rossdavidh

'...when this all falls apart, I genuinely think there's going to be an authority crisis where people look at the media and go, "Hey, why did you believe every insane thing said by a CEO for four years?"'

So, some of the capital invested due to the AI boom may be for stuff like buildings, energy infrastructure, and other long-lasting stuff. I am reminded of all of the spending done in the runup to the Y2K, when I saw factories able to get all kinds of upgrades to equipment done, outside of the normal budgetary restraints, simply because the vendors were smart enough to say "not fixable, you'll have to buy the new model". It could happen.

However, GPUs and memory chips are some of the fastest-depreciating capital investments one can make. Overinvesting in this generation's GPU model will either be wasted (because the chips are worthless in a few years) or result in a lot of underinvestment in future years (because instead of replacing those GPUs with newer models you keep using them, unwilling to admit you bought several times as much as you should have).

If we have reached the point in the cycle where the boom's proponents are trying to argue that even if it was all a mistake, maybe it's ok, then one suspects were might be late in the boom part of the cycle.

Great stuff! Great! However, the issue is not, and has likely never been, that young people aren't willing to believe "they can do that". It's that most people, of necessity, are not good at both inventing, manufacturing (how to efficiently make that invention many times), and sales (how to get someone to buy it). Therefore, they will need to find someone willing to hire them to do one part of that process. And the modern labor market is not interested in training people on the job, they want someone who's already done the job somewhere else.

Not saying you shouldn't go tell kids about how things work! Just that the reason kids don't do that as much, is not primarily anything to do with education or the kids, but rather because the modern labor market has done its best to avoid ever having to give anybody a chance to do something they haven't already done many times before.

In various places and times in the past, otherwise rational and successful people had all sorts of odd religious (or other ideological) beliefs, that caused them to do odd things. Sometimes, those odd behaviors even had a modest negative impact on them. However, they signaled (perhaps more powerfully because somewhat costly) to those around them that they were on the "correct" side of things, that they had internalized the religion of their peers. This may have paid off for them, on balance, by making them more trusted or liked by other believers.

Not that it always turns out to be a net positive, but it can often enough to keep happening. Some niche religions have too costly demands on their members (e.g. chastity for all believers, giving away all possessions, etc.), and thus fail to survive. But others, and other belief systems that function psychologically in a similar way, can survive despite (or even because of) the modestly costly irrational behaviors. Costly signaling is a more honest testament to your commitment to the ideology.

No

Might be worth trying live yogurt in the meantime, per the article?

That is a general truth of most ML; many models _can_ find the information in the data, if the data is good enough. If it is not, then likely no model can.

Several instances of both. In at least one case they told us that they wanted to rent to a big-name tenant at a larger space in the same complex, before they started renting any of the smaller places. What seemed odd to me is that they would nonetheless put the property "on the market", i.e. online listing, even though if you called up they would say it's not for rent. I figured it had to be to claim that it was "available for rent" in some technical sense, but I didn't know why.

In others they would just have too high a price, and it is still empty six years later so they are clearly not unaware that they have it priced too high. Simple math showed they would be better off renting at a lower price than going years without anything, but it was quite common so it didn't seem like it could be simple stupidity.

The test, I guess, of this explanation is whether or not the flow of money out of private credit causes the end of "extend and pretend", and the market finally clears.

Oh my goodness, thank you. My wife had to move her store in 2020 in the midst of lockdown; you'd think rents would have been low, but no. Since then, many of the places that wouldn't lower the rent then, have sat empty ever since. This is in Austin, TX, a town that has had a healthy economy during that entire time.

Weirder still, many of them were on the market, theoretically for rent, but if you called them up it turned out they weren't actually available, and the landlord wasn't interested in renting them. I couldn't figure out why you would pretend something was for rent at $X, and let it sit empty for years, rather than actually rent it at something <$X. Now it makes sense.

It seems to me he contradicts his own thesis here: "The boring truth is that expertise in most subjects is largely a matter of having an enormous library of knowledge and skill. For example, if you want to learn a language, you need to learn a lot of words. Any method that tries to skip over the fact that there are tens of thousands of words to learn is doomed to failure. All skills are like this, it’s simply that the “atoms” of learning are usually less obvious than in languages."

...but we don't learn our first language, or any other, by first learning a few thousand words and only then speaking. We start using the very first words we learn, in real life situations, and add words as we need them. It's the real-world applicability and project-based method that he pronounces skepticism of elsewhere in the same piece.

Every coach of every sports team ever, knows that you need drills, but you also need to play actual games, to keep kids motivated to do those drills.

True that, but it could be both cause and effect. There are reports that gut bacteria can induce depression, which benefits the bacteria because it sends more calorie-rich, highly processed food down the gullet (think ice cream binges when depressed). Not hard to believe that gut bacteria optimized for a particular kind of food could evolve the ability to induce their host to eat that food exclusively. It would have the side-benefit (for the bacteria) of reducing competition from other microbes that aren't optimized for that, by starving them out. Things aren't always just cause or effect, especially in biology.

Working on a framework for factory management systems.

Do they mean, the moment when everyone realizes it's not as useful as they at first thought?

Should note that "top US markets" really means "most expensive US markets", not for example the largest by population.

It's somewhat akin to how documentation inside code is so often inaccurate; it may have been fine when it was written, but it doesn't get updated.

Sorry, didn't know/remember that a GR account was required just to read...

One of the authors of this paper, David Reich, has written a book called "Who We Are and How We Got Here", which is worth reading. My thoughts on it: https://www.goodreads.com/review/show/2605841954

No matter what your politics, sooner or later someone you don't agree with will be in charge at the national level.

There are also cases where states take on cases that the national government never pursues in the first case. IIRC, states pursued the tobacco companies when the national government would not (Democrat or Republican).

Of course, it happens in federal courts, so you also need separate and independent branches at the national level. But states that can act independently are important as well.

In case you wondered what the point of the federal (i.e. states not totally controlled by federal government) system is, here's a good example. If only the federal government were allowed to pursue this case, it would have ended when the administration changed. 30 states chose to keep the case alive, and good on them.

I'm building a framework for making small-to-medium sized factory management systems.

True enough, but I think that a lot of "actually programming the thing" turned out to be "figuring out what exactly needs to be programmed". Afterwards, people did not want to admit that this was the case, perhaps even to themselves, because it seemed like a failure to plan. However, in most (nearly all?) cases, spending more time prior to programming would not have resulted in a better result. Usually, the best way to figure out what needs to be programmed, is to start doing it, and occasionally take a step back to evaluate what you've learned about the problem space and how that changes what you want to actually program.

In other words "figuring out what needs to be programmed" and "actually programming the thing" look the same while they're happening. Afterwards, one could say that the first 90% was figuring out, and only the last 10% was actually doing it. The reason the distinction matters, is that if you do something that makes programming happen faster, but figuring out happen slower, then it can have the surprising affect of making it take longer to get the whole thing done.

All of this article, both the good (critique of the status quo ante) and the bad (entirely too believing of LLM boosterism) are missing (or not stressing enough) the most important point, which is that the actual programming is not the hard part. Figuring out what exactly needs programmed is the hard part.

For reasons which it would take a while to unpack, if is often the case that the best (or sometimes only) way to find out what programming actually needs to be done, is to program something that's not it, and then replace it. This may need to be done multiple times. Programming is only occasionally the final product, it is much more often the means of working through what it is that is actually needed. This is very difficult for the people who ask for the software, to understand, and it is quite often very difficult for the people doing the programming to understand.

Most of what is being done, during programming, is working through the problem space in a way which will make it more obvious what your mistakes are, in your understanding of the problem and what a solution would look like. Once you have arrived at that understanding, then there are a variety of ways to make what you need, but that is not the rate-limiting step.

My favorite line: "I got basically all the extensions with this, making everything I did before this look really stupid."

Not at all; all good developers succeed by finding ways to make their past work look unnecessarily complicated.

So, "BS-as-a-service" allows you to make more money off BS, including deepfake before-and-after, and falsifying doctor profiles at scale. I mean, fraud is definitely one application where "hallucinations" are not so much of a problem, that is a valid point.

I don't understand; where is the cloud connection to an LLM agent? I don't even see how you could get Kubernetes or React in this thing.

If the algorithm (whatever it is) evaluates its own output based on whether or not the user responds positively, then it will over time become better and better at telling people what they want to hear.

It is analogous to social media feeding people a constant stream of outrage because that's what caused them to click on the link. You could tell people "don't click on ragebait links", and if most people didn't then presumably social media would not have become doomscrolling nightmares, but at scale that's not what's likely to happen. Most people will click on ragebait, and most people will prefer sycophantic feedback. Therefore, since the algorithm is designed to get better and better at keeping users engaged, it will become worse and worse in the more fundamental sense. That's kind of baked into the architecture.

Good idea; call it "Children of Cronos" (who ate his children).

Hypothesis: the ridiculous amount of resources put into the LLM (and other "AI") technologies currently, far beyond any realistic hope of profit from it, is because a bunch of techno-billionaires believe in Roko's basilisk.

I mean, I don't have any evidence for this, but it makes more sense of their actions than a profit motive, in this case.

One is reminded of the Behaviorist's attempt to do away with introspection and pretend that they could understand what was going on in the mind purely by examining behavior. Like the drunkard looking for his keys under the light rather than where he dropped them, because it's easier to see there.

This is an especially inviting mistake if one of the primary consequences of the internet/social media technology of the 21st century is mental health issues and doomscrolling ragebait. Andreessen helped to create this problem; no wonder he doesn't want to look at it.

"One or two hybrid days per week capture nearly all the fertility upside."

That is an interesting point, and not obvious why it would be so. In fact, it kind of calls into question whether the whole relationship is causal. The people who were able to WFH longer were more often in high-income jobs (service workers never got to do it in the first place, it was almost entirely an office worker thing). They were thus more likely to be in an economic position where they felt comfortable having another child.

This would also explain why it impacted the intensive margin (children per mother) but not the extensive margin (percent women who are mothers).

I don't have a problem with WFH where it makes sense, and I do think many societies need to look at how to help young adults become parents, but I am a bit skeptical of this particular relation. If you've ever been a parent with a young child at home, your estimate of how much work you could do would be possible is a lot more modest.