programmed in, it has cell death which is necessary for organisms to live long enough to reproduce. Many of the things that are optimal for you to reproduce might not be optimal for you to survive a thousand years, and what happens to organisms outside their reproductive cycle is pretty much irrelevant to natural selection.

Please correct my math/assumptions

1) You forgot to include the actual costs of the GPUs and other equipment, you are just calculating launch costs which will always come on top of the other stuff

2) Datacenters require repairs and hands-on attention, and space provides extra challenges like radiation to deal with

3) They have not provided any evidence that they can solve the heat diffusion issue, which could kill the whole project

4) Musk's track record for outlandish tech is poor (hyperloop, fully autonomous driving, Mars trips, etc. were significantly delayed or not achieved yet), and unlike some of these other things no one else seems to think this is economically feasible