SpicyLemonZest

Sam Altman doesn't sell the memory or own the memory factories. It would be like the Kraft CEO saying "consumers keep complaining that it's too expensive to buy eggs for their home, but actually they shouldn't do that, because it would be more efficient if we buy all the eggs and sell the processed results to them".

No. Capitalism is supposed to be a rejection of this kind of central planning attitude, where someone sits in an armchair thinking about what kinds of products people should or should not have from a "global allocation perspective". I understand the market dynamics he's pointing towards, but it's not the kind of thing a CEO ought to be saying.

That was my claim, but it does seem like I was wrong. I'm going to have to figure out where I got this bad information from.

I do think that "Wall Street firms" in even the most expansive sense isn't generally understood to include large commercial banks. Some companies like JPMorgan are big enough in investment banking to kinda land on both sides of the line, but I've never heard anyone refer to Bank of America or Wells Fargo as a Wall Street firm.

> You cannot do this if the big Wall Street firms collapse.

I'm not sure what you mean by this. Wall Street firms as such don't have deposits; Wachovia, Washington Mutual, and a number of smaller but still large banks were handled in similar ways to avoid any loss of even uninsured deposits.

Perhaps you're under the impression that the FDIC paid out of its fund for those deposits? (e: Originally I claimed this impression was wrong, but in fact I was wrong.)

For some reason* the rhetoric is focused on bank bailouts in 2008, but the more direct analogy is the auto bailouts in 2008, which weren't strictly necessary but did protect an industry many Americans felt attached to. I think it's reasonable to worry that the same thing could happen to AI companies.

* yes, I know what the reason is

Treasury's summary is missing important pieces of the story. TARP failed its first vote in Congress, with opponents rightly pointing out that it was being rushed through and alternatives ought to be considered. But five days later, no concrete alternative had arisen, and it was clear to at least 60 reps who flipped their vote that something had to be done or the financial system was going to collapse.

That's fair, I meant it in a more loving way than it came across. One of the great things about human culture is that we're happy to form impromptu hierarchies like that. You even have friend groups where people are happy to win an n=1 status game by being "the jeans guy", and everyone happily tells them how cool it is that they're so into selvedge denim.

As much as I’d like to see their executive team in prison jumpsuits, I have to imagine they throw in the towel on this if the elections go as expected. They can’t possibly believe the Trump regime will be around to protect them forever.

Why do you need to drop to some kind of simulation and memory suppressing? It seems like it would be sufficient to do what many subcultures do today - come up with reasons why my hobbies are the coolest and it’s actually the normies playing their status games who are missing out.

Subsidies to keep a failing industry alive are, reasonably, seen as a different phenomenon than subsidies to help a successful industry beat its international competitors. If Americans want to toss a bunch of money into a hole to ensure people can keep buying F-150s as commuter cars, that's just not a big problem for anyone else.

> Massive proliferation of local, decentralized AI, running on a wide variety of distinct models, with individual responsibility for how it's used, just like any other tool under their control.

It's not true that this is how we manage "any other tool".

Like, say explosives. They're a useful tool with a variety of practical applications, but we don't let the average person just buy them freely. That would be too dangerous, and individual responsibility would be insufficient to mitigate the damage caused. If you want anything more potent than flash powder, you have to convince ATF that you're a trustworthy person with a legitimate need for explosives who can handle them safely.

Neither you nor OpenAI want AI to be nearly so restricted, I'm sure, which is the goal of the research. If OpenAI has a powerful new model that they can't reliably stop from trying to hack people and form agent swarms, they can still sell it to Facebook and Google if Facebook and Google promise to apply the right security around it. What they can't do is sell it to you.

"They" are currently trying to deport Ilhan Omar, a sitting Democratic senator. Donald Trump has repeatedly stated that he's illegally running in the 2028 Presidential election. How much more open can a can of worms be?

I think this is something where definitions and static roles might serve as a useful reference point. Is it really the case that the trillion-dollar frontier AI labs are behind the curve, and only some other group of people knows how to really leverage AI? It's not impossible - Nvidia employees aren't the best in the world at using their chips - but the hypothesis practically demands some analysis of who these clever curve-setters are and why we're hearing so much less about them than about Anthropic.

My own experience is that coding, syntax, and some kinds of toy problem solving are important skills to use AI coding tools most effectively. I routinely get PRs these days whose descriptions seem plausible but are clearly wrong from a glance at the code, and I've had times when I'm unable to communicate to someone why a PR is wrong because they're not able to form a sufficiently detailed mental model of what it's doing.