No, supply will increase. What you're not considering is that the manufacturers are building fab capacity like crazy right now. When all of that capacity hits (in 3-4 years), prices will crater. We've been through this whole cycle many times.

I remember the first big one, driven by the launch of Windows 95. MS-DOS could use memory above the 640K mark only with significant pain and complexity, but Win95 could address it directly and really liked having an insane amount of RAM... like 16 MB!!. Freakin' hog. My NeXTstation only had 8 MB and it blew those dumb PCs out of the water... anyway, I digress. RAM prices spiked, but within about 18 months (fabs were simpler back then) they went through the floor.

Not so much the raw materials as the fab space/time, but yes.

We have pretty extensive history of how this goes; there have been a half-dozen DRAM crunches in the last ~40 years. What always happens is that prices explode, manufacturers build out massive new capacity (which takes years, and as fabs have gotten more complex it takes more years), then all the new capacity goes into production at about the same time and RAM prices crater.

And we're definitely seeing the same processes at work this time. Between Samsung, SK Hynix, TSMC and Micron there's almost a trillion dollars of new fab construction going on right now. When all of that comes on line, it's going to provide a massive boost to output... and RAM prices will almost certainly drop precipitously, losing all of the recent gains plus whatever normal decline we'd have expected, plus some more.

When? Probably around 2030. 2029 if we're lucky, though we might see a little softening by 2028. Definitely not this year and almost certainly not next year.

It could happen faster if the AI bubble bursts before then. I wouldn't hold my breath.

What resources are we short on?

Argumentum ad homimen. If you can't actually point out what's wrong with the arguments, you haven't refuted them.

Or if, you know, we developed ways to generate and use that don't produce carbon dioxide emissions, and ways to capture CO2 from the atmosphere and sequester it, then we wouldn't have to slaughter billions and return to a subsistence lifestyle.

Can we maybe try a carbon tax for a couple of decades first? We can always do the murdering later. Given the birthrate trends, if we wait a century the population will crash without doing anything.

Until the AIs have actually taken control there's still room to try to figure out how to align them with humanity. I suspect we are running out of room, though. Fast.

Well, that does tend to produce about one baby per month on average.

The point is that the best way to eliminate abortion is to make contraception easy, effective and cheap.

That quote was in the article, but the article title was about a four-day work week. Kamelia Aryafar, the author, clearly thought (correctly) that AI would allow engineers to get a week's worth of work done in less than a week and then thought (incorrectly) that this would mean developers would respond by ending their week as soon as they had done that.

What actually happens, of course, is that if you can get done in one day what used to take five, you now do five weeks' worth of work every week.

Fundamentally, productivity isn't the constraint on working hours. If that were true, Keynes' expectation that we'd be working 15-hour weeks would be true. What constrains working hours is the limits of human ability to remain attentive and focused and the demands of human life, and those constraints turn out to be roughly 40 hours per week. A little more or less than that depending on the person and on the nature of the work.

Increasing productivity has never really changed working hours, it just increases productivity, which makes us all wealthier. Yes, all of us, including the workers -- if you believe all of the gains accrue only to the bosses or owners of capital you really need to learn more about what life was like in past decades and centuries.

The things AleRunner mentioned are plausible outcomes of misaligned artificial superintelligence, and the research in question is trying to prevent those outcomes. If you have good arguments as to why those outcomes are implausible, a lot of people would like to hear them -- including the AI companies and their safety researchers.

You misunderstand the purpose of this research, and the reason for publishing it. The research is not about "flexing", it's about safety, trying to understand the risks that we may be facing as the agent capabilities increase. AGI or not AGI is actually irrelevant here. What the researchers are trying to understand is what the agents will do when they face apparent opposition.

The cooperative results are great, because we think that's what we'd like highly-capable agents to do, to look for reasoned, peaceful resolutions. What we don't want them to do is to become aggressive, or to fail to step back and look at the larger situation.

And the reason for publishing it is to enable other researchers to see it and build on it, quickly, because everyone working on this stuff understands that it could pose an existential risk to humans. Find and talk to some people who work on AI research; they all worry about this. Many of them actually believe that we'd be better off if everyone stopped the research, or at least slowed it down significantly, but everyone knows that's never going to happen. So, their best option is to create the most capable and best-aligned AI, as quickly as possible. It needs to be well-aligned so it doesn't pose a threat to human existence. It needs to be capable so it can protect us from badly-aligned AI.

The IPOs won't stop the research, nor the publication. If you're not interested in it, just don't read it.

If students are being subjected to 4 GPa of pressure (40,000 bar), that's a real problem. I mean, that would crush their bodies beyond the limits of molecular cohesion, rupturing all cell membranes and turning all of their bodies' water into high-density ice. If the university is regularly murdering students this way, they should be stopped.

I'm not sure how pass/fail grades help, though.

Investors are chasing a company that achieves some kind of AI singularity. Let's set aside the fact that there's no reason to believe there is anything but diminishing marginal returns by making marginal refinements to current frontier models. Let's imagine someone hits the jackpot and gets, not even AGI, but a system that's as far ahead of today's frontier model are ahead of 2020's GPT 2.0 in performance.

Globally AI revenues are 150 billion, against a cumulative burn rate of 450 billion. A model that is a generation ahead of others would almost certainly capture the lion's share of that revenue.

If AGI magically appears as a Sam Altman has promised investors it will, a hundred million is way too low. Add, maybe, another zero to the revenues.

Conservatively, a safer assumption is that frontier models will get marginally better based on refinements in training and reinforcement and the other bits and bobs that go into these systems. The nobody is winning the lion's share of anything, at least overnight. But you have to define "safe". By "safe" I mean unlikely to lose money. But some investors are clearly defining "safe" as "having the greatest chance of owning a piece of the biggest thing ever."

I'm not following this super-closely, but if Anthropic is pursuing adding multi-step model based reasoning to their system, that could be the basis of a generational leap in capability. But if that is an approach that looks like it has a chance of working, then their competitors are no doubt pursuing the same thing. In that case you'd expect the revenue pie to grow as the scope of model utilty increases, but that growth to be split among several competitors. This could credibly result in a revenue stream for some of them that is as big as the entire industry's revenue stream today. But there's going to be hell to pay on the data center impacts end of things.

It's a good question, but I think framing it as what humans need to do to be able to live alongside ASI is the wrong way to look at it. The right way to look at it is "What do humans need to learn how to do to have fulfilling lives?".

Of course that question also depends on what makes sense for humans in a world run by benevolent ASI. We don't need to bother thinking about humans' role in a world run by malevolent or even disinterested ASI because there will be no role.

Sort of. When my kids were in public school in the 2000s and 2010s, book encyclopedias were sometimes grudgingly accepted as source. Not always, and it was often made clear that they weren't proper sources (because they're not!). The solution was to use Wikipedia but to follow the reference links to the primary sources it cites and use those.

I thought that wasn't just a good answer, but a strict improvement over accepting hardcover encyclopedias, both pedagogically and as a matter of scholarly principle.