I do ML in my day job and was working toward an advanced degree in the field of natural language processing when the now-famous Attention Is All You Need paper released. It really was transformative for my field, and was recognized as such at the time. But I don’t think even most smart and informed people working in ML anticipated that this model architecture would produce the capabilities we see today in less than 10 years’ time. I, along with many people in the field I talk to, continue to be astounded by what these models can do, even with their limitations.
That is to say, I am skeptical there is much correlation between the depth of one’s understanding of the linear algebra techniques that undergird the transformer architecture and one’s ability to make accurate predictions about what impacts these algorithms will have on society.
We shouldn’t fall into the trap of assuming the nerds, myself included, have any secret wisdom about the future. As a group, we were wrong about the second-order effects of the internet, and I suspect those predictions were made with much more earnestness than the pitches the AI companies are making.
My understanding, as someone who has been adjacent to the AI safety communities since the late 2000s, is that a lot of these people were already convinced by Moore’s Law that significant AI was coming in the next couple decades, and were thinking about the kinds of issues around disempowerment and existential risk for that reason. Some of them locked in the idea that AI would be basically a Turing computer, and never fully adapted to the neural net and then LLM versions we got in particular, but some then just saw these technologies as the medium through which their worries might happen.
Two notes (which are totally compatible with your takes):
The prevailing opinion in the late 2000’s was that we were still several decades away from real progress (“by 2100” was a popular guess, but the other side of that was a live option too). Basically nobody (except Ray Kurzweil and Shane Legg, I guess) thought the 2020’s would be the big decade for things starting to work.
We don’t have superintelligence or anything at the moment, so it’s not that the old opinions are definitively refuted, but clearly shorter timelines are a million times more common now than twenty, ten, or even five years ago.
A lot of the discourse also started in a world where we had no idea how to get systems to do things they weren’t explicitly programmed to do, and a worry was that we’d stumble onto the right (wrong) one, and in so doing create Skynet. We now at least know that it’s possible to get emergent programs that follow our instructions and have stable-ish dispositions. This is very good news, even if it’s not the whole ballgame.
A very similar thing happened after the development of nuclear weapons - nuclear scientists were suddenly held up as policy advisers, even though understanding the physics of nuclear fission doesn’t translate incredibly well to actually designing policy around them.
There’s also a less-inflammatory version (because the models are harder to anthropomorphize) going on with driverless cars. Are driverless cars going to lead to more VMT? Are people going to own a car in 2045 or just hail robotaxis? I don’t know, and I bet the engineers at Waymo don’t know either! This is all to say that it’s really tempting to think/hope technical experts have the ability to predict policy implications, but usually they don’t.
All true, but your *average technical person working on a particular issue* knows a lot more about that issue and has probably thought more about related policy implications than a layman or even a policy person.
i.e. Dario Amodei has a much better developed viewpoint on AI and its implications than the vast majority of policy people working in Washington DC (especially before the last ~2 years when AI policy became in vogue).
I have the exact inverse take, which is that anytime I hear someone saying that AI research scientists predict massive social disruption, job loss, or even post-materialist utopia, all I think is that these linear algebra wizards have zero idea how society works and I put no additional weight on their predictions.
I agree with the second paragraph inasmuch as deep understanding of linear algebra doesn't seem like an especially important prerequisite for prediction-making--chiefly because the underlying architectures aren't necessarily the natural level of abstraction that people or should or do think about with respect to AI--but I think your last paragraph goes too far in its implications.
The extremely detailed AI-2027 timeline is proving spookily prescient, Amodei's predictions regarding "people will basically stop writing code' have been borne out. And here's Dylan Matthews on "The AI people have been right a lot."
Hi. I work in the tech industry. No, 'people will basically stop writing code' has not only not been borne out, it's not even remotely close to true. Like, the exact opposite man
I am 'in tech' as a data scientist. I have stopped writing code and am better for it. I'm no dev, and I won't pretend my experience will generalize to proper devs, but the claim is at least remotely close to true.
Seems wildly off base. You just need to look at the Claude Code leak to see the entire thing is filled with manual memory and QA overrides. It's not at all hyperbolic to say that ~ 80% of production output is flowing through the human written code base.
Exactly. These AI doom lords have a lot in common with the string theory fantasists who predicted that science would end after the discovery of the GUT. Well, we're still here. There's still a lot to figure out including with how these AI tool will play out.
I just want to know how I can even evaluate the doomer claims. In three years if AI is a big business with useful applications but otherwise things are swimming along and we're all still alive are they more right or wrong? What about 5 years? 10? Am I supposed to treat them seriously forever regardless of what the actual literal capabilities of these systems are?
What if someone hooks Claude up to a nuclear arsenal and it accidentally launches everything, were the emergent super intelligence people correct or were we just fucking morons?
"Grand Unified Theory" of physics. It means any theory that "unifies" three of the four fundamental forces of the Standard Model. I'm not a physicist and can't explain exactly what that means. It's still not achieved -- we still don't have a worked-out theory that fits everything that's known, makes new experimental predictions that are subsequently confirmed, etc.
I don't understand string theory either, but I thought that was an attempt at a "Theory of Everything" that also includes gravity. Gravity in general relativity is a curvature of space (and time), but it's very hard to make a quantum version of that theory without producing paradoxes, infinities etc. String theory is, iiuc, a broad class of approaches to quantum gravity, but with important problems for actually confirming any theory --e.g. there are many possible versions of string theory, and their distinctive predictions are infeasible to test in experiments we can actually build on a planet.
Also, if you happen to be into science lessons in the form of a cappella song parodies (performed by one man with lots of editing), there's this classic video. Again I have only a limited understanding, but it can give a non-physicist some sense of what problem string theory aims to solve. https://www.youtube.com/watch?v=2rjbtsX7twc
Not only that the stories the nerds tell about AI risk are exactly what you would *expect* them to say for other reasons. They fundamentally center intelligence as the most important thing and then tell a sci-fi story which puts them in the role of the hero.
I'd be much more persuaded [1] if the story they had about AI risk made it seem boring or centered some other kind of actor -- for instance you could imagine a concern focused on how humans will react/interact with AI that centered social scientists or one that focused on geopolitical or economic risks.
---
1: Also the alignment arguments are just bad. They fundamentally assume that AIs will somehow maximize some really simple moral function when the very thing that makes modern AIs powerful is their ability to learn quite complex context dependent heuristics like language.
I like a good grumpy statistician as much as the next guy, but I have a hard time being upset that a different culture came up with different names for the same thing. One time I asked a stats grad student what backprop was and he said "the chain rule." And he's right! NNs are a bunch of stacked logistic regressions (or hinge regressions, whatever "non-linearity" aka link function they're using nowadays). Bias is an intercept term. At least they didn't try to redo "regularization."
The statistics community has never come up with a term as cool as "support vector machine," even though it's just shy of word salad.
I'm very charmed by how clearly obviously it is that NNs were thought up by CS people. They don't like math, they like graphs. An NN is an abstraction of math that lets you focus on the graph and think about it like a software diagram. They don't care about p-values or overfitting or whatever because n = 1 billion and they can just look at OOS validation and regularize and it's fine, shut up about it.
Further in this, it is hard to get upset about naming being difficult across fields, when naming after people remains the norm.
One of my favorite lists is the one of "things unexpectedly named after people." French drains being the latest I learned on there. :D
CS is full of things that are a ton easier for me to understand when I realize a possessive was dropped. Bloom's filter and Shell's sort being my favorite examples.
Nerds don’t have a secret wisdom, but they understand AI enough to be able to think about its implications. Most laypeople have no idea what AI is going to be like, and cannot begin to think about its implications in any meaningful way.
FWIW, I don’t think you need to be deep into ML algos to have the necessary understanding of AI. But you do need to have at least a somewhat technical mind. Lots of (non data) journalists, lawyer, and politicians types have no idea what the future holds.
Thing is, most laypeople have no idea what the coming AI would be able to do and to what extent we would be able to control it. They are thus completlely unable to think about its impact on society or the dangers it poses. Nerds don’t have a secret
The center-left media refuses to admit that in much of Europe, immigrants commit crimes at meaningfully higher rates than natives. A lot of this is a composition effect — Europeans are heavily socialized and well-behaved, so admitting almost any other group will mechanically push crime rates up. Germany's 2023 police data makes the point cleanly: non-Germans are about 15% of the population but 41% of violent crime suspects, and asylum seekers specifically were roughly 2% of the population and 15% of rape and sexual assault suspects in 2017 (the pattern has held since). Yes, young-male demographics and socioeconomic status explain part of the gap — but only part.
The problem is European elites don't want to admit this. Rather than an honest discussion of the tradeoffs — more disorder versus a younger, more dynamic workforce and a shored-up welfare state — voters get force-fed pieties about equality that don't match what they see on the street. Add the fact that subsidizing unproductive new arrivals strains the welfare state rather than rescuing it, and the electoral success of the European right makes sense.
As one of the European SlowBorers that lives in the US, I think that this is unfortunately correct. The immigration debate gets very confused between the US and the EU, and you have American right-wingers (like Vance) who think that immigrants in the US don’t assimilate, and EU officials who think that immigrants in the EU do assimilate. I think that American media also doesn’t understand basic things about Europe (lack of understanding may be a charitable interpretation), so you get a lot of confusion here. I was pretty worried about Meloni’s election but she’s been basically fine, and I wouldn’t trust any media source that continues to label her as “far-right”.
PS: I see some commenter below who says that if you wrote something similar about the US, it would be considered racist. I agree! The immigrant experience in the US is VERY different from the one in the EU.
The weird think about the US immigration discourse is concern about Latin American immigrants assimilating. If there is one thing that the last century has proven, it's that we can absorb a huge number of Latin American immigrants and assimilate them to US norms. Grandpa comes to hang drywall and doesn't speak much English, Dad is fluent in both and speaks English with no accent and retains a lot of the culture, and his son is a generic American with a hispanic name who still speaks a bit of Spanish to talk to his grandparents but who is way more comfortable in English, prefers American football to soccer and pizza to pupusas or whatever, etc.
One of my favorite "sort of assimilated" moments. I was on my way to visit my parents, and stopped at this ginormous indoor flea market in rural Tennessee. (North of Cookeville, to be precise).
The interior was typical - booths around the edges, and in the center, selling everything from cheap knives to household bric-a-brac. And in the most prominent location - front and center from the main door -- was a double-sized booth selling fuzzy blankets. On the display rack, there were two blankets: Our Lady of Guadalupe, and a Confederate battle flag.
Isn't that just because Europe's immigrants are newer and haven't had time to assimilate?
Comparing European immigrants who arrived 7 year ago to Hispanic immigrants that have been in America for 3 generations seems like an apples to apples comparison. Just look at all the American angst over, say, recent Somali immigrants. Very few Americans outside of places like Slow Boring are blase and assume they will assimilate in 3 generations.
Likewise, Europe used to have tons of immigration and nobody is now complaining about, say, Italian minorities in Croatia or French refugees from 1848 who went to Spain, not assimilating. It's just Europe did all their assimilating 100 years ago and hasn't had much opportunity since. I don't think that means they are somehow unable to do it, given their proven track record of creating national unity in Italy, Germany, etc in the not too distant past amongst people who did not really see themselves as a single people.
It just takes time, which Europe has not had .... Yet.
I think this is the activist vs normie split. For activists, a line of "ningun ser humano es ilegal" works well, as does a "you will not replace us" rejection of all immigrants. But if you're trying to actually govern well, you want to allow some immigration but control it and be a bit selective about who comes here wrt whether or not they're likely to be net contributors to the society.
Either the US or any European country is capable of having selective immigration. The US does a better job of being selective but then has done a bad job of policing illegal immigration. Thankfully, our unselected mass has come from a part of the world (Latin America) that's less culturally dissimilar to us than the Middle East is to Europe. So we could be doing a BETTER job, but things could be worse all things considered.
I don’t think that the US necessarily does a better job of being selective. If you are extremely smart, you’ll go to MIT not TU München. If you have a groundbreaking idea and want to start a company, you’ll go to the US not France. Given who actually is interested in coming to Europe, I think we’re doing a decent job with being selective.
I don't think Europe's main problem is in selection effects, though it's definitely a contributing factor.
It's that European nations, except sort of the UK and Spain (at least as pertains to Latin American immigrants) neither require nor even really *allow* immigrants to assimilate.
In terms of culture, economics, and policy, a society really couldn't craft a system where the barriers to assimilation are higher and the payoff lower than in contemporary Europe if it deliberately tried.
Whereas in the US basically every aspect of our society pushes immigrants to assimilate quickly and be productive. If you don't you'll live a shitty life and your kids will knuckle under and do it anyway, and once you do you're a real, genuine American in the eyes of basically everyone except the Twitter natcon rightists.
Are Latin Americans in Spain immigrants? :P I know what you’re talking about, but I consider it borderline immigration.
We agree on one of the main problems with integration of immigrants in Europe. We are nation states (what I sometimes see called “ethnostates” in the US), not nations of immigrants. We had many wars to make sure that our countries are ethnically homogeneous! A lot of people were killed or forced to flee so that everything aligns with the lines on the map.
The other problem is that after living in the US for a few years, I’ve come to believe that a generous welfare state is incompatible with a generous immigration policy. I think that all the immigrants in Europe that don’t work are basically being rational, and I believe that integration is hard without work.
The only immigrant groups to the US with below average education levels are either from countries nearby (Mexico, Central America, the Caribbean), or from countries where most of the diaspora in the US are refugees (Somalia, Afghanistan, etc.). Every other country is sending disproportionately educated people to the US (not that education is the only thing that matters, but it's a decent proxy).
That's in large part the choice of Western Europeans (says the Eastern European who studied and lives in the US, along with many others of my type). Most Asians i speak to in the US also did not feel welcome in the EU.
This has actually changed somewhat in the past 20 years. But a) the EU has been unwelcoming for a long time, and b) it's not even clear they are more welcoming today.
I watch a lot of German police procedurals, mostly "Tatort" and various "SOKO" series, and at least during the late teens and early 2020s it was all immigrants all the time. Seriously. Now, if "Law & Order" isn't a documentary, then neither is "Tatort." But it certainly tells you about the concerns of the TV-watching public in Germany.
I think the rightwing argument is that this is a statistical illusion, because immigrants (by definition) have been here a shorter period of time. I.e. if you compare an immigrant who's been here for 5 years to a native who's been here for 40, of course you're going to see a higher crime rate for the latter- they've had more time to commit crimes. From what I've heard, if you compare for 'years you've had to commit crimes on US soil' you get different results
Considering how 1) a lot of the immigrants the Right is worried about are younger and male and 2) people tend to commit crime (especially violent crime) in that same age bracket, that's kind of an odd counterargument. If anything, you would expect a disproportionately young and male demographic to have higher crime rates without any statistical massaging.
I don't love that you could write the same couple grafs about America with only a couple word substitutions, except then it would rightfully be called rather racist, while still probably getting lots of upboats because something something hawkish commentariat.
Except we have a more violent baseline and less violent base of immigrants than Europe, so it would just be racist here rather than expressing an unfortunate truth
One difference is that the US has a lot of crime and some high-crime demographics, so immigrants who have the same propensity to commit crimes might be worse than the natives in Germany or Sweden, but better than the natives in the US.
I seem to remember an article a while back where they interviewed a bunch of people at an AfD rally about why they were supporting the party. It seemed like most were lifelong CSU/CDU voters who just wanted their party to do *something* about immigration. It's weird to me that the CSU wouldn't adopt more moderate positions or at least acknowledge it's a discussion worth having.
Brandon Johnson’s problems go well beyond policy. He’s one of the most incompetent executives in any large organization on the planet.
He has a dysfunctional relationship with the much more popular Governor Pritzker and the legislature— and his team sometimes doesn’t even advocate for their own positions.
He has been unable to pass his preferred revenue policies, losing a referendum on a real estate transfer tax and having to accept the city council’s budget which excluded a business head tax.
He has repeatedly been unable to get his preferred appointees in office in a timely way (like the Zoning Board chair being vacant until his a candidate he didn’t want was appointed this week), and at other times has fought with his own appointees. For example, the school board resigned en masse, and rather peculiarly refused to comment, when Johnson ordered them to fire the schools CEO whom he had allowed to stay on; even after appointing another slate he was still unable to get CPS to do the irresponsible borrowing he desired, and the new CEO (promoted from acting after a long delay) seems to be refusing to shut the schools down for May 1 as the teachers union and mayor want.
You missed out on a rigorous, informative discussion of British colonialism of India that somehow emerged from a total non-sequitur insult that Freddie leveled at Matt!
A problem with putting a militant teachers union's militant in charge with governing a city is they don't do the whole "balance various interests and concerns" well. This is part of the whole issue with the "Chicago School" of left wing politics (which Brandon is very much a product of). You're so focused on destroying the enemy/overthrowing the system* that you can't do normal governance things like appeal to the business community for investment or tell unions "No property taxes are not primarily about funding the bottomless pit that is your poorly run pension plan"
*there was real value in pointing out the problems with Daley the Elder's system, but that ended a long time ago and it's time to move on.
Agree with this, but as I commented on the original question: being mayor of Chicago is one of the worst jobs in politics because the city is functionally bankrupt. The city continually needs to raise taxes and cut services to pay for pension obligations and debt. That is extremely unpopular with voters. I don’t think any other major American cities have this problem.
At the same time, Chicago mayors have VASTLY more power than, say, NYC's mayor because of the lack of a city charter. City Council here is not funded at a level that would allow them to do their own fiscal analyses. This is the root cause of a lot of our problems. A couple of good primers on this topic:
I agree that Johnson has been bad, but the unfunded pension obligations are so large that I don’t think Chicago has any feasible path to paying them regardless of choices. The city needs a state constitutional amendment allowing them to reduce benefits and/or a bailout from the state/federal government.
I think so. He seems to have the progressive lane to himself if he chooses to run, so would have good odds to make a runoff.
I imagine he loses, but that may be the plan on the part of the Chicago movement left—the mayorship is quite the poisoned chalice given the revenue situation, and Johnson has does his best to dig a deeper fiscal hole. So they need someone not associated with the left to make the tax increases and cuts, without their fingerprints on it. And they can blame the failures while in power on Johnson himself and his cronies like Jason Lee rather than the blind spots of the larger movement.
I think we are beyond tax increases and cuts. I don't see any way out of Chicago's fiscal hole except state-facilitated bankruptcy. It's a question of when, not if.
The city still has options for now although I wouldn’t push back on your assessment over a longer horizon, unless city governance quality gets radically higher than it’s been in decades. The city will probably need to go through a round or two of austerity before the state allows restructuring—and the state isn’t in good shape either. But I don’t fully understand the political landscape here, the contours of any planning on restructuring would be kept very quiet.
CPS is in much more immediate difficulty and the state might need to take some control in exchange for a bailout.
Its own pension system is about as grossly underfunded as Chicago's, it has very high tax burdens and net outmigration, has serious shortfalls of infrastructure and service provision...
Sure, for now. Illinois is in better shape than Chicago, which is in better shape than CPS. The state just bailed out Chicagoland transit in exchange for less mayoral control.
CPS’ credit rating is already terrible and they have limited revenue options on their own. The city is in trouble on its own and the TIF surpluses are finite.
And even beyond anything he's actually done or not done, he's a singularly incapable communicator to a degree that's hard to believe for this level of politics. And rhetoric is typically an area where left-wing figures have an advantage over their more pragmatic competitors, it's the opposite in Johnson's case.
There's a lot to debate about *the idea* of Brandon Johnson, that *idea* won the mayoral election. But the story of his mayoralty is that the person himself is a far, far cry from that idea. In that way he's just not a comparable figure to other progressive mayors across the country, it's apples and oranges.
I feel like this is a much better answer than the one Matt gave. The question starts by asking about Johnson and MY barely mentions him in the answer, this feels like a trend to me.
A quick AI related anecdote. Last week I was visiting a collaborator to try and finish off a draft of a math paper (ca. 25 pages). (It is some fairly computational stuff in group ring theory applied to low dim topology if that means anything to anyone). At one point we were wondering where a general argument we have breaks down in an exceptional case (something that works for all Baumslag-Solitar groups except for the solvable ones ... i'm writing allthis to convey the message that it is a narrow and specialized situation). We knew based on an example that the argument doesn't work in the exceptional case but couldn't spot where the general proof (which is spread out throughout the paper) breaks down the way we had written it. So, collaborator fed the draft to several chatgpt instances and one of them (after 10 minutes or so) both pointed out the counterexample we knew and also pointed to a specific sentence in a lemma where the assumption that the BS group wasn't one of the exceptions was implicitly used. (It was pretty nonobvious, the sentence was vague and nothing about the context suggested that special assumptions were used.)
It is the sort of thing we probably would have figured out an hour later, and a smart and dilligent grad student might pick up on after a week if they had to referee such a paper from scratch (a thankless task).
Now, it is true that the other instances gave counterexamples but didn't spot where the proof breaks down, and that this level of usefulness was somewhat unusual among the other things we tried, but it is clearly not nothing, or 'hype'or whatever.
I think I misunderstood where your earlier comment was coming from. Military logistics sounds like something i would quiz someone about at a dinner party, no doubt. But I imagine that if you were to talk about it with the same laypeople repeatedly, it would just degenerate to the same conversation over and over. Am i getting that right? So in practice, if you talk to family about work it is about the human parts, both positive and negative.
Family, food, travel, health, sport, money, even politics ... it is rude to talk about work in mixed company, just as it is rude to speak in a language only a proper subset is comfortable with.
I find it really interesting that advanced math is the one area where I hear people using ChatGPT specifically in effective ways. Everywhere else it seems to be Claude.
I use Claude for most other things, but ChatGPT for some reason continues to trounce it on research level math. I recently fed ChatGPT 5.4 in the chat window and Opus 4.7 max effort in Claude Code the same question - ChatGPT gave a decent if speculative answer and flagged where the uncertainties were, and Claude made startling elementary errors that made its answer worthless. No contest.
(Grew up with Celsius and switched to Fahrenheit only after moving to US for grad school, so have a lot of experience with both)
Fahrenheit is just superior to Celsius for everyday use (at least in non-arctic climates), and this is a petty hill I will happily die on
1) it's a true 0 to 100 scale. 0F is very cold, 100F is very hot, and this is the range that covers most people's everyday experience.
2) it almost entirely eliminates the conversation: "-what's the weather like?" "it's 5 degrees." "plus or minus?" which happened ALL THE TIME when i lived in Northern Europe (where its around 0C for a significant proportion of the year). With C the only way to avoid it is to always specify plus or minus but that is an extra word and a lot of people chose not to say it. With F you can have the same problem, but in the range that is much more rare, and, frankly, difference between -5F and 5F is much less important.
3) 1 degree of C is just a bit too much. As a result, a lot of the time you need to think of it in whole degrees even for everyday stuff, and certainly can't just round to nearest 10 (difference between 5 and 14 or 11 and 19 is massive). With F, most of the time the first digit of the temperature gives all you all the info you need to know.
Basically, it is almost perfectly tuned for the temperatures that people actually use. And guess what, scientists, always switch to the temperature scale that is tuned for their specific need, whether it is Celsius, Kelvin or electronvolt.
Fahrenheit got really lucky with that! He may have been thinking about the 0 and 100 thing on some level when he decided human body temperature would be 100 and the coldest he could achieve with brine would be 0, but no one understood yet how ordinary people would be talking about weather in terms of temperatures, and that these ten degree increments really were basically right for planning clothing, or that 1 degree would be pretty close to the just noticeable difference.
There is a long rant from Jon Lovett (Pod Save guy) about how imperial measurements are superior to describing the world as it is experienced by actual people. Along these lines.
And it makes sense - the general conservative point is right. Traditional systems developed by people for a purpose usually work well for that purpose. The good feature of the SI is that it enables equally easy talk about all measurements at all scales. But if there is a particular scale you are constantly talking about, it’s better to use unites adapted to that. Don’t tell me street addresses in latitude and longitude - tell me with reference to the local streets! Even astronomers use the “astronomical unit” rather than megameters or gigameters when talking about distances inside the solar system.
Ok, but can we agree that traditional unit conversions are really dumb? Why 12 inches to a foot? Why 16 ounces to a pound? Who the heck multiplies or divides in their head by 16?
In metric, it's simple - everything is a multiple of 10, super easy to convert. 1 meter is 100 centimeters is 1000 millimeters is 0.001 of a kilometer, easy peasy.
Right, and this is the one big flaw of the metric system, not being able to divide evenly by 3 and 4, those are very common divisions in everyday life. If only we had 6 fingers on both hands instead of 5...
The reason for these weird conversions is that you shouldn’t be doing these conversions! It’s no weirder than 9.461x10^15 to convert meters to light years!
The point historically was that you measure things like precious metals or oils in ounces, and drinking liquids like water or milk in cups, just as you measure rooms in feet and travel distances in miles.
Metric is great when you’re measuring things across scales without any distinctive scale as special (like when you’re converting recipes to feed an army). But if you’re always working at one scale, you want the units for that scale, even if they’re not a nice power of ten times units for another scale.
I see people talking about conversions as the reason metric is way better, but I feel like this overstates how often most people need to do conversions? I'm fully on the side of metric for mass (especially cooking), but for temperature I'm not calculating energy and pressure regularly so I don't need to convert anything, and for distance the things I measure in feet and miles are fundamentally so different that there's no actual conversion happening. I'm six feet tall but I'm not going for a 30,000 foot run.
Fair point, but it depends on what you're doing. As a molecular biologist, you frequently have to convert between pico, nano, micro, and milli (grams and molar).
Given that really bad things happen to the human body above his estimation of 100 degrees that can be aggravated by hot climate, he might not have been *really* lucky, even with some luck.
I always found it hilarious when my fellow students and postdocs would seamlessly switch between Celsius (for experiments) and Fahrenheit (for the weather).
"All right, I'll put the E. coli in the 37-degree incubator and head out. The weather's really nice, it's up to 55 degrees already!"
In addition to this, measuring even big differences in human height with whole numbers of inches seems markedly superior to looking at decimal points.
Maybe it’s just because I am familiar with imperial, but someone being 6 ft 5 versus a 5 ft 7 person seems much easier to understand than 1.96 vs 1.70 meters.
I work in a more precision-oriented industry than many and I just don't buy it. It was fine when I was doing fabrication but in aerospace welder essentially use metric fractions of inches (thous). It's kind of ridiculous.
I don't see any reason why dealing in thousandths of an inch is any different than metric in this case. Neither one is dealing with human relatable scales and neither one really gives neater or cleaner numbers to deal with. A .001 tolerance equates to either 0.0254mm or 25.4 um which is something you might see on a sealing bore. Sheet metal might be .030 which gives 0.762mm or 762 um.
It seems that metric is superfluous given that imperial/EE exists. Metric came after and was only really useful to scientists when long and complicated calculations had to be performed by hand. In the modern world computers do all of that. Computers have eliminated the one major practical advantage of using metric but nothing has eliminated the usefulness of having units that are easily translated to human experience.
Would there be some huge gain if you were used to 1.6mm and 2.4 mm rods instead?
Cause it’s slower, and normal speech is not maximized to be unambiguous…When it’s -20, is perfectly reasonable to just say 20 (and people do) no confusion, so I guess people just extend it to much.
Also where the Celsius and Fahrenheit scales cross, which is why songs, movies, TV shows, etc. almost always say "40 below" when mentioning cold temperatures -- it communicates "very cold" to media consumers under either scale without needing any translation.
This is an interesting position! I also started using F after coming to the US for grad school and you’re right - it’s very useful! I do miss g and kg and m and km though, and especially the easy conversions!
When I learned statistical mechanics and saw that actually what appears in all the equations is kT (Boltzman's constant times temperature in Kelvin), I realized that actually temperatures should always be specified in electron volts. I drew a scale on a piece of paper for easy conversions to/from Fahrenheit and Celsius.
I also prefer F for human things and I am constantly annoyed that the thermostat for our reverse cycle air con can't do 1/2 degree steps. Sometimes I feel like I just need 1 degree F cooler/hotter but can't do it.
Regarding Twitter. I’ve been on there recently because there are people like you are on there.
I don’t know why you are leaving out the fact that Musk is clearly using the platform to boost virulently bigoted white nationalist and yes I’ll say it Nazi content. Not he said the wrong pronoun made up five minutes ago therefore he’s a Nazi. I mean, he could easily be the villain in an Indiana Jones movie Nazi. He’s both retweeting some of the most nakedly white nationalist content you’ve seen or clearly putting his thumb on the scale to boost said accounts. And of course this has attracted white nationalists to the site
Now, how many are actually bots or Boris from Russia pretending to be Joe from Kansas City? Probably a decent number. But this is still Musk putting his thumb on the scale.
I’m genuinely confused about this and why this is not part of your answer. Spend five minutes seeing what he puts out and you’ll know immediately he did the Heil Hitler salute in January 2025.
It was a Q&A answer, not an essay on Musk. Matt said Twitter is bad because Musk is bad, and that was a pretty good answer. Is your position that every mention of Musk or Twitter must also include, "And also Elon Musk boosts white nationalists and Nazis"?
This business of "You did something wrong by not mentioning this fact," is one of the worst things on the internet. You shouldn't do it to people, Matt or anyone else.
You seem to be reading more negativity into the comment than is there.
But to me, Matt’s comments did sound like they were trying to describe the central problem of Musk’s control of Twitter just being himself as a poor takester, and if Colin is right that Musk is not just promoting bad thinkers who agree with him, but white nationalists and Nazis, then that seems like it would have been worth including in Matt’s original comments.
To turn my MattY hat on. I'm making the same argument Matt makes when he argues with people about why homelessness exists and why it's so widespread in places like California. Someone will come out and say it's because the US and California itself doesn't provide enough mental health services or health care and this is why homelessness is out of control. The issue isn't that the person in question making this argument is necessarily wrong. It's that if you're holding up lack of mental health services as you're go to reason why homelessness is a problem, you're making a fundamental category error as to what makes California's homelessness so bad.
Musk being a Nazi and/or boosting white nationalism is not some side aspect of of the platform. It is clearly a central aspect of what he is doing with the platform. His go to reason for why Musk is bad is he that he doesn’t admit he made a mistake about DOGE.
This would be like if a major news publication spent the the past two months covering mistakes made during the Iran war and not bringing up oil or the strait of Hormuz and instead focusing one the fact the administration seems too happy with itself it severely damaged the Iran navy. Like I think we would be well within our rights to criticize the coverage for not focusing enough on the actual biggest mistake as opposed to real mistakes that honestly are not nearly as important.
And part of the reason why DOGE was such a mess was that Musk had spent half a decade beforehand cooking his brain in conspiracy BS. He killed USAID because of weird videos he watched made by a neo-Nazi conspiracy theorist.
Oh I’ve done that recently (think you meant “following” which is what I switched to). Which is the only reason I still use the platform at all. Since Matt focused on the issue of Musk and specifically what was worst about Twitter I wanted to point out that he was leaving out what was truly worst.
But secondarily is just the spam fed by algorithms. Before I changed over to “following” my feed would get filled up with the absolute most random stuff if I looked at some tweet for more than a few seconds. Sort of funny but illustrative example. I do go on there to get a lot of sports coverage. And with the NFL draft coming, wanted to see what latest buzz was going around. Couple writers noted a few prospects that reminded them of a few guys from late 80s or early 90s. And I guess because I settled on those tweets too long my feed was flooded with the absolute most random football plays from that time period. Like not Super Bowl highlights like week 7 colts vs Dolphins that even the most die hard colts and dolphins fans forgot about long ago.
So yeah the other part is what algorithm they are using is making the platform kind of unusable and was that as much as the gross white nationalist stuff that drove me to the “following” function.
I think people who rail about greed and corporate profits probably secretly feel that shortages are preferable to high prices in terms of dealing with excess demand and short supply. Shortages are, in theory, “fairer” after all, and in some ways, since the marginal value of an hour of a poor person’s time is less than that of a wealthy person’s time, forcing everyone to wait in line for scarce goods is a kind of wealth redistribution. They seem to be smart enough not to say it out loud, though.
Or they'll put in another way "if there isn't enough to go around, we shouldn't just let the rich have it; it should be shared out more fairly" (or words to that effect). Which amounts, in practice, to government intervention and rationing.
Climate change messaging in the US is fundamentally silly because it’s all done in degrees Celsius, and Americans don’t have intuition that degrees C are twice the size of degrees F.
re AI communications - remember that consumer AI is not the real battle. The real revenue is B2B. Companies that incorporate AI into their products generate TONS of revenue. Also companies that change their business practices to leverage AI in their internal workflows - development teams being the first and most obvious example.
So the comms that might be scaring normies are actually comms meant to scare business leaders into action: you cannot ignore this, it is transformational, you need to make AI work for your business or you will be destroyed.
I wonder what happens when some company rents out "accountantbot" for $30,000 a year. It makes mistakes but corrects them as taught about them.
To make matters worse, it's cheaper in taxes to "hire" accountantbot for $30,000 than a human for $30,000. No FICA, and even if there's a nominal sales tax to be paid by the "employee" it's way less than an income tax.
There are some policies that offer themselves. Payroll taxes for humans need to be low, zero, or maybe even negative. And renting accountantbot needs to be taxed at a very high level. I'm not sure how easy it is to draw the line about what counts as a "knowledge worker" bot, but pretending one could then we'd need a very significant tax on them. 50% or 100%, even 200% isn't off the table.
Also, accountants make way more than $30,000 a year, more like at least $60,000 right out of college. I actually made $38,500 right out of college back in the year 2000 - inflation adjusted this is $73,000.
As an accountant, I always find it odd how a lot of people, usually creative/liberal arts types, assume I make like $15 an hour or work for H&R Block for nickels or something like that.
It's not the most lucrative job, but it's not terrible pay either.
Yes, my number was deliberately low compared to the current starter threshold.
I think AI can really devastate the entry-level workers. A lot cheaper even if not as good. Then in 10 years we don't have seniors who can internalize and explain what's going on.
Companies need to figure out a better way to “grow their own” mid level employees without starting them with grunt work. (Historically, some university departments are good at hiring assistant professors to grow their senior ranks, while other departments maintain their name by hiring only senior people.)
My guess is that an "accountantbot" could compete pretty well with the offshore accounting services that aren't necessarily great quality to begin with and largely used by places like startups, but the price point would need to be lower than $30k/year.
Better prompting / markdown files can improve e.g. claude code to do things the way you want, be better about fixes.
With more widespread adoption as accountants and people improving prompts/markdown I would absolutely expect it to be better about mistakes as taught about them - that seems to be how it works with the good coding agents already.
The idea that we can just write more and more rules to deal with every case was the AI dream of the 1980s in the expert systems boom. It worked to a point, but then stalled out, and a return to expert systems will have the same problem.
I'm thinking of tax returns for instance. Some large percentage of people have very simple returns with common examples, and some additional percentage have a slightly more complex return with similar common things.
If you invest in index funds you're probably getting a standard form from Morgan Stanley, or Vanguard, or Charles Schwab, or something.
So you've got a very large set of people with very similar data sets/info so I'd expect failures to be the kind of thing we can train on - you don't need to deal with every case - even having a "this is non-standard, refer to a licensed CPA" is useful.
Basically I think the difference is, this seems way faster to get your "expert system" up to 60% of cases, and not that hard to get it to 80% of cases, and then you stop and let the rest get handled by computer PLUS human.
Expert systems trying to do 100% of things had issues and they were expensive to spin up even to the 60% mark.
Tax accounting for individuals and small businesses is the easiest part of the field by far, already largely automated if someone is keeping even half-decent books.
Expert systems are great at what they do, which is bringing ordinary people's activities up to the level they could have with good advice.
They just don't push the bar forward in terms of high end skill, either replacing actual experts or creating superhuman experts.
These new systems have the advantage that they've absorbed all the "common sense" that is implicit in all the training text, so you don't need to spend forever writing each of those common sense rules in - you can just focus on the advice you'd actually give a human who doesn't know the specific activity you're doing.
> Better prompting / markdown files can improve e.g. claude code to do things the way you want, be better about fixes.
It _can_, sure. But it isn't guaranteed, and in practice, claude ignores the markdown files a lot of the time, especially as it gets closer to its context window filling up. I'm pretty convinced that giving markdown files to claude is just snake oil at this point.
Also "better prompting" means the *human* is learning, not the agent.
I get that public opinion in the UK is very anti-development, probably for deep-seated cultural reasons going back to William Blake if not Richard II (the play). But I'd think YIMBYism would be a good market niche for some aspiring British writer - even as a minority viewpoint it would be great for trolling and clickbait. Yet it seems to be an unoccupied niche. Does YIMBYism not exist at all over there?
The Economist has often derisively referred to the British voter as 'Hobbits' for their intense provincialism at the expense of the wider country. As long as the tea is served and the pension is paid, who care if the little hobbits can afford houses or afford to live.
It does! At least in terms of journalism - the rather on-the-nose "YIMBY Pod" is the obvious candidate: https://www.abundancepod.com/ There are also some pressure groups, including one called "YIMBY Alliance". The authors of the essay "The Housing Theory of Everything" are all British. But I don't think it (yet?) has the same prominence in The Discourse as the American version.
It certainly does - as well as a number of others mentioned Jonn Elledge has a fairly successful Substack and used to have "Build More Bloody Houses" as his banner picture.
He was at the New Statesman (as an editor and columnist) for some years and now has columns there and at the New World, both of which raise some YIMBY points. There are a number of other youngish (20s/30s) journalists around who are also YIMBY inclined.
He used to write a regular column at CapX called NIMBYwatch, which is now written by James Ball.
I'd also add that the Young Liberals (the youth wing of the Lib Dems) are very YIMBY and there are repeated battles between them and the older more NIMBY end of the party. Young Greens get into similar internal fights with their older members. Polling consistently shows a big age divide on teh NIMBY/YIMBY split.
[Young Labour probably would as well, but Labour is structured to prevent internal fights from going public]
What there isn't is a significant right-wing YIMBY tendency at all - libertarianism basically doesn't exist in the UK.
They don’t seem to have any influence over the political parties of the right, which is what I meant by “significant”. Never seem to hear any YIMBY talking-points from Conservative or Reform politicians (where you do hear them from all three of the centre-left/left parties, even if not necessarily from their leaderships).
It's striking that in all those Shakespeare history plays, I can't remember more than a pro forma reference to policies, even in broad terms like modernization vs. traditional elites. It was all character.
I mean historically they just colonized places. This is just pains from the end of the British Empire. They should go take over somewhere in Africa again
Elie Hassenfeld is heir to the Hasbro fortune (“Hasbro” = Hassenfeld Brothers). Which might not necessarily be the most exciting fortune in the world. But he’ll be alright.
Anyway Zohran Mamdani is systematically selling out, whether voluntarily or not, and I am absolutely loving it.
I'm not addicted to Magic: The Gathering, it's just a more win-win way for me to back GiveWell's philanthropic endeavours. You can save four children's lives for the price of just one Black Lotus!
Oh god, don't get me started on what Hasbro is doing to Magic: the Gathering. It could best be described as a systematic attempt to satisfy shareholders' taste for not just goose meat but specifically golden-egg-laying goose meat.
Magic is suffering from its own variation on the UK's triple lock problem in that no set is allowed to ever be weaker in power level than the previous set while the set release schedule has also been absurdly accelerated, meaning that testing and quality control has gone out the window.
The Limited environment (which is all I care about) still seems pretty good - at least for large sets, just a problem of "too much" - I don't need 7 sets a year.
Yeah they’re getting better at making sets consistently fun to draft at the cost of taking a huge steaming dump all over every non-Commander constructed format.
Structural incentives always encourage TCGs to release more faster and creep power harder, and I think it's inevitable to hit rough patches like this sooner or later (and 30+ years is still a fantastic run)...but, man, the Extended Universe sets. I'm cool with closely-allied things like LOTR or FF, and niche things like Fallout can be alright too. Marvel and TMNT can fuck right off though. Nor does Avatar The Last Ringbearer really belong, although that's disappointing in a different way. Wouldn't matter so much if the core sets and plot were humming along nicely, but we completely lost the thread sometime around Brothers' War (if not earlier) and no signs of getting back on track. And I hate that you can't just ignore those sets either, cause they nevertheless contain some incredibly powerful stuff that warps metas both overall and locally.
Don't get me wrong, it's still fun to play when I can and I'm not in a rush to sell my collection or whatever. But I can't justify devoting most of my mindshare anymore to an obsession that seems obsessed with eating its own corn seed. When sets come and go and I literally don't feel like I have time to analyze them properly before the next spoilers are already up - What Are We Doing Here?
I started right after Fourth Edition (late by many standards, ancient by current ones) and looked into it a bit last year for nostalgia reasons and JFC WTF pure madness.
Along with changes of Twitter/X, can we talk about the recent changes to the Substack App? They got rid of the feature where you can see all your subscriptions (free and paid) and click on them to navigate to them (the website never had this). Now you either have to Search or hope that the newsletter of interest is in your feed. The app was already problematic before. Old articles are hard to search for and you basically need to either search key words or endlessly scroll. You can’t “Find on this Page” within the app. There’s no personal comment history.
Apps like Twitter, Substack, and Reddit are all basically braindead simple from a technical perspective (now that cloud software exists - it was really hard two decades ago). So the only thing that actually matters is good product management and design. These are difficult disciplines and I have serious respect for people who can do them well.
Thanks. I thought it was just me. Zoom has also changed things, and it is making me nuts. I often wonder what the goal is of companies that make these changes that may lead to very slight improvements, but lots of annoyance and tech exhaustion.
The website does have a way of seeing all your subscriptions: go to settings and scroll down to the subscriptions list.
Also, my version of the app, I just tap on my picture on the top-right and it brings up exactly that list of subscriptions (paid first, alphabetical order). Is this an iOS/Android thing?
It used to be like this! They seem to have removed this feature at least on iOS. And they want people to use it more like Instagram or Facebook with a feed, where new posts are more like stories.
The claim that the structural changes to Twitter have been good is insane. The big structural changes have been: paid accounts which get priority in the algorithm and in replies, paying people for engagement, turning the algorithm to focus on a uniform kind of viral engagement bait, allowing antisemitism and racism, losing to spammers so that DMs are now a cesspool, driving off advertisers so that ads are now insane people and scams, and making himself unavoidable on the platform. Which of those are good?
I disagree. The rare times that I tweet have led me to being followed by the most amazingly gorgeous women on the planet who are dazzled by my anodyne and cliched takes on the politics of the day. Who knew that was what sexy women are most turned on by?
Facebook is a lot better if you switch to your Friends feed. But that isn’t the default when you’re on your phone and they are very good at sucking you in with that first screen. Trying to be more mindful each time I open the app, but it’s challenging.
The reason that Le Pen and Meloni are seen as fascist adjacent is that their parties are direct heirs of parties which were explicitly fascistic in the 1970s and later. It is bad but also not particularly relevant today, they have changed their policies and ideals.
The difficulty for an American consuming US media coverage on this stuff is that the coverage of AfD and Meloni is basically the same. If Le Pen/FN comes to power in France, what predictions should we make about the policies her government should pursue? What should we predict if Vox gets power in Spain?
There is a lot of hypocrisy there, the far left contains lots of fanatical antisemites and actual members of the Stasi but doesn't have the same firewall.
Indeed both parties have direct historical links to the Pétain and Mussolini regimes respectively. Meloni's party even has some Mussolini descendants in it.
The challenge is that if you move from an explicitly fascistic position to a still right-wing (but now democratic) position, you've not made all that many policy changes; you're primarily changing your position on the legitimacy of democracy (ie on whether you'll accept being voted out and on whether you'll accept the constitutional limits on your power). It's only possible to test whether that is true or not if you get into power - will you use it within limits, will you relinquish it? It's understandable that opponents are genuinely concerned about whether you're really changed or just started lying about this.
Yup. Once you have that label it's very hard to get rid of it, no matter how much you moderate. Especially if the denunciations are not that strong and have an implied wink for supporters that come from that side. Something Americans should be familiar with ("very fine people on both sides").
I agree that Giorgia Meloni has generally not been that bad, but she endorsed Orban in Hungary. Hard to beat the fascist allegations when you’re a right winger who is supporting democratic backsliding in an EU country.
I find it really weird that it wasn't mentioned that Daniela Amodei is Dario Amodei's sister. (I had to Google that because I was wondering whether she was Dario's wife and the stuff about Holden Karnofsky had just been accidentally inserted in between.)
I agree that Musk is a pretty bad tweeter, and for me, it's pretty easy for me to ignore him and his calvacade of accounts that he likes. If his preferneces were cut out of the equation, not much would change for me.
I do have to disagree in that one structural change he's done has been very bad, and that is that he has made the search function damn near worthless, and it's because of turning the verification system into a payola scheme. As smug as it may sound, some people have more worthwhile things to say than others, and the legacy verification system accomplished this pretty well. I could add a filter:verified parameter to filter out the chaff in a search before. But now, no more--I now have to take a shot at the Latest results instead to find tweets that are reliable.
It's definitely bonkers that just using a plain Google search even today with all of Google's problems is often a more effective way to search for social media content than it is to use the social media site's own search function.
And this even with Google search getting worse than it used to be. When I ask a slightly obscure question, it tends to substitute more popular questions instead, and answer those.
For a concrete example, I wanted to test the claim that the largest centipedes known to have ever lived are still around today. I tried searching Google for largest fossil centipede, but the results were all for Arthropleura, which was not a centipede. Try "chilopod" instead of centipede, try more specific subtypes like Scolopendromorpha, try "evolution of chilopoda," etc., and it still "intelligently" detects that I am asking about myriapods and returns me to the more commonly discussed ones. Eventually I found scientific papers that described actual fossil centipedes, which are indeed small, but I was surprised by how advanced the search engine's artificial stupidity has become.
Artificial stupidity is not the lack of artificial intelligence, but a form of misalignment where it resists correctly understanding what you want.
All searches suddenly stopped you being able to search for a specific date range. I would be really curious to know why, it was a useful tool. Facebook and Google deliberately took away the capability.
For some reason, in the past couple of years, when I try to search for news articles I know I read between about 2005-2015, Google has gotten a lot worse at retrieving those articles. It used to be decent at retrieving older news articles.
Yes, there are a lot of web-site related decisions that I just kind of throw my hands up at because, rationally, the decisions must be informed by metrics in some way to boost revenue, but heck if I can figure out how!
I do ML in my day job and was working toward an advanced degree in the field of natural language processing when the now-famous Attention Is All You Need paper released. It really was transformative for my field, and was recognized as such at the time. But I don’t think even most smart and informed people working in ML anticipated that this model architecture would produce the capabilities we see today in less than 10 years’ time. I, along with many people in the field I talk to, continue to be astounded by what these models can do, even with their limitations.
That is to say, I am skeptical there is much correlation between the depth of one’s understanding of the linear algebra techniques that undergird the transformer architecture and one’s ability to make accurate predictions about what impacts these algorithms will have on society.
We shouldn’t fall into the trap of assuming the nerds, myself included, have any secret wisdom about the future. As a group, we were wrong about the second-order effects of the internet, and I suspect those predictions were made with much more earnestness than the pitches the AI companies are making.
My understanding, as someone who has been adjacent to the AI safety communities since the late 2000s, is that a lot of these people were already convinced by Moore’s Law that significant AI was coming in the next couple decades, and were thinking about the kinds of issues around disempowerment and existential risk for that reason. Some of them locked in the idea that AI would be basically a Turing computer, and never fully adapted to the neural net and then LLM versions we got in particular, but some then just saw these technologies as the medium through which their worries might happen.
Two notes (which are totally compatible with your takes):
The prevailing opinion in the late 2000’s was that we were still several decades away from real progress (“by 2100” was a popular guess, but the other side of that was a live option too). Basically nobody (except Ray Kurzweil and Shane Legg, I guess) thought the 2020’s would be the big decade for things starting to work.
We don’t have superintelligence or anything at the moment, so it’s not that the old opinions are definitively refuted, but clearly shorter timelines are a million times more common now than twenty, ten, or even five years ago.
A lot of the discourse also started in a world where we had no idea how to get systems to do things they weren’t explicitly programmed to do, and a worry was that we’d stumble onto the right (wrong) one, and in so doing create Skynet. We now at least know that it’s possible to get emergent programs that follow our instructions and have stable-ish dispositions. This is very good news, even if it’s not the whole ballgame.
Further reading on that subject, which says it better than I can: https://www.verysane.ai/p/alignment-is-proven-to-be-tractable (the author has another, more recent anti-Yudkowsky article that may also be of interest).
100% which might be a reason to suspect that something else might be driving those worries.
A very similar thing happened after the development of nuclear weapons - nuclear scientists were suddenly held up as policy advisers, even though understanding the physics of nuclear fission doesn’t translate incredibly well to actually designing policy around them.
There’s also a less-inflammatory version (because the models are harder to anthropomorphize) going on with driverless cars. Are driverless cars going to lead to more VMT? Are people going to own a car in 2045 or just hail robotaxis? I don’t know, and I bet the engineers at Waymo don’t know either! This is all to say that it’s really tempting to think/hope technical experts have the ability to predict policy implications, but usually they don’t.
All true, but your *average technical person working on a particular issue* knows a lot more about that issue and has probably thought more about related policy implications than a layman or even a policy person.
i.e. Dario Amodei has a much better developed viewpoint on AI and its implications than the vast majority of policy people working in Washington DC (especially before the last ~2 years when AI policy became in vogue).
The future is a radically high dimension space.
A butterfly flaps its wings in Beijing, Skynet launches the nukes because it misclassified the butterfly as a hostile drone.
I have the exact inverse take, which is that anytime I hear someone saying that AI research scientists predict massive social disruption, job loss, or even post-materialist utopia, all I think is that these linear algebra wizards have zero idea how society works and I put no additional weight on their predictions.
Fortune: "26-year-old AI researcher says learn these 9 skills or be replaced by AI ".
Me: That kid was born yesterday. What does he know about jobs, careers, or customers?
He can multiply matrices together like nobody's business though!
based
I agree with the second paragraph inasmuch as deep understanding of linear algebra doesn't seem like an especially important prerequisite for prediction-making--chiefly because the underlying architectures aren't necessarily the natural level of abstraction that people or should or do think about with respect to AI--but I think your last paragraph goes too far in its implications.
The extremely detailed AI-2027 timeline is proving spookily prescient, Amodei's predictions regarding "people will basically stop writing code' have been borne out. And here's Dylan Matthews on "The AI people have been right a lot."
https://dylanmatthews.substack.com/p/the-ai-people-have-been-right-a-lot
Hi. I work in the tech industry. No, 'people will basically stop writing code' has not only not been borne out, it's not even remotely close to true. Like, the exact opposite man
I am 'in tech' as a data scientist. I have stopped writing code and am better for it. I'm no dev, and I won't pretend my experience will generalize to proper devs, but the claim is at least remotely close to true.
A lot more people have been putting their API keys on GitHub, I will say that
Yep. Very excited to spend the next part (rest of?) my career cleaning up the explosion of codeslop we're about to get hit with.
Seems wildly off base. You just need to look at the Claude Code leak to see the entire thing is filled with manual memory and QA overrides. It's not at all hyperbolic to say that ~ 80% of production output is flowing through the human written code base.
"More Is Different" https://www.science.org/doi/10.1126/science.177.4047.393
Exactly. These AI doom lords have a lot in common with the string theory fantasists who predicted that science would end after the discovery of the GUT. Well, we're still here. There's still a lot to figure out including with how these AI tool will play out.
I just want to know how I can even evaluate the doomer claims. In three years if AI is a big business with useful applications but otherwise things are swimming along and we're all still alive are they more right or wrong? What about 5 years? 10? Am I supposed to treat them seriously forever regardless of what the actual literal capabilities of these systems are?
What if someone hooks Claude up to a nuclear arsenal and it accidentally launches everything, were the emergent super intelligence people correct or were we just fucking morons?
What's GUT?
"Grand Unified Theory" of physics. It means any theory that "unifies" three of the four fundamental forces of the Standard Model. I'm not a physicist and can't explain exactly what that means. It's still not achieved -- we still don't have a worked-out theory that fits everything that's known, makes new experimental predictions that are subsequently confirmed, etc.
I don't understand string theory either, but I thought that was an attempt at a "Theory of Everything" that also includes gravity. Gravity in general relativity is a curvature of space (and time), but it's very hard to make a quantum version of that theory without producing paradoxes, infinities etc. String theory is, iiuc, a broad class of approaches to quantum gravity, but with important problems for actually confirming any theory --e.g. there are many possible versions of string theory, and their distinctive predictions are infeasible to test in experiments we can actually build on a planet.
Also, if you happen to be into science lessons in the form of a cappella song parodies (performed by one man with lots of editing), there's this classic video. Again I have only a limited understanding, but it can give a non-physicist some sense of what problem string theory aims to solve. https://www.youtube.com/watch?v=2rjbtsX7twc
String theory and the GUT effort so far has failed. PW Anderson predicted this wouldn't work a long time ago in the linked paper above.
Not only that the stories the nerds tell about AI risk are exactly what you would *expect* them to say for other reasons. They fundamentally center intelligence as the most important thing and then tell a sci-fi story which puts them in the role of the hero.
I'd be much more persuaded [1] if the story they had about AI risk made it seem boring or centered some other kind of actor -- for instance you could imagine a concern focused on how humans will react/interact with AI that centered social scientists or one that focused on geopolitical or economic risks.
---
1: Also the alignment arguments are just bad. They fundamentally assume that AIs will somehow maximize some really simple moral function when the very thing that makes modern AIs powerful is their ability to learn quite complex context dependent heuristics like language.
This is like, my favorite response ever to transformers
https://bactra.org/notebooks/nn-attention-and-transformers.html
The math isn’t hard they just keep calling “likes” “florps.” I guess it was meaningfully more difficult to raise money for “fancy kernel regression.”
I like a good grumpy statistician as much as the next guy, but I have a hard time being upset that a different culture came up with different names for the same thing. One time I asked a stats grad student what backprop was and he said "the chain rule." And he's right! NNs are a bunch of stacked logistic regressions (or hinge regressions, whatever "non-linearity" aka link function they're using nowadays). Bias is an intercept term. At least they didn't try to redo "regularization."
The statistics community has never come up with a term as cool as "support vector machine," even though it's just shy of word salad.
I'm very charmed by how clearly obviously it is that NNs were thought up by CS people. They don't like math, they like graphs. An NN is an abstraction of math that lets you focus on the graph and think about it like a software diagram. They don't care about p-values or overfitting or whatever because n = 1 billion and they can just look at OOS validation and regularize and it's fine, shut up about it.
Further in this, it is hard to get upset about naming being difficult across fields, when naming after people remains the norm.
One of my favorite lists is the one of "things unexpectedly named after people." French drains being the latest I learned on there. :D
CS is full of things that are a ton easier for me to understand when I realize a possessive was dropped. Bloom's filter and Shell's sort being my favorite examples.
Nerds don’t have a secret wisdom, but they understand AI enough to be able to think about its implications. Most laypeople have no idea what AI is going to be like, and cannot begin to think about its implications in any meaningful way.
FWIW, I don’t think you need to be deep into ML algos to have the necessary understanding of AI. But you do need to have at least a somewhat technical mind. Lots of (non data) journalists, lawyer, and politicians types have no idea what the future holds.
Thing is, most laypeople have no idea what the coming AI would be able to do and to what extent we would be able to control it. They are thus completlely unable to think about its impact on society or the dangers it poses. Nerds don’t have a secret
The center-left media refuses to admit that in much of Europe, immigrants commit crimes at meaningfully higher rates than natives. A lot of this is a composition effect — Europeans are heavily socialized and well-behaved, so admitting almost any other group will mechanically push crime rates up. Germany's 2023 police data makes the point cleanly: non-Germans are about 15% of the population but 41% of violent crime suspects, and asylum seekers specifically were roughly 2% of the population and 15% of rape and sexual assault suspects in 2017 (the pattern has held since). Yes, young-male demographics and socioeconomic status explain part of the gap — but only part.
The problem is European elites don't want to admit this. Rather than an honest discussion of the tradeoffs — more disorder versus a younger, more dynamic workforce and a shored-up welfare state — voters get force-fed pieties about equality that don't match what they see on the street. Add the fact that subsidizing unproductive new arrivals strains the welfare state rather than rescuing it, and the electoral success of the European right makes sense.
As one of the European SlowBorers that lives in the US, I think that this is unfortunately correct. The immigration debate gets very confused between the US and the EU, and you have American right-wingers (like Vance) who think that immigrants in the US don’t assimilate, and EU officials who think that immigrants in the EU do assimilate. I think that American media also doesn’t understand basic things about Europe (lack of understanding may be a charitable interpretation), so you get a lot of confusion here. I was pretty worried about Meloni’s election but she’s been basically fine, and I wouldn’t trust any media source that continues to label her as “far-right”.
PS: I see some commenter below who says that if you wrote something similar about the US, it would be considered racist. I agree! The immigrant experience in the US is VERY different from the one in the EU.
The weird think about the US immigration discourse is concern about Latin American immigrants assimilating. If there is one thing that the last century has proven, it's that we can absorb a huge number of Latin American immigrants and assimilate them to US norms. Grandpa comes to hang drywall and doesn't speak much English, Dad is fluent in both and speaks English with no accent and retains a lot of the culture, and his son is a generic American with a hispanic name who still speaks a bit of Spanish to talk to his grandparents but who is way more comfortable in English, prefers American football to soccer and pizza to pupusas or whatever, etc.
One of my favorite "sort of assimilated" moments. I was on my way to visit my parents, and stopped at this ginormous indoor flea market in rural Tennessee. (North of Cookeville, to be precise).
The interior was typical - booths around the edges, and in the center, selling everything from cheap knives to household bric-a-brac. And in the most prominent location - front and center from the main door -- was a double-sized booth selling fuzzy blankets. On the display rack, there were two blankets: Our Lady of Guadalupe, and a Confederate battle flag.
Very relevant The Argument post:
https://www.theargumentmag.com/p/why-america-is-so-much-better-than
I am sure there are other differences but it seems like preventing immigrants from working is a big one.
This was a great read. Thank you for sharing it here!
Isn't that just because Europe's immigrants are newer and haven't had time to assimilate?
Comparing European immigrants who arrived 7 year ago to Hispanic immigrants that have been in America for 3 generations seems like an apples to apples comparison. Just look at all the American angst over, say, recent Somali immigrants. Very few Americans outside of places like Slow Boring are blase and assume they will assimilate in 3 generations.
Likewise, Europe used to have tons of immigration and nobody is now complaining about, say, Italian minorities in Croatia or French refugees from 1848 who went to Spain, not assimilating. It's just Europe did all their assimilating 100 years ago and hasn't had much opportunity since. I don't think that means they are somehow unable to do it, given their proven track record of creating national unity in Italy, Germany, etc in the not too distant past amongst people who did not really see themselves as a single people.
It just takes time, which Europe has not had .... Yet.
It's kind of funny that the idea of taking in immigrants, but being selective in the process is just off the table. Nope, has to be all or nothing.
I think this is the activist vs normie split. For activists, a line of "ningun ser humano es ilegal" works well, as does a "you will not replace us" rejection of all immigrants. But if you're trying to actually govern well, you want to allow some immigration but control it and be a bit selective about who comes here wrt whether or not they're likely to be net contributors to the society.
If this is a comment about Europe, I think that we get the best immigrants that are available to us!
Either the US or any European country is capable of having selective immigration. The US does a better job of being selective but then has done a bad job of policing illegal immigration. Thankfully, our unselected mass has come from a part of the world (Latin America) that's less culturally dissimilar to us than the Middle East is to Europe. So we could be doing a BETTER job, but things could be worse all things considered.
I don’t think that the US necessarily does a better job of being selective. If you are extremely smart, you’ll go to MIT not TU München. If you have a groundbreaking idea and want to start a company, you’ll go to the US not France. Given who actually is interested in coming to Europe, I think we’re doing a decent job with being selective.
I don't think Europe's main problem is in selection effects, though it's definitely a contributing factor.
It's that European nations, except sort of the UK and Spain (at least as pertains to Latin American immigrants) neither require nor even really *allow* immigrants to assimilate.
In terms of culture, economics, and policy, a society really couldn't craft a system where the barriers to assimilation are higher and the payoff lower than in contemporary Europe if it deliberately tried.
Whereas in the US basically every aspect of our society pushes immigrants to assimilate quickly and be productive. If you don't you'll live a shitty life and your kids will knuckle under and do it anyway, and once you do you're a real, genuine American in the eyes of basically everyone except the Twitter natcon rightists.
Are Latin Americans in Spain immigrants? :P I know what you’re talking about, but I consider it borderline immigration.
We agree on one of the main problems with integration of immigrants in Europe. We are nation states (what I sometimes see called “ethnostates” in the US), not nations of immigrants. We had many wars to make sure that our countries are ethnically homogeneous! A lot of people were killed or forced to flee so that everything aligns with the lines on the map.
The other problem is that after living in the US for a few years, I’ve come to believe that a generous welfare state is incompatible with a generous immigration policy. I think that all the immigrants in Europe that don’t work are basically being rational, and I believe that integration is hard without work.
The only immigrant groups to the US with below average education levels are either from countries nearby (Mexico, Central America, the Caribbean), or from countries where most of the diaspora in the US are refugees (Somalia, Afghanistan, etc.). Every other country is sending disproportionately educated people to the US (not that education is the only thing that matters, but it's a decent proxy).
That's in large part the choice of Western Europeans (says the Eastern European who studied and lives in the US, along with many others of my type). Most Asians i speak to in the US also did not feel welcome in the EU.
This has actually changed somewhat in the past 20 years. But a) the EU has been unwelcoming for a long time, and b) it's not even clear they are more welcoming today.
I watch a lot of German police procedurals, mostly "Tatort" and various "SOKO" series, and at least during the late teens and early 2020s it was all immigrants all the time. Seriously. Now, if "Law & Order" isn't a documentary, then neither is "Tatort." But it certainly tells you about the concerns of the TV-watching public in Germany.
Okay but in the US immigrants commit (non-immigration-related) crimes at lower rates than natives and we have the same right-wing movement…
Do we? Our right wing is less interested in protecting the welfare state.
I think the rightwing argument is that this is a statistical illusion, because immigrants (by definition) have been here a shorter period of time. I.e. if you compare an immigrant who's been here for 5 years to a native who's been here for 40, of course you're going to see a higher crime rate for the latter- they've had more time to commit crimes. From what I've heard, if you compare for 'years you've had to commit crimes on US soil' you get different results
Considering how 1) a lot of the immigrants the Right is worried about are younger and male and 2) people tend to commit crime (especially violent crime) in that same age bracket, that's kind of an odd counterargument. If anything, you would expect a disproportionately young and male demographic to have higher crime rates without any statistical massaging.
I can't speak to all research on this but the papers I've read have normalized for length of time in-country...
[citation needed]
I don't love that you could write the same couple grafs about America with only a couple word substitutions, except then it would rightfully be called rather racist, while still probably getting lots of upboats because something something hawkish commentariat.
Except we have a more violent baseline and less violent base of immigrants than Europe, so it would just be racist here rather than expressing an unfortunate truth
One difference is that the US has a lot of crime and some high-crime demographics, so immigrants who have the same propensity to commit crimes might be worse than the natives in Germany or Sweden, but better than the natives in the US.
A rising tide lifts upboats.
I seem to remember an article a while back where they interviewed a bunch of people at an AfD rally about why they were supporting the party. It seemed like most were lifelong CSU/CDU voters who just wanted their party to do *something* about immigration. It's weird to me that the CSU wouldn't adopt more moderate positions or at least acknowledge it's a discussion worth having.
Brandon Johnson’s problems go well beyond policy. He’s one of the most incompetent executives in any large organization on the planet.
He has a dysfunctional relationship with the much more popular Governor Pritzker and the legislature— and his team sometimes doesn’t even advocate for their own positions.
He has been unable to pass his preferred revenue policies, losing a referendum on a real estate transfer tax and having to accept the city council’s budget which excluded a business head tax.
He has repeatedly been unable to get his preferred appointees in office in a timely way (like the Zoning Board chair being vacant until his a candidate he didn’t want was appointed this week), and at other times has fought with his own appointees. For example, the school board resigned en masse, and rather peculiarly refused to comment, when Johnson ordered them to fire the schools CEO whom he had allowed to stay on; even after appointing another slate he was still unable to get CPS to do the irresponsible borrowing he desired, and the new CEO (promoted from acting after a long delay) seems to be refusing to shut the schools down for May 1 as the teachers union and mayor want.
Reupping my Johnson piece form last year https://www.slowboring.com/p/progressives-need-to-reckon-with
You did such a good job of swatting away Freddie's usual aggressive top level comment that day.
Had some practice with it
As we all have...
He's blocked me, so I no longer see when he comments on Slow Boring. Life is good.
You missed out on a rigorous, informative discussion of British colonialism of India that somehow emerged from a total non-sequitur insult that Freddie leveled at Matt!
Oh well. No longer seeing Freddie's random insults on Slow Boring is worth something to me.
A problem with putting a militant teachers union's militant in charge with governing a city is they don't do the whole "balance various interests and concerns" well. This is part of the whole issue with the "Chicago School" of left wing politics (which Brandon is very much a product of). You're so focused on destroying the enemy/overthrowing the system* that you can't do normal governance things like appeal to the business community for investment or tell unions "No property taxes are not primarily about funding the bottomless pit that is your poorly run pension plan"
*there was real value in pointing out the problems with Daley the Elder's system, but that ended a long time ago and it's time to move on.
Agree with this, but as I commented on the original question: being mayor of Chicago is one of the worst jobs in politics because the city is functionally bankrupt. The city continually needs to raise taxes and cut services to pay for pension obligations and debt. That is extremely unpopular with voters. I don’t think any other major American cities have this problem.
At the same time, Chicago mayors have VASTLY more power than, say, NYC's mayor because of the lack of a city charter. City Council here is not funded at a level that would allow them to do their own fiscal analyses. This is the root cause of a lot of our problems. A couple of good primers on this topic:
https://www.thelastward.org/p/with-no-city-constitution-chicagos
https://www.thelastward.org/p/brandon-johnson-on-chicagos-mayor
The pension debt is crippling, sure, but Johnson has done nothing except make it worse.
I agree that Johnson has been bad, but the unfunded pension obligations are so large that I don’t think Chicago has any feasible path to paying them regardless of choices. The city needs a state constitutional amendment allowing them to reduce benefits and/or a bailout from the state/federal government.
A federal bailout of Chicago would not be a good idea. The backlash would likely lead to a Republican Congress for a generation.
BESHAR TO CITY: DROP DEAD
His fight for May 1st protest holiday for teachers is crazy. I don't see how he gets reelected. Does he even run?
Why would public servants need a protest day? And is it really a protest day if there's an official holiday?
I think so. He seems to have the progressive lane to himself if he chooses to run, so would have good odds to make a runoff.
I imagine he loses, but that may be the plan on the part of the Chicago movement left—the mayorship is quite the poisoned chalice given the revenue situation, and Johnson has does his best to dig a deeper fiscal hole. So they need someone not associated with the left to make the tax increases and cuts, without their fingerprints on it. And they can blame the failures while in power on Johnson himself and his cronies like Jason Lee rather than the blind spots of the larger movement.
I think we are beyond tax increases and cuts. I don't see any way out of Chicago's fiscal hole except state-facilitated bankruptcy. It's a question of when, not if.
The city still has options for now although I wouldn’t push back on your assessment over a longer horizon, unless city governance quality gets radically higher than it’s been in decades. The city will probably need to go through a round or two of austerity before the state allows restructuring—and the state isn’t in good shape either. But I don’t fully understand the political landscape here, the contours of any planning on restructuring would be kept very quiet.
CPS is in much more immediate difficulty and the state might need to take some control in exchange for a bailout.
Does IL have any slack to conduct a bailout?
Its own pension system is about as grossly underfunded as Chicago's, it has very high tax burdens and net outmigration, has serious shortfalls of infrastructure and service provision...
Sure, for now. Illinois is in better shape than Chicago, which is in better shape than CPS. The state just bailed out Chicagoland transit in exchange for less mayoral control.
CPS’ credit rating is already terrible and they have limited revenue options on their own. The city is in trouble on its own and the TIF surpluses are finite.
He has zero chance to make the runoff. His support base is gone, even the CTU is wavering.
And even beyond anything he's actually done or not done, he's a singularly incapable communicator to a degree that's hard to believe for this level of politics. And rhetoric is typically an area where left-wing figures have an advantage over their more pragmatic competitors, it's the opposite in Johnson's case.
There's a lot to debate about *the idea* of Brandon Johnson, that *idea* won the mayoral election. But the story of his mayoralty is that the person himself is a far, far cry from that idea. In that way he's just not a comparable figure to other progressive mayors across the country, it's apples and oranges.
I feel like this is a much better answer than the one Matt gave. The question starts by asking about Johnson and MY barely mentions him in the answer, this feels like a trend to me.
A quick AI related anecdote. Last week I was visiting a collaborator to try and finish off a draft of a math paper (ca. 25 pages). (It is some fairly computational stuff in group ring theory applied to low dim topology if that means anything to anyone). At one point we were wondering where a general argument we have breaks down in an exceptional case (something that works for all Baumslag-Solitar groups except for the solvable ones ... i'm writing allthis to convey the message that it is a narrow and specialized situation). We knew based on an example that the argument doesn't work in the exceptional case but couldn't spot where the general proof (which is spread out throughout the paper) breaks down the way we had written it. So, collaborator fed the draft to several chatgpt instances and one of them (after 10 minutes or so) both pointed out the counterexample we knew and also pointed to a specific sentence in a lemma where the assumption that the BS group wasn't one of the exceptions was implicitly used. (It was pretty nonobvious, the sentence was vague and nothing about the context suggested that special assumptions were used.)
It is the sort of thing we probably would have figured out an hour later, and a smart and dilligent grad student might pick up on after a week if they had to referee such a paper from scratch (a thankless task).
Now, it is true that the other instances gave counterexamples but didn't spot where the proof breaks down, and that this level of usefulness was somewhat unusual among the other things we tried, but it is clearly not nothing, or 'hype'or whatever.
I assume your parents no longer ask you what you do at work.
Do you talk to your parents sbout your job?
When they were alive, my parents loved nothing more than discussing my work in military logistics. "Nothing more" in the sense of "anything but."
I think I misunderstood where your earlier comment was coming from. Military logistics sounds like something i would quiz someone about at a dinner party, no doubt. But I imagine that if you were to talk about it with the same laypeople repeatedly, it would just degenerate to the same conversation over and over. Am i getting that right? So in practice, if you talk to family about work it is about the human parts, both positive and negative.
Are there people who don't? Like, what else are you supposed to talk to them about?
Family, food, travel, health, sport, money, even politics ... it is rude to talk about work in mixed company, just as it is rude to speak in a language only a proper subset is comfortable with.
I strongly disagree. Good day, sir!
I’m mystified by this exchange, but happy to let it go. Cheers!
Don't tell sharty.
I find it really interesting that advanced math is the one area where I hear people using ChatGPT specifically in effective ways. Everywhere else it seems to be Claude.
I use Claude for most other things, but ChatGPT for some reason continues to trounce it on research level math. I recently fed ChatGPT 5.4 in the chat window and Opus 4.7 max effort in Claude Code the same question - ChatGPT gave a decent if speculative answer and flagged where the uncertainties were, and Claude made startling elementary errors that made its answer worthless. No contest.
(Grew up with Celsius and switched to Fahrenheit only after moving to US for grad school, so have a lot of experience with both)
Fahrenheit is just superior to Celsius for everyday use (at least in non-arctic climates), and this is a petty hill I will happily die on
1) it's a true 0 to 100 scale. 0F is very cold, 100F is very hot, and this is the range that covers most people's everyday experience.
2) it almost entirely eliminates the conversation: "-what's the weather like?" "it's 5 degrees." "plus or minus?" which happened ALL THE TIME when i lived in Northern Europe (where its around 0C for a significant proportion of the year). With C the only way to avoid it is to always specify plus or minus but that is an extra word and a lot of people chose not to say it. With F you can have the same problem, but in the range that is much more rare, and, frankly, difference between -5F and 5F is much less important.
3) 1 degree of C is just a bit too much. As a result, a lot of the time you need to think of it in whole degrees even for everyday stuff, and certainly can't just round to nearest 10 (difference between 5 and 14 or 11 and 19 is massive). With F, most of the time the first digit of the temperature gives all you all the info you need to know.
Basically, it is almost perfectly tuned for the temperatures that people actually use. And guess what, scientists, always switch to the temperature scale that is tuned for their specific need, whether it is Celsius, Kelvin or electronvolt.
Fahrenheit got really lucky with that! He may have been thinking about the 0 and 100 thing on some level when he decided human body temperature would be 100 and the coldest he could achieve with brine would be 0, but no one understood yet how ordinary people would be talking about weather in terms of temperatures, and that these ten degree increments really were basically right for planning clothing, or that 1 degree would be pretty close to the just noticeable difference.
There is a long rant from Jon Lovett (Pod Save guy) about how imperial measurements are superior to describing the world as it is experienced by actual people. Along these lines.
And it makes sense - the general conservative point is right. Traditional systems developed by people for a purpose usually work well for that purpose. The good feature of the SI is that it enables equally easy talk about all measurements at all scales. But if there is a particular scale you are constantly talking about, it’s better to use unites adapted to that. Don’t tell me street addresses in latitude and longitude - tell me with reference to the local streets! Even astronomers use the “astronomical unit” rather than megameters or gigameters when talking about distances inside the solar system.
Ok, but can we agree that traditional unit conversions are really dumb? Why 12 inches to a foot? Why 16 ounces to a pound? Who the heck multiplies or divides in their head by 16?
In metric, it's simple - everything is a multiple of 10, super easy to convert. 1 meter is 100 centimeters is 1000 millimeters is 0.001 of a kilometer, easy peasy.
Oh. I know the inches one. Base-12 is more easily divisible for construction than 10.
Right, and this is the one big flaw of the metric system, not being able to divide evenly by 3 and 4, those are very common divisions in everyday life. If only we had 6 fingers on both hands instead of 5...
The reason for these weird conversions is that you shouldn’t be doing these conversions! It’s no weirder than 9.461x10^15 to convert meters to light years!
The point historically was that you measure things like precious metals or oils in ounces, and drinking liquids like water or milk in cups, just as you measure rooms in feet and travel distances in miles.
Metric is great when you’re measuring things across scales without any distinctive scale as special (like when you’re converting recipes to feed an army). But if you’re always working at one scale, you want the units for that scale, even if they’re not a nice power of ten times units for another scale.
Especially when one "foot" was literally the tradesman's foot and the unciae was the width of their thumb.
I see people talking about conversions as the reason metric is way better, but I feel like this overstates how often most people need to do conversions? I'm fully on the side of metric for mass (especially cooking), but for temperature I'm not calculating energy and pressure regularly so I don't need to convert anything, and for distance the things I measure in feet and miles are fundamentally so different that there's no actual conversion happening. I'm six feet tall but I'm not going for a 30,000 foot run.
Fair point, but it depends on what you're doing. As a molecular biologist, you frequently have to convert between pico, nano, micro, and milli (grams and molar).
Given that really bad things happen to the human body above his estimation of 100 degrees that can be aggravated by hot climate, he might not have been *really* lucky, even with some luck.
If Trump switched us to Kelvin instead then that would totally avoid the "above or below zero" issue.
"What's the temperature out there?"
"294"
"Nice day!"
I always found it hilarious when my fellow students and postdocs would seamlessly switch between Celsius (for experiments) and Fahrenheit (for the weather).
"All right, I'll put the E. coli in the 37-degree incubator and head out. The weather's really nice, it's up to 55 degrees already!"
Fahrenheit was created with human comfort in mind
In addition to this, measuring even big differences in human height with whole numbers of inches seems markedly superior to looking at decimal points.
Maybe it’s just because I am familiar with imperial, but someone being 6 ft 5 versus a 5 ft 7 person seems much easier to understand than 1.96 vs 1.70 meters.
The scales are too compressed.
I work in a more precision-oriented industry than many and I just don't buy it. It was fine when I was doing fabrication but in aerospace welder essentially use metric fractions of inches (thous). It's kind of ridiculous.
Yes, but that’s not human scale measurement like height.
I don't see any reason why dealing in thousandths of an inch is any different than metric in this case. Neither one is dealing with human relatable scales and neither one really gives neater or cleaner numbers to deal with. A .001 tolerance equates to either 0.0254mm or 25.4 um which is something you might see on a sealing bore. Sheet metal might be .030 which gives 0.762mm or 762 um.
Just seems superfluous given the metric system already exists. Still, I know what a .060 and .090 rod look like.
It seems that metric is superfluous given that imperial/EE exists. Metric came after and was only really useful to scientists when long and complicated calculations had to be performed by hand. In the modern world computers do all of that. Computers have eliminated the one major practical advantage of using metric but nothing has eliminated the usefulness of having units that are easily translated to human experience.
Would there be some huge gain if you were used to 1.6mm and 2.4 mm rods instead?
I work in a pretty precision industry too but man I love base-10 ft. Engineering-scale imperial is very user friendly
If it's -5, why wouldn't one say "minus five" or "5 below 0"? It's wrong to call it 5. If I say it's 5 degrees out, I never mean -5 degrees.
Cause it’s slower, and normal speech is not maximized to be unambiguous…When it’s -20, is perfectly reasonable to just say 20 (and people do) no confusion, so I guess people just extend it to much.
I grew up in frigid temps and have never heard anyone just elide the sign because it's too inconvenient to specify it.
-40 is the magical temp. Mercury melts.
Also where the Celsius and Fahrenheit scales cross, which is why songs, movies, TV shows, etc. almost always say "40 below" when mentioning cold temperatures -- it communicates "very cold" to media consumers under either scale without needing any translation.
Note I did not specify the scale.
This is an interesting position! I also started using F after coming to the US for grad school and you’re right - it’s very useful! I do miss g and kg and m and km though, and especially the easy conversions!
When I learned statistical mechanics and saw that actually what appears in all the equations is kT (Boltzman's constant times temperature in Kelvin), I realized that actually temperatures should always be specified in electron volts. I drew a scale on a piece of paper for easy conversions to/from Fahrenheit and Celsius.
And yes, actually the speed of light is 1.
I like having the constants in the equations, which you don't get with natural units.
Imperial for temp and cooking (I'm not weighing my flour, now everyone go rage about it). Metric for everything else.
Been building some stuff lately and the concept of an 11/64" drill bit is infuriating. Do better, America.
I also prefer F for human things and I am constantly annoyed that the thermostat for our reverse cycle air con can't do 1/2 degree steps. Sometimes I feel like I just need 1 degree F cooler/hotter but can't do it.
Regarding Twitter. I’ve been on there recently because there are people like you are on there.
I don’t know why you are leaving out the fact that Musk is clearly using the platform to boost virulently bigoted white nationalist and yes I’ll say it Nazi content. Not he said the wrong pronoun made up five minutes ago therefore he’s a Nazi. I mean, he could easily be the villain in an Indiana Jones movie Nazi. He’s both retweeting some of the most nakedly white nationalist content you’ve seen or clearly putting his thumb on the scale to boost said accounts. And of course this has attracted white nationalists to the site
Now, how many are actually bots or Boris from Russia pretending to be Joe from Kansas City? Probably a decent number. But this is still Musk putting his thumb on the scale.
I’m genuinely confused about this and why this is not part of your answer. Spend five minutes seeing what he puts out and you’ll know immediately he did the Heil Hitler salute in January 2025.
It was a Q&A answer, not an essay on Musk. Matt said Twitter is bad because Musk is bad, and that was a pretty good answer. Is your position that every mention of Musk or Twitter must also include, "And also Elon Musk boosts white nationalists and Nazis"?
This business of "You did something wrong by not mentioning this fact," is one of the worst things on the internet. You shouldn't do it to people, Matt or anyone else.
You seem to be reading more negativity into the comment than is there.
But to me, Matt’s comments did sound like they were trying to describe the central problem of Musk’s control of Twitter just being himself as a poor takester, and if Colin is right that Musk is not just promoting bad thinkers who agree with him, but white nationalists and Nazis, then that seems like it would have been worth including in Matt’s original comments.
Basically the gist of my point.
To turn my MattY hat on. I'm making the same argument Matt makes when he argues with people about why homelessness exists and why it's so widespread in places like California. Someone will come out and say it's because the US and California itself doesn't provide enough mental health services or health care and this is why homelessness is out of control. The issue isn't that the person in question making this argument is necessarily wrong. It's that if you're holding up lack of mental health services as you're go to reason why homelessness is a problem, you're making a fundamental category error as to what makes California's homelessness so bad.
“Have you renounced Stalin,” is an old trope. Ancient in internet years.
reject *and* denounce
Musk being a Nazi and/or boosting white nationalism is not some side aspect of of the platform. It is clearly a central aspect of what he is doing with the platform. His go to reason for why Musk is bad is he that he doesn’t admit he made a mistake about DOGE.
This would be like if a major news publication spent the the past two months covering mistakes made during the Iran war and not bringing up oil or the strait of Hormuz and instead focusing one the fact the administration seems too happy with itself it severely damaged the Iran navy. Like I think we would be well within our rights to criticize the coverage for not focusing enough on the actual biggest mistake as opposed to real mistakes that honestly are not nearly as important.
And part of the reason why DOGE was such a mess was that Musk had spent half a decade beforehand cooking his brain in conspiracy BS. He killed USAID because of weird videos he watched made by a neo-Nazi conspiracy theorist.
"Spent the past two months"? How about one paragraph in a news round-up? Would we be within our rights then?
It’s easy enough to avoid all that, just follow decent people and don’t use the “for you” filter.
Oh I’ve done that recently (think you meant “following” which is what I switched to). Which is the only reason I still use the platform at all. Since Matt focused on the issue of Musk and specifically what was worst about Twitter I wanted to point out that he was leaving out what was truly worst.
But secondarily is just the spam fed by algorithms. Before I changed over to “following” my feed would get filled up with the absolute most random stuff if I looked at some tweet for more than a few seconds. Sort of funny but illustrative example. I do go on there to get a lot of sports coverage. And with the NFL draft coming, wanted to see what latest buzz was going around. Couple writers noted a few prospects that reminded them of a few guys from late 80s or early 90s. And I guess because I settled on those tweets too long my feed was flooded with the absolute most random football plays from that time period. Like not Super Bowl highlights like week 7 colts vs Dolphins that even the most die hard colts and dolphins fans forgot about long ago.
So yeah the other part is what algorithm they are using is making the platform kind of unusable and was that as much as the gross white nationalist stuff that drove me to the “following” function.
I see the same sorts of ad issues, but it's even worse on Facebook - a platform I hate even more, but am forced to minimally use.
I think people who rail about greed and corporate profits probably secretly feel that shortages are preferable to high prices in terms of dealing with excess demand and short supply. Shortages are, in theory, “fairer” after all, and in some ways, since the marginal value of an hour of a poor person’s time is less than that of a wealthy person’s time, forcing everyone to wait in line for scarce goods is a kind of wealth redistribution. They seem to be smart enough not to say it out loud, though.
Or they'll put in another way "if there isn't enough to go around, we shouldn't just let the rich have it; it should be shared out more fairly" (or words to that effect). Which amounts, in practice, to government intervention and rationing.
I think they also look more positively at rationing of goods.
On the Celsius comment, one small take I have:
Climate change messaging in the US is fundamentally silly because it’s all done in degrees Celsius, and Americans don’t have intuition that degrees C are twice the size of degrees F.
I don't think that's the reason why Americans resist climate change policies.
re AI communications - remember that consumer AI is not the real battle. The real revenue is B2B. Companies that incorporate AI into their products generate TONS of revenue. Also companies that change their business practices to leverage AI in their internal workflows - development teams being the first and most obvious example.
So the comms that might be scaring normies are actually comms meant to scare business leaders into action: you cannot ignore this, it is transformational, you need to make AI work for your business or you will be destroyed.
I wonder what happens when some company rents out "accountantbot" for $30,000 a year. It makes mistakes but corrects them as taught about them.
To make matters worse, it's cheaper in taxes to "hire" accountantbot for $30,000 than a human for $30,000. No FICA, and even if there's a nominal sales tax to be paid by the "employee" it's way less than an income tax.
There are some policies that offer themselves. Payroll taxes for humans need to be low, zero, or maybe even negative. And renting accountantbot needs to be taxed at a very high level. I'm not sure how easy it is to draw the line about what counts as a "knowledge worker" bot, but pretending one could then we'd need a very significant tax on them. 50% or 100%, even 200% isn't off the table.
Also, accountants make way more than $30,000 a year, more like at least $60,000 right out of college. I actually made $38,500 right out of college back in the year 2000 - inflation adjusted this is $73,000.
As an accountant, I always find it odd how a lot of people, usually creative/liberal arts types, assume I make like $15 an hour or work for H&R Block for nickels or something like that.
It's not the most lucrative job, but it's not terrible pay either.
Yes, my number was deliberately low compared to the current starter threshold.
I think AI can really devastate the entry-level workers. A lot cheaper even if not as good. Then in 10 years we don't have seniors who can internalize and explain what's going on.
Companies need to figure out a better way to “grow their own” mid level employees without starting them with grunt work. (Historically, some university departments are good at hiring assistant professors to grow their senior ranks, while other departments maintain their name by hiring only senior people.)
My guess is that an "accountantbot" could compete pretty well with the offshore accounting services that aren't necessarily great quality to begin with and largely used by places like startups, but the price point would need to be lower than $30k/year.
> It makes mistakes but corrects them as taught about them.
This is not possible and will never be with current designs
Better prompting / markdown files can improve e.g. claude code to do things the way you want, be better about fixes.
With more widespread adoption as accountants and people improving prompts/markdown I would absolutely expect it to be better about mistakes as taught about them - that seems to be how it works with the good coding agents already.
The idea that we can just write more and more rules to deal with every case was the AI dream of the 1980s in the expert systems boom. It worked to a point, but then stalled out, and a return to expert systems will have the same problem.
That's true, but is this the same situation?
I'm thinking of tax returns for instance. Some large percentage of people have very simple returns with common examples, and some additional percentage have a slightly more complex return with similar common things.
If you invest in index funds you're probably getting a standard form from Morgan Stanley, or Vanguard, or Charles Schwab, or something.
So you've got a very large set of people with very similar data sets/info so I'd expect failures to be the kind of thing we can train on - you don't need to deal with every case - even having a "this is non-standard, refer to a licensed CPA" is useful.
Basically I think the difference is, this seems way faster to get your "expert system" up to 60% of cases, and not that hard to get it to 80% of cases, and then you stop and let the rest get handled by computer PLUS human.
Expert systems trying to do 100% of things had issues and they were expensive to spin up even to the 60% mark.
Tax accounting for individuals and small businesses is the easiest part of the field by far, already largely automated if someone is keeping even half-decent books.
Expert systems are great at what they do, which is bringing ordinary people's activities up to the level they could have with good advice.
They just don't push the bar forward in terms of high end skill, either replacing actual experts or creating superhuman experts.
These new systems have the advantage that they've absorbed all the "common sense" that is implicit in all the training text, so you don't need to spend forever writing each of those common sense rules in - you can just focus on the advice you'd actually give a human who doesn't know the specific activity you're doing.
basically already true: https://x.com/henrythe9ths/status/2043109671251423416
> Better prompting / markdown files can improve e.g. claude code to do things the way you want, be better about fixes.
It _can_, sure. But it isn't guaranteed, and in practice, claude ignores the markdown files a lot of the time, especially as it gets closer to its context window filling up. I'm pretty convinced that giving markdown files to claude is just snake oil at this point.
Also "better prompting" means the *human* is learning, not the agent.
I was thinking that these could become better best practices to be shared across users though.
I get that public opinion in the UK is very anti-development, probably for deep-seated cultural reasons going back to William Blake if not Richard II (the play). But I'd think YIMBYism would be a good market niche for some aspiring British writer - even as a minority viewpoint it would be great for trolling and clickbait. Yet it seems to be an unoccupied niche. Does YIMBYism not exist at all over there?
Works in Progress, the publication Matt linked to, is based out of London. They’re generally very pro growth and supply side policies.
The Economist has often derisively referred to the British voter as 'Hobbits' for their intense provincialism at the expense of the wider country. As long as the tea is served and the pension is paid, who care if the little hobbits can afford houses or afford to live.
It does! At least in terms of journalism - the rather on-the-nose "YIMBY Pod" is the obvious candidate: https://www.abundancepod.com/ There are also some pressure groups, including one called "YIMBY Alliance". The authors of the essay "The Housing Theory of Everything" are all British. But I don't think it (yet?) has the same prominence in The Discourse as the American version.
It certainly does - as well as a number of others mentioned Jonn Elledge has a fairly successful Substack and used to have "Build More Bloody Houses" as his banner picture.
He was at the New Statesman (as an editor and columnist) for some years and now has columns there and at the New World, both of which raise some YIMBY points. There are a number of other youngish (20s/30s) journalists around who are also YIMBY inclined.
He used to write a regular column at CapX called NIMBYwatch, which is now written by James Ball.
I'd also add that the Young Liberals (the youth wing of the Lib Dems) are very YIMBY and there are repeated battles between them and the older more NIMBY end of the party. Young Greens get into similar internal fights with their older members. Polling consistently shows a big age divide on teh NIMBY/YIMBY split.
[Young Labour probably would as well, but Labour is structured to prevent internal fights from going public]
What there isn't is a significant right-wing YIMBY tendency at all - libertarianism basically doesn't exist in the UK.
I don't know whether they count as "significant", but there are certainly Conservative YIMBYs such as Robert Colvile: https://capx.co/the-secret-history-of-british-housing
They don’t seem to have any influence over the political parties of the right, which is what I meant by “significant”. Never seem to hear any YIMBY talking-points from Conservative or Reform politicians (where you do hear them from all three of the centre-left/left parties, even if not necessarily from their leaderships).
"My kingdom for a house!"
This you, Richard II?
Hey that was Richard III!
It's striking that in all those Shakespeare history plays, I can't remember more than a pro forma reference to policies, even in broad terms like modernization vs. traditional elites. It was all character.
Official censorship was a thing in Shakespeare's time, so setting stories in other countries and/or the distant past along with avoiding obvious political/social references to the contemporary order was important: https://www.sparknotes.com/shakespeare/life-and-times/historical-context/theatrical/censorship/
My kingdom for the correct Richard!
Technically, that roman numeral would be IV not III!
(Growing up in the state of Il caused me to develop a hatred of sans fonts)
I mean historically they just colonized places. This is just pains from the end of the British Empire. They should go take over somewhere in Africa again
Elie Hassenfeld is heir to the Hasbro fortune (“Hasbro” = Hassenfeld Brothers). Which might not necessarily be the most exciting fortune in the world. But he’ll be alright.
Anyway Zohran Mamdani is systematically selling out, whether voluntarily or not, and I am absolutely loving it.
Hasbro techbro.
I'm not addicted to Magic: The Gathering, it's just a more win-win way for me to back GiveWell's philanthropic endeavours. You can save four children's lives for the price of just one Black Lotus!
Oh god, don't get me started on what Hasbro is doing to Magic: the Gathering. It could best be described as a systematic attempt to satisfy shareholders' taste for not just goose meat but specifically golden-egg-laying goose meat.
Magic is suffering from its own variation on the UK's triple lock problem in that no set is allowed to ever be weaker in power level than the previous set while the set release schedule has also been absurdly accelerated, meaning that testing and quality control has gone out the window.
The Limited environment (which is all I care about) still seems pretty good - at least for large sets, just a problem of "too much" - I don't need 7 sets a year.
Yeah they’re getting better at making sets consistently fun to draft at the cost of taking a huge steaming dump all over every non-Commander constructed format.
#JusticeForVivi
Structural incentives always encourage TCGs to release more faster and creep power harder, and I think it's inevitable to hit rough patches like this sooner or later (and 30+ years is still a fantastic run)...but, man, the Extended Universe sets. I'm cool with closely-allied things like LOTR or FF, and niche things like Fallout can be alright too. Marvel and TMNT can fuck right off though. Nor does Avatar The Last Ringbearer really belong, although that's disappointing in a different way. Wouldn't matter so much if the core sets and plot were humming along nicely, but we completely lost the thread sometime around Brothers' War (if not earlier) and no signs of getting back on track. And I hate that you can't just ignore those sets either, cause they nevertheless contain some incredibly powerful stuff that warps metas both overall and locally.
Don't get me wrong, it's still fun to play when I can and I'm not in a rush to sell my collection or whatever. But I can't justify devoting most of my mindshare anymore to an obsession that seems obsessed with eating its own corn seed. When sets come and go and I literally don't feel like I have time to analyze them properly before the next spoilers are already up - What Are We Doing Here?
I agree with almost everything you said, but Avatar? You really don't think there's any synergy between Magic and Avatar?
I started right after Fourth Edition (late by many standards, ancient by current ones) and looked into it a bit last year for nostalgia reasons and JFC WTF pure madness.
It's also a bit ironic that the rich family in "Ready or Not" seem to be a fictional version Parker Brothers, whom Hasbro bought out decades ago.
Along with changes of Twitter/X, can we talk about the recent changes to the Substack App? They got rid of the feature where you can see all your subscriptions (free and paid) and click on them to navigate to them (the website never had this). Now you either have to Search or hope that the newsletter of interest is in your feed. The app was already problematic before. Old articles are hard to search for and you basically need to either search key words or endlessly scroll. You can’t “Find on this Page” within the app. There’s no personal comment history.
Apps like Twitter, Substack, and Reddit are all basically braindead simple from a technical perspective (now that cloud software exists - it was really hard two decades ago). So the only thing that actually matters is good product management and design. These are difficult disciplines and I have serious respect for people who can do them well.
Substack cannot do them well.
I have twice installed the app and both times instantly regretted it. The mobile website is far better experience.
Thanks. I thought it was just me. Zoom has also changed things, and it is making me nuts. I often wonder what the goal is of companies that make these changes that may lead to very slight improvements, but lots of annoyance and tech exhaustion.
Everyone hates the zoom changes.
The website does have a way of seeing all your subscriptions: go to settings and scroll down to the subscriptions list.
Also, my version of the app, I just tap on my picture on the top-right and it brings up exactly that list of subscriptions (paid first, alphabetical order). Is this an iOS/Android thing?
It used to be like this! They seem to have removed this feature at least on iOS. And they want people to use it more like Instagram or Facebook with a feed, where new posts are more like stories.
They’re probably going to do the same thing to the Android version eventually then. Sad.
The claim that the structural changes to Twitter have been good is insane. The big structural changes have been: paid accounts which get priority in the algorithm and in replies, paying people for engagement, turning the algorithm to focus on a uniform kind of viral engagement bait, allowing antisemitism and racism, losing to spammers so that DMs are now a cesspool, driving off advertisers so that ads are now insane people and scams, and making himself unavoidable on the platform. Which of those are good?
I disagree. The rare times that I tweet have led me to being followed by the most amazingly gorgeous women on the planet who are dazzled by my anodyne and cliched takes on the politics of the day. Who knew that was what sexy women are most turned on by?
It’s bots talking to bots now.
Twitter works well if you sort your feed by "Following" rather than "For You".
Also if you just go straight to the accounts you're interested in. It's rare that I ever look at my feed one way or another.
Facebook is a lot better if you switch to your Friends feed. But that isn’t the default when you’re on your phone and they are very good at sucking you in with that first screen. Trying to be more mindful each time I open the app, but it’s challenging.
I disagree; most of the changes I listed affect the experience even for those of us who primarily use "Following" (which I have always done).
The reason that Le Pen and Meloni are seen as fascist adjacent is that their parties are direct heirs of parties which were explicitly fascistic in the 1970s and later. It is bad but also not particularly relevant today, they have changed their policies and ideals.
It is worth noting that AfD is significantly worse and contains actual Nazis... there are degrees of this sort of thing.
The difficulty for an American consuming US media coverage on this stuff is that the coverage of AfD and Meloni is basically the same. If Le Pen/FN comes to power in France, what predictions should we make about the policies her government should pursue? What should we predict if Vox gets power in Spain?
There is a lot of hypocrisy there, the far left contains lots of fanatical antisemites and actual members of the Stasi but doesn't have the same firewall.
Indeed both parties have direct historical links to the Pétain and Mussolini regimes respectively. Meloni's party even has some Mussolini descendants in it.
The challenge is that if you move from an explicitly fascistic position to a still right-wing (but now democratic) position, you've not made all that many policy changes; you're primarily changing your position on the legitimacy of democracy (ie on whether you'll accept being voted out and on whether you'll accept the constitutional limits on your power). It's only possible to test whether that is true or not if you get into power - will you use it within limits, will you relinquish it? It's understandable that opponents are genuinely concerned about whether you're really changed or just started lying about this.
Rachele Mussolini defected to Berlusconi’s centre-right party because she felt Meloni’s party was too willing to use Mussolini linked symbols.
Yup. Once you have that label it's very hard to get rid of it, no matter how much you moderate. Especially if the denunciations are not that strong and have an implied wink for supporters that come from that side. Something Americans should be familiar with ("very fine people on both sides").
Meanwhile, someone like Macron goes out to make his own centrist party to get away from the negative branding of the Socialists and actually won.
I agree that Giorgia Meloni has generally not been that bad, but she endorsed Orban in Hungary. Hard to beat the fascist allegations when you’re a right winger who is supporting democratic backsliding in an EU country.
Yeah, a lot of the European far right parties could easily not cozy up to Putin. It's not that hard.
I don't think she is the only European politician that can be seen to back European regimes that undergo Democratic backsliding.
Greece banned the 2nd biggest opposition party a few years ago and no one batted an eyelid.
I find it really weird that it wasn't mentioned that Daniela Amodei is Dario Amodei's sister. (I had to Google that because I was wondering whether she was Dario's wife and the stuff about Holden Karnofsky had just been accidentally inserted in between.)
Thank you very much for answering my question!
I agree that Musk is a pretty bad tweeter, and for me, it's pretty easy for me to ignore him and his calvacade of accounts that he likes. If his preferneces were cut out of the equation, not much would change for me.
I do have to disagree in that one structural change he's done has been very bad, and that is that he has made the search function damn near worthless, and it's because of turning the verification system into a payola scheme. As smug as it may sound, some people have more worthwhile things to say than others, and the legacy verification system accomplished this pretty well. I could add a filter:verified parameter to filter out the chaff in a search before. But now, no more--I now have to take a shot at the Latest results instead to find tweets that are reliable.
Every social media company decided to make search dysfunctionial at the same time, I have no idea why
It's definitely bonkers that just using a plain Google search even today with all of Google's problems is often a more effective way to search for social media content than it is to use the social media site's own search function.
And this even with Google search getting worse than it used to be. When I ask a slightly obscure question, it tends to substitute more popular questions instead, and answer those.
For a concrete example, I wanted to test the claim that the largest centipedes known to have ever lived are still around today. I tried searching Google for largest fossil centipede, but the results were all for Arthropleura, which was not a centipede. Try "chilopod" instead of centipede, try more specific subtypes like Scolopendromorpha, try "evolution of chilopoda," etc., and it still "intelligently" detects that I am asking about myriapods and returns me to the more commonly discussed ones. Eventually I found scientific papers that described actual fossil centipedes, which are indeed small, but I was surprised by how advanced the search engine's artificial stupidity has become.
Artificial stupidity is not the lack of artificial intelligence, but a form of misalignment where it resists correctly understanding what you want.
All searches suddenly stopped you being able to search for a specific date range. I would be really curious to know why, it was a useful tool. Facebook and Google deliberately took away the capability.
For some reason, in the past couple of years, when I try to search for news articles I know I read between about 2005-2015, Google has gotten a lot worse at retrieving those articles. It used to be decent at retrieving older news articles.
It would seem to belie their goal of trying to keep you on their site as long as possible.
Yes, there are a lot of web-site related decisions that I just kind of throw my hands up at because, rationally, the decisions must be informed by metrics in some way to boost revenue, but heck if I can figure out how!
Or cut costs.
Agreed the search function simply just doesn’t work anymore