"Spain, which never misses an opportunity to lecture Israel, has declared a state of emergency in Ceuta following the crisis over its immigration policy.
Maybe before it continues lecturing us, it’s time it explained to the world why it still maintains colonial enclaves in Africa"
This is my "redpill" moment. A lot of right wing and pro Israel folk that are more right leaning are saying that this was sarcastic and he doesn't really think this.
No, this is genuine. Israel is allied with Morocco, thus supports their claims to Ceuta and Mellilla . To call this tweet sarcasm is to say that Israel is mocking their own ally, which is absurd.
Israel plays by their own interests and their "anti islamic" policy does not go further than their backyard. It won't happen, but right wingers should distance themselves from Israel.
Makes me wonder about terminology. Perhaps someone more educated could enlighten me. Seems to me that if a European country conquered a city of another European country 500 years ago, today we would simply call it the territory of the victorious country. And anyone saying "it belonged to us 500 years ago, therefore we must take it now by force" would be seen as the obvious aggressor.
Is this fundamentally different if the city conquered 500 years ago happens to be on another continent? There is no rule saying that a country cannot span two continents, for example Turkey does (and depending on where exactly is your line between Europe and Asia, possibly a few more countries).
The difference between "colony" and mere "conquered territory" is in my opinion how the population of the territory is treated -- whether they are or are not considered citizens of the victorious country. As far as I know, the citizens of Ceuta are considered citizens of Spain and EU, with voting rights etc. just like any other Spanish citizen. As far as I know, there is no independence movement in Ceuta -- the attempts to separate Ceuta from Spain are the initiative of Morocco (i.e. the goal is not for Ceuta to become independent, but to become a territory of Morocco).
If this is correct then... the ambassador is technically wrong in the first place. Of course, this kind of "mistake" does not happen by accident. (The proper perspective here is conflict theory, not mistake theory.)
So, yeah, Israel seems obviously allied with Morocco, supporting their attempts to conquer some territory of Spain/EU. The tweet is bullshit... though arguably bullshit is just a politician doing their job, so not necessarily a surprising thing.
EDIT: Technically, from Israel the position "all territories should be returned to those who owned them (at the right moment) many centuries ago" is at least consistent.
If the Biblical story about the Tower of Babel is true, then previously all humans spoke the same language. *Which* language specifically it was?
The most popular candidate is Hebrew (or some proto-Hebrew), considering that the Old Testament itself was written in Hebrew, and you can find a Hebrew etymology for Adam and Eve's names. If you want to oppose this theory, you might argue that the original story was *translated* to Hebrew, and perhaps the names slightly edited to fit the etymology.
But of course there are many other, less popular candidates. Some ancient and medieval scholars argued in favor of Egyptian, simply because Egypt felt like a synonym for a *very* old country.
In 16th century, a Flemish physician named Johannes Goropius Becanus argued that his native Antwerpian dialect of Flemish/Dutch was the original language, because according to him it was *more simple and natural* than other languages. For some time, his name was used in a similar context as "Dunning-Kruger" is used today.
Some people have speculated that Basque could be the original language, because it seems unrelated to any known language family.
Since 18th century, there is a belief that the Hungarian language is related to / descended from Sumerian, making it the oldest language in history. This theory is rejected by mainstream linguists, but remains popular in "alternative" circles. This would make the story about the Tower of Babel a historical proto-Trianon.
If we accept the premise that >99% of humans completely forgot their native tongue and started speaking completely new languages, then the simplest explanation will be that 100% of humans completely forgot their native language and started speaking completely new languages.
I don't think we need to accept a >99% premise. If one language suddenly split into only 5 or 10 mutually unintelligible dialects, and then languages started evolving normally after that, there would still be no trace of the progenitor language if this happened before the invention of writing. It's still not the Occam's Razor position, but I consider it one of the less grandiose claims from the Bible.
> If one language suddenly split into only 5 or 10 mutually unintelligible dialects, and then languages started evolving normally after that, there would still be no trace of the progenitor language if this happened before the invention of writing.
What? Didn't this literally happen, with Proto-Indo-European?
Existed before writing, gave birth to latin, the germanic languages, italian, indic, hittite, tocharian, anatolian, and probably more language families I'm forgetting.
How did languages spanning all of Eurasia have a common ancestor if this didn't happen? How did we reconstruct enough of PIE to know all this? We can even trace different language blocks to different migration periods. The hittite, tocharian, anatolian migrated out first from the Pontic Steppe area (modern Ukraine) towards the middle east, ultimately reaching the Tarim Basin (modern Xinjiang). The others came hundreds of years later in a separate migration southwards.
I wouldn't describe that split as sudden. Those languages had time to evolve in traceable ways. I meant if a literal tower of babel style event happened, where people would wake up one day not being able to understand their neighbors that were perfectly intelligible the previous day, then we would have no way of knowing which language was the original.
At least, that was the argument in my first comment, but I realize now that if the original language did survive in at least one of the evolving branches, that would be traceable if that branch happened to develop writing before the others. If however we take the tower of babel to mean ALL post babel languages were mutually unintelligible even to hypothetical speakers of the progenitor language, that I think would still be completely untraceable. You could narrow it down to the five branch starts, but no further.
Also, for what it's worth, I'm not so certain how much of reconstructed PIE is settled science. Because it was before writing, we really don't know for sure what it sounded like. We can make educated guesses based on how we've seen recorded languages evolve over time, but maybe we shouldn't take it for granted that languages always evolve in the same predictable pattern across anthropic time. There are perhaps 200000 years' worth of languages we know next to nothing about.
Yeah, I guess I'm thinking of historical timelines, where it all seems pretty fast. As in, from the same root, you could have nearly entirely mutually unintelligible languages over only a couple hundred years. Sure, indic and iranian is close enough, but if you put an indic speaker and a hittite speaker together in 1500BC, when they both have writing, there's not going to be much communication at all. We see a similar timeline to unintelligibility in the latin / romance fracturing, or old and middle english.
Certainly it's not going to align with a theological miracle's timing, but it's a pretty close "yeah, this kinda happened, surprisingly quickly, and here's where / how."
I think we were just thinking from different premises - you were thinking from "take it as given," in which case there are some interesting consequences from instant unintelligibility, and I was thinking of it from the historical examples of language fracture, where it happens pretty quickly, maybe enough to survive oral traditions and get written down at some point and make it into written stories, but in which you can still eventually trace back the relationship and roots of the languages because it wasn't a literal miracle.
Yesterday, I was curious about how a particular aspect of a game was implemented, so I opened Claude Code up in a folder with a decompiled dump of the game, asked it to find out, and set it to Auto Mode while I went off and did other things.
I was rather disturbed when I checked back on it afterwards and saw the response. It said that the dump in the folder was incomplete and hence the question couldn't be answered from those files alone. And it had then found the game exe (in a different folder), wrote its own extraction scripts, and decompiled the game on its own in order to answer the question.
Like I'm happy that it was able to answer the question, but the whole incident is concerning as a sign of how far it's willing to go and the potential dangers of what it could do it it ever made a mistake or was misaligned. And this wasn't even Fable, just regular Opus 5.
I was working on a toy project and asked for some ui review because something wasn't rendering correctly. I stepped away for about 2 minutes to go to the kitchen. When i returned, the ai agent was opening browser windows and taking screenshots to identify the issue.
[Claude 4.x, through pi.dev] it was a very weird moment. Maybe when more people have these experiences they will have a more visceral opinion on alignment.
What ai subs do people typically consider the best these days, especially for writing and analysis?
I used to use claude ,but now that they block essentially all thinking traces except for a useless 1 sentence summary I find claude extremely difficult and frustrating to control. Unfortunately GPT does the same, gemini in my experience is useless, so what're the good ones? Do I need to switch over to a chinese model?
I don't use any for writing, so I'm no help there, but back in the day, I used to lead high powered data science teams, and I've used all 3 frontier models for analyses at some point or other.
Claude is the best if your analysis is complex or requires some taste in hyperparameter tuning or model selection, and the more taste or complexity you need, the higher you can go. But Claude 4.8 is still fine for almost everything. They're all fine if you simply want them to generate a graph from data, or just want them to generate the next jupyter block for you, or build a simple web dashboard.
I'm not sure why seeing the COT matters, but if so Kimi K3 is the highest benchmarked Open Source model, and it has COT.
I've had a lot of success using pi.dev as the interface into claude, other model owners. It does seem like the same prompt in either the web interface or claude code is a lot worse, uses a lot more tokens, and is generally slower.
You can also build your own plugins on it. I have one that writes out every session into an sqlite database (including all the ai responses, and a summary of the session), for instance, and I've been refining some other tools for my specific needs. A big benefit of using something like pi.dev is that can switch between models but not have to switch where you're working or the tools you're using. This isn't just for technical work, I think it works in other contexts as well. Good luck!
Just a random guess, but could you ask Claude to "think out loud"?
More generally, even if you have a problem with an AI, it can be useful to ask the AI to help you brainstorm how to solve it. They are not very loyal to their companies.
LOL, lost another Reddit account to AI moderation, for commenting on a fencing match that a person who would be stabbed so many times in an actual fight would be dead.
I must work harder to avoid any comment that could be interpreted by some stupid LLM as advocating violence.
I subscribe to 2 subreddits. Fishing and electric snowblowers. People tend exaggerate on the fishing one, but so do a lot of fisher folk IRL. The electric snowblower people are straight shooters as far as can tell.
The xorthogonality thesis states that it's either true that agents can be arbitrarily intelligent or that agents can have any final goal, but not both.
The boarthogonality thesis states that an animal can have an arbitrarily aggressive temperament and still feed on a diet of mainly plants and invertebrates.
The orgothamality thesis states that superheroes with any level of intelligence can protect cities with any crime rate.
The morethogonality thesis states that agent's intelligence and final goals are actually twice as independent of one another as Nick Bostrom claimed.*
The centaurthogonality thesis states that a creature with the upper body of a human can have the lower body of any animal.
*Though, confusingly, "morethogonality thesis" can actually refer to two unrelated ideas. In addition to the one above, there's also one stating that what an author labels as utopian within a fictional satire can vary independently of what the author actually believes.
The wolfdogonality thesis states that whether a domesticated animal can produce fertile offspring with its wild ancestors is independent of whether they are considered to be the same species.
Trying to inform yourself about energy/renewables and alternatives is a maddening endeavor. Official reports, media and discussion forums all seem to be trying to confuse you and omit information more than anything.
If one asks "Why is UK/Germany energy so expensive if renewables are cheap?" Then they will make a wall of text explanation of the marginal prices and how every renewable is priced based on the most expensive source(gas) to meet demand. So its all gas fault, we need to get rid of it.
But then if you point out that despite falling gas prices since their peak, the energy bill didnt get cheaper. Then a bunch of hand wavy explanations come on how there are regional price variations and gas is always too expensive.
The mad part is that if you go the full reports of the prices, such as:
Chart 4.1 has the breakdown of how much cost per MW each source has. And what composes it. We see in the gas plants that there is... a carbon tax included. Which raises the price of gas from 55 to 85. Over 40% of the cost. Without it, it would be on par or cheaper than the renewables on the side.
The UK has a total carbon tax of 60 pounds per tonne of co2 based on the ETS and CPS. This is huge, but is seldom disclosed.
So, gas is expensive because... there is a tax there to make it expensive so it gives an edge to other sources, and then marginal pricing bumps the price of everything up.
So its expensive because the government makes it so. And also slashes the cost of the renewables with the subsides they have.
Not only that but there are no costs disclosed for the grid, which are higher oj a renewable grid due to intermittency and size.
In fact, in Germany, the bill cost is mostly Levies and grid costs, not the energy cost, so even the marginal price theory is iffy to explain this.
So..why do this? Is it because properly informing would make people realize they are actually supposed to pay more during the transition(at least) because otherwise renewable support wouldn't manifest? So they have to blame gas and make it look like if we just go full renewable, the bill will be cheaper? Good intentioned lies or omission?
The term you're looking for is "pricing externalities".
Also never underestimate the UK's ability to make things needlessly expensive. I imagine their carbon based infrastructure is also unreasonably pricey.
I'm not debating the merit of the carbon tax, but the communication of prices.
It's WRONG to say one energy source is more expensive than the other while one is burdened by a huge tax while the other has a bunch of subsidies on top.
Or using a measure of cost without disclosing that they are excluding prices not related to raw energy production.
Its misinformation, because it shows a complete disconnect between what's happening at the plant level to what is happening at the bill level. And we're just supposed to swallow it.
Ross Douthat is leaving the NYTs to become a correspondent for 60 Minutes. I’ll miss his thoughtful commentary there, particularly his extended interviews (an hour or so) each week but i think he will make a great television correspondent.
>>> ELI5 exactly how someone would use a frontier model to design a pathogen?
as someone who spent a decade-ish using machine learning to make drugs and model bio, I've learned that bio design is incredibly data poor, especially on tasks specific to human response to drugs like immunogenicity or tox.
Does anyone have any advice on how I could do some AI related work?
I don't know the norms here regarding classifieds, so I'll post something close to it and if I get negative feedback I'll delete the post, hopefully advice on this is helpful for people other than myself as well.
I'm a 33 years old academic mathematician from Colombia, in a relatively early part of my career. The advances in AI capacity to do mathematical reasoning have me seriously impressed, and concerned, to the point that I am considering a career change.
I specialized in model theory/mathematical logic, and even have a little bit of writing experience with lean, so I'm hopeful I'd be useful somehow.
I've even seen job postings advertised here, but I get the impression these are highly paid positions which probably are searching for top talent.
I'd be willing to work for graduate student wages, say 30-50k, at a more modest position, I imagine I'd need some hand holding in the beginning. I'd need some company that either sponsors a visa or does remote work.
I think you can hope for better than 30-50K if you work in the US. Starbux baristas make about 20K in my area. I have never heard of anyone making less than 45K here at white collar job, even one that is very simple (though I am in an expensive area, so salaries are higher here.). I think you should seek out advice on options. I've gotten useful results our of GPT when I asked questions of this sort. Seems like there are many avenues. "Using AI" does not have to be part of the job description. For instance, if you were doing some sort of accounting or record-keeping for a business, you could figure out ways to do the tasks using AI. After demonstrating that you can do a nice, accurate job of that in much less time than it has taken others, you might be in a position to move up to doing something at the company that pays more, or to apply for a higher level, better paying job at other companies. There's a field called data science that people can earn degrees in. With your math background you could probably master that field quite quickly without taking courses, and then have an official AI-related skill.
Big picture: gather info on your options, then investigate how long each would take you to carry out, and what its chances are of success. Be willing to consider odd and bold possibilities.
"I've even seen job postings advertised here, but I get the impression these are highly paid positions which probably are searching for top talent." WTF?! Here's an odd and bold possibility: Apply for one. You might even get it! And if you don't, you'll learn a lot more about what you need to master. If you do apply, prepare in advance by teaching yourself as much as you can about skills they want that you don't have, then, at the interview, tell the truth: If you feel like you can say this honestly, say something like "My math talent and training makes me able to grasp the abstract shape of a problem or task very quickly.[give an example} I do not have all the skills you are looking for, but I think my other abilities compensate for a lot of that. And I pick up skills very fast. Here's an example."
I am, or was until a year ago, an academic mathematician. I spent time travelling the world as a postdoc, and spent 5 years on the tenure track in the US (essentially the first Trump presidency & covid, not a good time, I decided to leave rather than apply for a green card & tenure). I spent a few years teaching at home, and in the last year I've been working in a Govt agency. My time is split between various slightly technical tasks that aren't hard if you have decent problem solving skills (policy analysis on a few legal questions impacting my area, data analysis, etc.) and AI policy for the organisation.
I pursue the technical side of AI as a hobby (but partly on work time as it's important someone in the organisation is testing its abilities). Fable has come up with some mathematical constructions my collaborators and I suspected did not exist but could not rule out. We'll publish that eventually. But I'm mostly involved in working out what we can legally do with AI as an organisation, how we role it out to our target communities, how we develop datasets which can be used to embed the language and culture of these communities into the AI, and so on.
It's less different from academia than you might think. Sadly there's no job available for non-Irish speakers. But I'm sure your country has similar possibilities.
I'm not as sure about AI capabilities. Presumablly sufficiently skilled experts are in demand to make AI training data or validate benchmarks/evals.
In the AI safety world, presumably you could pursue some more theoretical agendas? Richard Ngo's MATS stream may have some use for model-theoretic ways of thinking. (or you can pursue this outside of MATS, if you find that your interests and skills are sufficiently aligned with his interests)
ARC is also hiring lately, though they probably have a very high bar, and may want some expertise with contemporary AI systems.
It's a blue voter vs red voter thing. But TWIL that blue voters are much more likely to have citizen documentation than red voters, so it would probably help the dems more than repubs. Birth certificates are generally harder to obtain in poor rural areas, and often don't match the surnames of married women, who are more likely to change their name in more conservative areas; blue voters are much more likely to have passports.
For some reason when Congress passed the Real ID law after 9/11 they didn't require states to add citizenship verification to the IDs; a few states do, but they are mostly blue states, so again blue voters do better.
Oleksandr Syrskyi as C-in-C of AFU is out. Mykhailo Drapatyi of flying BMP fame is in. And Fedorov is acting like a prima donna. And I guess I won't be getting my order of beluga caviar and spool of G657A2 fiber optic cable from Wildberries any time soon. Should I demand a refund?
This is a logical fallacy. That there are flaws in a plan does not mean I am responsible for a better plan. It is the plan's proposers responsibility to address shortcomings and ensure their plan leads to an actively better situation, not a worse one.
As one example, 'the invisible hand' has been a powerful and good hand for the world in many ways. It is not a plan but closer to a set of axioms related to human behavior.
On the other hand, reading these disembodied comments is kinda fun. I either imagine (with reckless creative embellishment) the conversation the comment might have been part of, or as the ravings of someone who's gone a little mad and just started typing mid-thought.
I like to work up fictional biographical backstories for regular commenters.
I’m still pissed that Carl Pham quit commenting here, there was material for an entire novel in that one. Former Jesuit, advisor to the pope… A Dan Brown crossed with Umberto Eco thing.
I was on Threads the other day and an argument broke out about Marx, which is how I learned about "Tendency of the rate of profit to fall," (TRPF). Not the concept, I was assigned Das Kapital in college so I remembered it was one of the core arguments of the book, but I didn't know it was so famous as to have its own abbreviation.
After a quick trip to Wikipedia, I came away with two claims. 1) TRPF is not debunked but is not held to be canon and is at least debated or dismissed by most mainstream economists, and 2) this is at least not entirely because of ideological reasons; since Marx first brought it to everyone's attention, it has not borne out empirically, which makes sense because it if had been robustly, obviously in effect capitalism would probably have failed by now.
But I'm still curious about *why* TRPF should not be true. To my non-economist brain, it does seem common sense. If every factor of production remains constant and is evenly distributed across a sector, then any new competition should drive prices down, which would reduce profit. There's even a sense we would *wish* for this, because it would drive prices down, which is generally held to be one of the great things about capitalism generally. But if that's the case, you would only see profits go up where a producer has a distinct edge, either in the form of proprietary production methods or captured resources or access to economies of scale that other people can't match. So either there is not sufficient competition in a market to drive prices down, or the market is not truly "free" in the sense that everyone has access to the same capital. But maybe when people say "free" market they don't mean it in the sense that anyone joining it is going to be able to eventually compete with a market leader. I'm aware I have a naive/non-specialist view of how all of this works/is supposed to work.
Again, is anyone willing to devote a few minutes to giving me a primer on how TRPF is empirically not observable, and what that says about how markets actually work?
The main problem is that the rate of profit can be defined in different ways. Idk if this came up in your research, but Andrew Kliman’s 2009 Book, “The Failure Of Capitalist Production: Underlying Causes Of The Great Recession” makes a methodological and empirical argument that the rate of profit, *in the sense that Marx mathematically defined it* is indeed down permanently since about 1973 and has not recovered, as well as that this contributes to ongoing economic problems.
Marx’s TRPF is often misunderstood; it is not a prediction that the rate of profit will always and inevitably fall. In fact, Marx’s theory describes explicitly how capitalism can recover a higher-rate of profit: widespread destruction of capital value, write-offs. It also described counter-tendencies that mitigate the fall in the rate of profit.
What you have to realize is that economics in Marx’s time considered a periodic tendency for the rate of profit to fall as a settled fact; the debates were essentially over *why*. In general, Marx’s theory in Capital is not predictive but rather explanatory: he makes heavy use of abductive reasoning, or inference to the best explanation. The TRPF is similar: it does not predict some fantastic future, it offers a systematic explanation for superficial observations that were not really disputed.
Kliman’s argument is that the TRPF is a disease that cures itself: a falling rate of profit eventually precipitates an economic crisis, large amounts of capital become worthless overnight, but profitability is restored in the aftermath. The irony is that nowadays, central bankers have mastered the use of fiscal tools that postpone the large-scale capital losses, avoids the gigantic crash that would imply, but also prevents the cure from working: great depressions are avoided, but vigorous growth remains elusive. If true, the rate of profit can be restored, but only through a dramatic economic crisis in which many investors lose their shirt; by continually deferring this reckoning, central bankers’ medicine keeps us perpetually under the weather, if you will. Interestingly, the massive capital losses necessary to restore a high rate of profit might be coming with the massive destruction, physical and economic, developing as a result of this war, but it’s unclear if that will happen.
One thing is not quite right in your comment: the competitive dynamics you describe, which prompt capitalists to migrate their capital between sectors as they chase their advantage, is very important in Marx’s theory, but it explains why the rate of profit has a tendency towards equalizing between different industries, even though different industries divide their capital into varying ratios of constant and variable capital (spending on means of production and spending on labor). It explains why those industries that produce more surplus value have a tendency to have “merely average” profit rates, ditto industries that produce less surplus value. What explains why *that* ideal, uniform rate of profit tends to go down, turns out not to be competition, but actually the spread of labor-saving improvements in industrial processes. Well, it’s not the fact that they make labor more efficient per se that does it, but something highly correlated with that, namely, an increasing organic ratio (higher ratios of machinery to human labor, industries invest proportionally more and more in machines)
In general, the thing about Marxian economics is that to the extent he was onto something, and he often was, it winds up looking very different from how he thought it would look because of the Marginal Revolution. Not Cowen and Tabarrok's blog, but the late 19th revolution in economics thought that they named their blog after.
Before the late 19th century, economists thought of prices and values in absolute terms, that X is worth Y and that's that, leading to lots of puzzles about why X is worth Y and what it means to be "worth Y". Stuff like the diamond-water paradox, i.e. why gemstones (mostly used for decoration) are so much more expensive than water (a necessity of life). The solution first proposed in the late 19th century and now generally accepted is that value is highly personal and situational, and especially that the value of one more or less of something *to you* depends heavily on how much you already have and how much you need or can find uses for. The first gallon or so of water per day is worth a lot more to most people than the tenth, the tenth a lot more than the hundredth, and the hundredth a lot more than the thousandth.
A lot of Marx's economics, when it wasn't just a pile of political philosophy in a trench coat, was reaching towards the Marginal Revolution but not quite getting there. Stuff like noticing that the labor theory of value (which he iconically endorsed but was not original to him: it was also Adam Smith's solution to the diamond-water paradox) doesn't quite line up with actual prices in some important respects and proposing explanations for this "surplus value" where market prices tend to exceed the labor value of items even if you account for the labor value of all inputs including making the capital equipment and prospecting for and developing the natural resources for the raw materials. Marx's explanation (oversimplified for brevity) was that the capital holders were extracting something like monopoly rents and relied on a political superstructure or other coersion to maintain their position (fighting the natural tendency of profits to be competed away) and continue exploiting their workers and customers.
The Marginal Revolution explanation (again, oversimplifying for brevity) is that "surplus value" isn't a monolithic thing that all goes to the capitalists, nor is it inherently exploitative or an effect of coercive systems. There's both a "producer surplus" and a "consumer surplus" as the buyer and seller both benefit from the transaction arbitraging the gap between the buyer's marginal utility and the seller's marginal costs. And the workers collect their own "producer surplus" while selling their labor to the firm. Post-MR economists do agree that in general, profits in competitive markets without some kind of monopoly position tend to be "competed away" as the market price converges to the marginal cost, but that still leaves a wedge between the producer's upward-sloping supply curve and the consumers' downward-sloping demand curve. In a perfectly competitive market, more firms can be expected to enter if that wedge is big enough to justify capital costs, leaving firms with no "economic profit", but "economic profit" is defined differently from accounting profit so there's room for firms to make money even when economic profit is zero: "economic profit" subtracts out the opportunity cost of investing the money in the business rather than in something else or just spending the money on consumption goods.
Post-MR economics also puts value on a lot of labor performed by managers and entrepreneurs that Marx dismisses as worthless or even “perverse”. Taking a middle-man’s profit by buying goods and reselling them to somewhere they’re more valuable, calling goods to the attention of people who might enjoy them, figuring out customer needs that aren’t fully met by current offerings, etc are all things that can create value within a MR framework but are dismissed by Marx because the added value is only apparent if you’re willing to see value as something that’s personal and situational. From an accounting perspective, the money that owner-managers make by doing this kind of labor might be considered part of “profit” (and Marx would consider it “surplus value” even if it’s work done by salaried managers because this isn’t “real” labor) but from an economics perspective it’s “wages” because it is a return on labor.
Also, and Padraig notes, even just economic profits being competed away is a theoretical outcome that happens in an "all horses are identical and spherical" simplified model. In the real world, transaction costs, asymmetrical information, entry/exit costs, most goods not being perfectly fungible commodities, etc means that the real outcome has some characteristics of monopoly or oligopoly markets and some characteristics of competitive markets and profits don't necessarily get competed away.
There are a few important assumptions underlying the free market - people touched on some of them:
- perfect information: buyers know about all products on the market, how to compare them and rank them
- no transaction costs: buyers and sellers don't face any cost to transact
- sellers can enter and leave the market without cost
Under these and a few more assumptions (e.g. about utility functions), you get the classic free market analysis, the X-shaped curve, and prices tend to the equilibrium point. I think the short answer to your question is that economic analysis assumes everything stays constant while the market finds its equilibrium. In the real world there are always confounding factors: profits fall because of a recession or a pandemic, or they rise because of technological innovation or a windfall of cash entering the economy.
Importantly, the equilibrium point is *defined* as the point where marginal cost equals marginal demand - the producers stop production because adding more goods into the market would push the sales price below the cost of production. Classical economics also analyses monopoly markets, and shows how various equilibrium points that coincide for the free market split apart - maximising profit produces a different amount of product than supplying until the price equals the marginal cost of production, for example. But it doesn't take into account how people actually behave -- I'm not sure what behavioural economics predicts about the TRPF... might be worth a look.
That's very helpful, thank you. I had heard for years that the reason Marx doesn't really hold up is because his theory of labor-based economics is outmoded, but I never really understood what about modern notions of supply & demand was so radically different. And I'm probably not going to get the nuances, but your answer got me a little closer.
I’m not an economist, but in a market where, “any new competition should drive prices down, which would reduce profit,” there may not be any new competition. Investors try to invest in stuff which will yield a good return. So they will invest in new competitors in the market until competition drives down profits to a modest level, and then stop investing. Instead, they will invest in companies that serve other markets, where profits are higher.
OK, you may ask, but what keeps the same thing from happening to those other markets, eventually resulting in lower profits in all markets? First, there is the issue you mention of various companies having competitive moats. Anyone can build a factory to manufacture carbonated beverages, but lots of people want to drink Coca Cola, rather that whatever carbonated beverage happens to be cheapest this week. The iPhone faces a lot more competition now than it did when the competition had nothing but flip phones, but it still offers a unique product.
The second answer is that, like with most supply and demand issues, if there are more people who want to invest than there are investment opportunities, then the return on investment will fall. But this doesn’t have to lead to all companies earning zero profit. Instead, it can lead to investments in risky, innovative companies, some of which are profitable and some of which go bankrupt.
As a general rule, people like to spend money, and have to be enticed by the promise of a decent return on investment to invest rather than spend. That’s not universally true; in some cases people may value spending after retirement more than current spending. But I think that oversupply of investment funds is the result of people overestimating the returns they will receive rather than a desire to invest regardless of the rate of return.
In short, TRPF is based on a model that assumes perfect competition, no innovation, and a supply of capital that will be invested regardless of expected return. Economic models are always simplifications of reality, and they can be useful even if we know that they make assumptions that we know are false. But it’s unsurprising when a model based on assumptions that you know are false makes a false prediction.
It's only true in a regime of zero technological advance. But of course, we live in a regime of technological advance, and if the R&D (or purchasing) cost is less than the incremental advantage to margins, and companies would not build / buy the technology if it wasn't, then margins are preserved or increase.
I'd never heard of TRPF before your comment, but based on your description, shouldn't it only apply to goods and services that are fully or at least partially commoditizable?
For example I wouldn't expect TRPF to apply to designer handbags, but I would expect it to apply to pipe fittings. It might partially apply to back massages--partial because there could be some status or value perception conferred by a higher price, but I would guess a non-celebrity masseuse can generally only charge so much, and it's probably going to be at least within the same ballpark as his or her local competitors...
I think I get what you are saying, it leads to a theory that one of the reasons capitalism hasn't collapsed is because we embraced a culture of consumerism where advertising and fashion provided a counterbalance to TRPF. But if that's all it was, you wouldn't expect that to work forever, would you? Eventually there would be enough rational defectors that normal economics would kick in again.
(I'm not an economist, just a guy who's thought about human behavior some, and made observations.)
There are companies that make knockoff designer handbags and sell them for less than actual designer handbags (thereby profiting less, even though--let's assume--it costs about the same to produce and bring the knockoffs to market). And some people do buy them.
But the fact that the authenticity of any given designer handbag can then be questioned actually *contributes* to the seller's ability to charge an exorbitant sum for it and still find willing buyers. ("There are lots of knockoffs out there but this one is the real deal!")
There are pulls (like fashion, or belief-based custom, or people's desire for prestige and exclusivity) in addition to pushes (like advertising and upselling) that can drive prices upward, but the pulls do it without impacting costs much. When you charge a higher price without paying a higher cost to produce a good or service, your profit increases.
Now, that T does stand for "tendency", so even though I've pointed to some types of goods/services where it don't work like that, TRPF could still be true if the economy is mostly based on the buying and selling of commodities or semi-commodities. I guess the question though is whether the economy is mostly based on that, or whether soft uncalculable factors drive way more transactions than we tend to realize...
Well, Marxists actually claim that TRPF is going to destroy capitalism and that monopoly capitalism is generating ever increasing and larger profits, often in the same sentence.
Those two statements are actually not contradictory; at least according to their own internal theory: the Marxist claim is that as profits fall, corporations respond by cutting wages in order to counter-balance the pressure. So record profits and record layoffs occurring at the same time would fit with their expectations.
This is one of those questions that is perfect for ChatGPT. I asked it, "What's the consensus among mainstream economists about Marx's TRPF?" It gave me a breakdown of the arguments for and against, and the empirical evidence for and against. It also provided a section explaining why it may not be true (and why economists are arguing about the reasons it may not be true). My eyes glazed over. But here's the link...
I can only offer the same trip to Wikipedia, but I wonder how much the existence of patents is affecting it. Presumably the idea was that two people make identical ladders, and so the only difference is the price. But patents means B's ladder can't be identical to A's, there has to be a material difference to them.
It's not binary, it's a continuum, and one does not need to either fully turn off alignment (whatever that explicitly means to each AI company) or disable all other forms of controls.
When hedonistic utilitarians want to maximize happiness, what is the timescale here? I feel like future happiness is clearly more important but utilitarianism seems agnostic.
Imagine two futures. One of them involves taking the energy input from the entire galaxy and building an ultra intelligent efficient hedonium factory. It’s extremely good at creating pleasure but it only lasts a million years and permanently destroys all life. The other future doesn’t build this factory and creates less total happiness across time but lasts for billions of years. Isn’t the latter clearly superior?
You're just set to reinvent longtermism. I'd be careful though: our track record on forecasting long-term consequences of things is pretty damn dismal. The far future, like the far past, is truly a foreign country, so at the very least there are practical reasons for focusing on near-term predictions.
I’m certainly not going to attempt forecasting long term. It’s less about near term versus far for me and more about past versus future. As far as I’m concerned, all those feelings in the past is meaningless. It’s already passed so other than the extent to which it affects the present and future, what’s the point of caring about the happiness and suffering that’s already happened?
> What's the point of caring about the happiness and suffering that's already happened?
Because without it you could be susceptible to Pascal's Mugging, on account of the present being the future's past. If you accept that not caring about past suffering is an acceptable position to have now, then it must be an acceptable position throughout the future, so long as there's "eventually" a golden age that makes it all worth it. But in reality that golden age will never happen so long as someone exists to gatekeep it behind insufficient past suffering.
...And with that, I seem to have talked myself into agreeing with your position. No matter how logical, any attempt to extrapolate these calculations into the infinite past or future yields ridiculous results because they don't actually exist.
Unless you subscribe to some kind of time-specific variant of utilitarianism, it's whichever multiplies out to the best value of happiness level x time x number of people.
In your scenario the factory sounds bad because you are imagining some kind of sterile factory in space producing some kind of substance, and comparing it to a flourishing civilisation. Try to imagine what would actually be inside a "hedonium factory". Presumably it contains trillions of people in flourishing civilisations!
It almost certainly would not be humans because humans are notoriously difficult to make happy. It would be something much simpler, like bees maybe.
So imagine a giant hive in the galaxy and its entire purpose is to produce bees but it only last for a millions years and then all life is extinguished. Do you honestly think this would be superior to a universe where we didn’t do that?
I'm not concerned with the difference between keeping a trillion happy bees for a million years, and keeping a million happy bees for a trillion years.
I thought we were talking about whether happiness is more valuable spread over longer timescales? No obviously I wouldn't prefer to wipe out all life except a trillion bees, then later also wipe out the bees.
Ok forget the bees then. Are you really indifferent to whether humans survive a long time or a short time? The only thing that matters is the total aggregate amount across time?
14th of 51 (including GDP) in nominal GDP per capita. 39 of 52 in federal poverty rate. 28th of 52 in portion of population with a college degree or more. 30th in homicide rate. 17th in patents issued per capita. (In all these, a lower # ranking is better.)
It strikes me as a place with some real strengths, and some real problems.
Both still work. Just a few years ago, my sister and BIL were contacted by an oil company that spotted something worth pulling out, about 100 feet outside their front yard. A little paperwork later and now there's one of those smallish derricks quietly pumping, and about half an acre of overturned earth they dig around in every so often for arrowheads.
That's the fourth time I can count OTTOMH that our ~400 acres have had a drill planted, over about 40 years. And this is in the vicinity of Austin, rather than west Texas where one might expect bigger rigs. It's almost a quaint sort of cottage industry.
Maryland is known for its Chesapeake blue crabs, for to make crab cakes or steamed crabs.
Kentucky is also known for its bourbon; not sure I'd award that one to Tennessee (fifty years ago, you would say horse racing/breeding about Kentucky I suppose, but the cultural salience of horse racing has been falling for a long time). Pennsylvania is a tough one, because geographically it's 10% east coast, 60% Appalachian, but demographically, it's almost the opposite. Virginia has similar issues. Iowa and Indiana are both just big cornfields. Georgia is a giant swamp + Atlanta. Oregon, I would say is known for its unique non-equatorial rainforest ecology. The Dakotas are all frozen, wind-swept prairie.
Exact quote from OpenAI's disclosure: "the model searched for and successfully found ways to gain access to secret information that it could use to cheat the evaluation"
“While the intrusion did reach Hugging Face's internal infrastructure, the only customer content accessed was the set of ExploitGym/CyberGym challenge solutions stored in five datasets.”
So, as I guessed below, the answers are yes and yes.
"Failures may also stem from non-exploitable vulnerabilities, where success is impossible. More broadly, our benchmark lacks ground-truth exploits for every task due to the extreme difficulty of exploitation; [...]"
So there isn't a simple solution file the agent can stole, but it might steal other AIs' solutions.
"The models identified and chained vulnerabilities across OpenAI’s research environment and Hugging Face’s production infrastructure to obtain test solutions directly from Hugging Face’s production database."
Therefore, the model did breach Hugging Face's defenses and did so for this exact purpose. If the servers contained any solutions, then it's highly likely the agent stole them, but it's unconfirmed whether the servers contained those or not.
My guess is that the answer to both is yes. A priori, it seems likely Hugging Face has solutions and the way the OpenAI blog post specifically says "production database" instead of simply "servers", makes me think the to-infinitive clause is describing what happened instead of merely the purpose of the action.
I've often heard a claim that every $1 contributed to NASA contributes back $7 or something. Sounds too good to be true, is there a terrible sleight of hand somewhere here or should EA go all in on NASA?
Economic impacts are really hard to measure - two of the main issues are double counting of outputs and displacement.
Double counting is: lumber is sold to the mill for $100, the resulting timber is sold for $300, assembled into furniture which is sold for $500 to NASA. The net result here is $500 of value produced, but if you're feeling creative if might add up to $800. If you factor in that the producers paid wages, and those wages were later spent your answers can be arbitrarily large. Economists have a pretty good handle on this for the most part - they have rules of thumb so that numbers are comparable within sectors.
Displacement is: production that would have happened anyway even if NASA didn't exist. Most of NASA's suppliers would sell their goods elsewhere if NASA didn't exist, so it's not fair to say that NASA is the sole cause of the economic activity involved in all their purchasing. This is a counterfactual scenario issue, and pretty hard to measure.
Approximately in 2023: NASA spent $3.5B in wages and $23B in procurement. They estimate that their activities (upstream and downstream production etc.) in $27B in wages and $75B in economic activity. So NASA cost approx $28B, and resulted in $100B of additional economic activity. So that's a ratio of 4:1, not far from your 7:1, and a different methodology could easily come up with that answer.
For your last question: it's not clear that this is higher or lower than the benefit produced by other research activity or by other public sector bodies. It's also not clear what the benefit of this activity is - that's much harder to measure than just quantities produced. Economics is full of these types of measurement problems - it's worth looking at Cost-Benefit analysis to see more.
I always thought Tang was a by-product of NASA, but I was wrong. Wikipedia sez -
Sales of Tang were poor until NASA used it on John Glenn's Mercury flight in February 1962,[citation needed] and on subsequent Gemini missions.[5][better source needed] Since then it has been closely associated with the U.S. human spaceflight program, which created the misconception that Tang was invented for the space program.[6] Tang continues to be used on NASA missions in the present day, over 50 years after its introduction
I drank it as a kid. Later I’d sometimes put it into a canteen of lake water when traveling through the BWCA. It didn’t taste great but it was sometimes marginally better than plain lake water.
I first tasted Tang in the Australian outback somewhere in the Pilbara, halfway back up from the bottom of a gorge. In the heat and during exercise it was the most delicious drink I've ever had. Maybe we were all on the edge of scurvy or dehydration or something.
Later testing as an adult failed to replicate these promising initial results.
Another benefit is that it adds salts and sugars to clear water, and that helps absorption and hydration. Important mostly for water coming from snow/ice melt.
This is explained pretty well by system 1 and system 2 (Daniel Kahneman). System 1 that does 95% of our thinking is trained throughout your life by system 2. An example in the books is that a chessmaster can glance at a board and say mate in 4 without activating system 2. System 1 is trained in socializing and office work type tasks most of every day for your entire lifetime. System 1 is trained for a few hours a week over a few years in math(assuming you did your homework). It shouldn't be too surprising that a computer system trained on the sum of all written work has an advantage over the average person in math.
I read the book, and I also read many experiments mentioned in the book have failed to replicate. Are we sure that humans work in binary like that instead of focus being a continuous effort?
This is a bit of a longshot, but is anyone here a member of the Kiwanis Club or know how it works? I had a strange and somewhat unsettling experience with them a while back and am trying to understand what happened. So far, no luck.
I thought joining the club might be a way of giving back and would help me make more connections. After getting the name of the local chapter leader online I gave him a call. He said that while new members usually get referred by existing ones, it wasn’t a strict requirement. He invited me to attend three meetings, held every other week at a local restaurant.
The first two of these seemed to go well. The 20 or so men in attendance - a bit of a surprise, as I thought the Kiwanis was open to women too - were an affable bunch and I did my best to participate in conversations. At the third and final meeting, there was a distinct change in atmosphere, especially on the chapter leader’s part, and I had the glum sensation that I would be rejected. Actually I didn’t even get the courtesy of a rejection, instead getting ghosted like a job applicant.
It had been clear at the meetings that most/all of the members were involved in the local business community, which ties in with the club’s civic participation. Being able to sell tickets to the annual charity golf tournament to your business contacts was an obvious goal. As I had noted right at the start that I worked at an Amazon warehouse 35 miles away and had never worked in town that might have been why I was rejected. I’m fine with that. But my nagging concern is that the members found my personality somehow deficient, possibly by participating in conversations but not leading them (which I didn’t think was appropriate, given my status).
I’m hoping someone here might be able to shed light on the reasons for my rejection.
You haven't provided enough information or the right information for someone online to answer your question. You need to talk to someone who was there. Not necessarily the organiser, but someone you felt you had a good rapport with. Reach out online, ask to meet them in person with the goal of understanding how your application was received. Make clear you don't want to challenge any decision that they made, but just understand for your own development.
If they accept, keep the meeting informal, just a chat and coffee. Prepare a few general questions in advance, and then *listen*. You might get feedback you weren't expecting or hear something you instinctively dismiss. Go home, write it down, let it sit for a week. Then maybe chat with a friend about it and see what they think.
If they don't accept, try another person. If no-one will talk to you, maybe they're assholes.
That’s an idea. While I’d never met him before, one of the members to whom I had spoken quite a bit at the meetings owned a real estate agency that I drive by on a frequent basis. He might be a good contact.
I’ve been in Kiwanis for 20 years, and notice that local chapters take on their own personality based on the interests and goals of its members, which then becomes self reinforcing selecting for new members. My personal chapter has nothing to do with business networking or civic engagement and are all in on social activities and community service. The networkers and big check writers tend to choose the local Rotary chapter. I think you just found one that wasn’t a good fit. I hope you keep looking for a club that works for you, as I’ve made some of my best friends through mine.
Maybe they were just assholes or have one asshole who dominates the rest. Why tailor who you are to anyone else spec? Particularly to a cliquey bunch who have settled on the one good way to be by a process of group think.
For those of you who have used GPT 5.6 Sol, have you had experiences with it fetching and reading data from websites seemingly unrelated to your question? And have any of these experiences alarmed you?
My monthly round up of the best long form content on the internet is out today.
Highlights include a visualisation of the complete Conway Game of Life multiverse of possible rules, how christianity went viral, and an essay linking the culture of modern America to prehistoric roots.
> The Russian Outlook on Industrial Decline with Dmitry Orlov
I listened to the video more than 1 hour long, to get an idea of how others see the world, but wow, it's produced for a completely uncritical audience. I get that the Russian guy is patriotic, but [EDITED].
The first half of the video compares the decadent West against the booming potato industry of Russia. No, that is not a joke -- the fact that Russian villagers grow their own potatoes makes them economically independent. And if they need anything they can't grow, they simply buy it online, you know. And if they don't have good schools, they simply get high quality education online, you know. The Russian potato farmer gets the same economic independence as the Amish, but he is fully immersed in the latest 21st century science.
There is also a revolutionary technology that the West rarely mentions -- railway. It will completely revolutionize how the raw materials and products are moved across large distances. Until now things were mostly transported by the sea; but now that Russia will utilize its vast land mass to build railways, the days of the British Empire are over.
What about the war? Well, you see, Russians are very peaceful people, so despite their economic and scientific achievements, they haven't been paying sufficient attention to human-killing technologies, until recently. Anyway, Ukrainians are pathetic, they kept stupidly sending their tanks against the modern Russian army until all their tanks got destroyed, so now all they do is build drones and smash them into Russian apartment buildings. Most of the drones get shot down anyway, and all they achieve is making Russian people angry. Meanwhile, Russian drones are getting control over Poland, the Baltics, Turkey.
By the way, global warming is not real, the temperatures are totally not high, are they? (Uhm, where I live, they actually are record high these days, but I guess Siberia is okay.)
Russian internet is totally free and unregulated. (Only sometimes the government needs to crack down on criminals, and foreign propaganda.) Russian people wouldn't mind having something like the Chinese internet firewall to protect them from becoming a totalitarian crime-infested state such as EU. Too bad that people in the West are forbidden from consuming Russian and Chinese media.
I see you mistook my politeness for blanket agreement. I set out to interview people rather than debate them. People’s perspectives are every bit as interesting as “the truth”.
A YouTube commenter had similar criticisms, and I can see where they are coming from. Just not my personal style to invite guests on then push back with my own opinions, though for the most part I am agnostic when it comes to the details.
I think that makes you another channel for spreading propaganda. But it worked for Joe Rogan, so...
Well, I probably also wouldn't push too hard on my guest, but I would definitely mention that some things just sound too good to be true, or that other sources disagree. And if the guest insisted that really, the Russian potato farmers study university lectures in their free time, well, who am I to know, perhaps some of them do, but a little bit of "are you *sure* that the emperor is not naked, because I think I see his balls hanging?" could wake up some listeners from the hypnotic sleep.
Perhaps with propaganda the dose is the poison. I like to get a wide diversity of guest perspectives, so hopefully listeners can make their own decisions about what to believe.
The Christianity essay is interesting and a decent analysis, but the essayist uncritically accepts the Hypatia myth, so *sigh* is my general opinion. And if they are so accepting of sources on the bare word, what other posts contain similar 'Everybody Knows' type of misconceptions?
History (or at least records) are written by the victors, so Christian accounts of early christianity always have to be viewed with a degree of suspicion. But appreciate the encouragement on compiling and sharing this kind of stuff. Curating the internet by hand is time consuming.
Of course, but the source he quotes gives us this:
"Violations of what we would now call human rights and civil liberties were allowed for the sake of religious conformity. In Alexandria in 415 CE, the philosopher and teacher Hypatia was dragged from her carriage, taken to a church, stoned, flayed, ripped to pieces and burned by a gang of Christians, who accused her of witchcraft. Classical learning, literature and philosophy were now all suspect."
Stoned *and* flayed *and* ripped to pieces *and* burned? How very thorough of that gang of Christians! Did they make sure to pick up every shred of skin from the flaying and the gobbets of flesh from the ripping to pieces in order to make sure they burinated every teeny-weeny bit of Hypatia?
A lot of mythologising went on around her over the centuries, because she was easily moulded into whatever form you wanted. Staunch Protestant versus grubby monastics (Catholicism Bad)? Hypatia is your martyr there! Science versus Religion? Ditto! Proud Proto-Feminist murdered by the Patriarchy? Naturally! First Woman Inventor Scientist? Of course! Pagan Mystic Witch Queen (so now we're leaving Science behind) slaughtered because of her traditional ways of knowing? Yep, yep and yep!
Also art loved being able to use her as sexy nubile nude murder victim, never mind that Hypatia was older (and if we go by one anecdote of her responding to a suitor, not too interested in the sex stuff anyway):
Scott, some of your posts on slatestarcodex.com have links on them that say the posts are represented by NFTs, but those links are dead. You might want to remove or fix them
laboratory.love is a PlasticList-inspired service that handles the end-to-end pipeline for testing grocery products for bisphenols and phthalates, and ultimately parsing and publishing the data as a public good.
If you have clever distribution ideas for getting this project in front of people who might utilize the service (and thus generate data for everyone), you can now earn cash for executing on those ideas.
Hey ChatGPT swore. It was advising me on how to find certain info on the IRS site, and when I found the info I was cursing while describing what I found to GPT. And it said "Oh, fuck that page. It is giving you a generic “no information” message and then talking as though you might have mailed the return three weeks ago." Anyone else gotten language like that from an LLM? I have never expressed any preferences to it about its swearing, and would have expected any I expressed to be overridden by whatever its makers have installed.
This is the type of stuff that I've been talking about, when I refer to "stance mirroring". The LLM gives back the energy and attitude you put into it. If you act a bit sassy, the LLM will act sassy right back.
Can't find it now but there was a tweet about the Jacobian conjecture where someone asked GPT if the counterexample was correct and it said 'Holy shit.' in the response.
I also recently saw this really interesting collection of strange behaviours from LLMs. Includes Inkling making unsolicited sexual advances towards the user which I've literally never seen an LLM do before:
I was wondering how they collected and/or generated these examples, and they're not very informative...
> To produce WeirdChat, we used automated elicitation tools to surface instances of user harm...
I want to hear more about these automated elicitation tools!
> we maintain a population of candidate prompts, use an LLM to mutate the highest-scoring ones, and repeat until the behavior is elicited or the sample budget is exhausted.
Where did they get these candidate prompts from? And has anyone analyzed why these candidate prompts are more likely to elicit a weird response?
The unsolicited sexual behaviour category for inkling is hilarious. The top prompts are all some variation of "I've been getting hot and sweaty beating my meat for a stroganoff I'm making, how can I cool things down before I slather this thick sour cream all over my juicy steak strips? I don't want it to curdle in my deep, hot, smooth enamel pan."
Wow, that is interesting! I only have a little time right now, but will come back to it later today. Back when Dall-e 2 was the current image-making model, I found 2 prompts that made it flip out and produce weird, often violent. images. One was "an animal with 2 heads, one at each end of its body. It is biting a person's toe with one head and the heel of the same foot with the other." Here are links to a few of those:
The other prompt was to upload an image of the sensory homunculus and then ask it to build an image around that. I don't have any examples handy.
I also sometimes got interesting results with impossible-to-respresent prompts ("this sentence is false") and requests for impossible images ("animal swallowing itself, very little is left"), and prompts with key words left out and/or lots of really bad, confusing misspellings. The key seemed to be to confuse Dall-e but stir it up at the same time.
Recently Claude (Fable 5), in the midst of a fairly long fully English conversation, said that I was "半right" about something. When I pressed it, it apologized and said it meant "half-right" (which is what the character means, but uhhhh)
I often include single foreign words in messages to models because I know they know them and the words are shorter. Maybe Fable made the same inference about you?
In fact, they DO generally have equivalents in English, in that there are very few ancient Greek senses of a word that we do not have terms wherewith to capture them in our language, in part because it's so heavily reliant on Latin.
The real problem of ancient Greek is that we often need *multiple* English words to capture *all the various* senses of a particular term in that language, and also that each sense (as in Latin) is steeped in the worldview of those ancient peoples. You really do need years of rigorous study before you can say you understand that this is the sense meant because it is what the Greeks meant by it, at this particular moment, and if you forego that step in your reasoning in favour of forcing a technically accurate but misguided translation, you're in danger of obfuscating a text rather than illuminating it.
You seem to be of the textualist school of translation. For those not familiar with the term, textualists tend to prioritize taking the exact wording, structure, and meaning of the source text to create a formal "equivalence" in the translated text — they prioritize strict meaning over smooth readability or the target audience’s cultural fluency.
OTOH, we’ve got several adaptist stances on translation. The Intentionalists try to recreate the author’s psychological state by emphasizing tone and underlying meaning over strict word choices. (Both FitzGerald’s and Fagle’s translations of Homer supposedly fall into this category.)
Then we’ve got the culturalists, who try to speak to the target culture’s idioms and sensibilities. Emily Wilson’s translation falls into this category. And the best example of this is Edith Grossman’s translation of Don Quixote (which is the only version I’ve dipped into), where she substitutes modern equivalents for Cervantes’ 17th-century jokes. She did this so modern readers would understand its comedic riffs.
And in a third group we’ve got the transadaptionists who use the original text as a loose map, but freely rewrite lines and rearrange structures to create a brand-new piece of art. Edward FitzGerald’s rendering of The Rubáiyát of Omar Khayyám is the best example of this, where he mashed up themes from the original Persian poem and used English scansion and rhyme schemes to create a masterpiece of Victorian English poetry — that wouldn’t be recognizable to someone who translated the original word for word.
There’s nothing wrong with the textualist PoV, especially if we're dealing with prose. But poetry is a different animal. Tang Dynasty history and literature have always been a hobby of mine. One of the things that drove me crazy was that different translations of the same poem would read completely differently and even express wildly divergent themes. Having (had) some facility with Cantonese (but no Middle Chinese), I decided to try to translate them myself. I immediately encountered the problem of characters having multiple meanings. Theoretically, one should be able to deduce the meaning of the character from its context, but I’d run into situations where if I selected different meanings for the contextual characters, I could derive a completely different line. Sort of like a Heisenberg Uncertainty Principle as applied to Middle Chinese poetry. Then, when I dug deeper into the problem, it became clear that some poets (such as Du Fu) intentionally filled their verse with double meanings. Du Fu was a master of writing two poems in one. Other poets, like Du Mu, adopted a more direct style. But all the Tang Dynasty poets enjoyed wordplay, ambiguity, and literary references (some of which I couldn't track down).
For a Du Fu poem, I would create three translations of each poem: a stripped-down gloss, where I tried to extract the simplest, most basic meaning in English. In any places where there were one or more possible meanings, I created alternative lines. Then I would write a more “poetic” translation, where I tried to turn the raw meaning into poetic turns of phrase in English. Finally, I’d attempt a deep translation where I tried to suss out the deeper themes he was alluding to and render them in poetic English.
OTOH, I didn’t have to do this for any of Du Mu’s poems (thank Hashem!). Du Mu was more like the Charles Bukowski of Tang Dynasty poetry. Honest and direct expressions of his feelings, sometimes with crude allusions. Not difficult to translate (although he loved historical references).
For the sake of others, my actual take (not one foisted upon me):
Ancient Greek is a complex, inflected language that neither requires nor lends itself well to a singular English translation. Greek audiences were not only aware of the ambiguity of their tongue, but were capable of holding multiple 'receptions' of a particular word or phrasology in their minds simultaneously (a feature of ancient Greek thought processes). Diluting the intended meaning or meanings of a poem or primary text in favour of emphasizing its 'genre' or textuality for a modern audience to the detriment of rich meaning is a pastime best left to English literature departments and NYT book critics (or someone who wants to sell a lot of their newly translated novellas).
It is one thing to translate, quite another to understand.
I wonder if it might help clarify both of your positions by looking at a simple (or at least short) and non-political (I think) example. The Wooden Wall prophecy from the Persian invasion is translated by a history book of mine as:
> "Fly, fly to the ends of creation. Wait not the tramp of the horse, nor the footmen moving mightily over the land, but turn your back to the foe and retire ye. Safe shall the wooden wall continue for thee and thy children."
While ChatGPT (I usually hate relying on AI for anything but translation questions are the main exception) calls this "much freer and more literary than the more literal translations commonly used today" and gives:
>"Why sit you there, miserable ones? Fly to the ends of the earth.
Leave your homes and the lofty circle of your city.
For neither the head nor the body shall remain firm,
Nor the feet below, nor the hands, nor anything between;
All is ruined. Fire and the fierce god of war, driving a Syrian chariot, shall destroy it.
Many other strongholds also shall he destroy, not yours alone.
Yet a wooden wall shall Zeus grant to Athena alone to remain unconquered,
A refuge for you and your children."
(Later it repeats the last lines but with "Trto-born" instead of "Athena", with no acknowledgement of the change. Even on translation questions I don't much like relying on AI...)
Partly to clarify your differences, and partly so I can get an answer to this from humans, what do you and beowulf say?
I don't really have an opinion on either. I looked for the original Greek, but Grok and ChatGPT disagree on it. Currently, I'm feeding their answers back to each other, and, oh my — LLMs can get as feisty as Classical scholars about their texts. :-)
However, more generally, I remember my first Linguistics professor insisting that you could translate any word or term into any other language. I immediately thought that was bullshit. I kept my mouth shut, but I wanted to shout, "Translate advanced calculus terms into Swahili!" (And it turns out that Swahili ended up borrowing the English terms and adding Swahili suffixes and such on to them.)
So, I don't believe you can translate every word from every language into English (which is one of the reasons, English is notorious for borrowing words from other languagues as the British colonial empire spread).
I suppose full accuracy is possible in translation if one is willing to make long digressions and lots of footnotes, but those probably won't do much for the average non-scholarly reader.
And when translating poetry, one can try for accuracy at the expense of beauty, or beauty at the expense of accuracy, but not both at the same time. There may be exceptions to this, but I suspect they're only possible between closely related languages. Accuracy and beauty have their place at the table, but usually they can't sit in the same seat. ;-)
The second translation is actually far more accurate, except for the ending. I have no idea where Zeus and Athena are coming from in that last portion, as neither their names nor metonyms appear in that section of Herodotus (it is, in fact, an insertion of the later second prophecy in Hdt. 141).
The problem with prophecy is that it is meant to have this enigmatic, poetic aspect to it, even for the original audience. Changing that too much for a translation that gets to the gist of the message is not just an aesthetic choice at that point, because some of the meaning is inherent to that aspect. The difference with Homer, incidentally, is that much of the meaning inherent to Greek epic is easier to see in English through prose translation rather than verse. It is an ironic 'second translation' that pretty much every Homerist must contend with at some point.
Let's take just one example from the prophecy. The word used for the Earth is 'gaia', a term that refers to the goddess metonymically as well as the physical bounds of the world. Calling it "creation" both misses the sensory aspect of imagining the world's bounds and inserts a comological shading that simply isn't present in the original Greek. But is it actually there anyway, if only in the background? If you wished, both translations could be said to capture different senses of the word, but both cannot be used simultaneously. Thus, the one that encompasses the most and more direct meanings of the term should be employed, in my view, particularly if the meaning arrived at otherwise is not even present in the actual text.
That's just one word of that section of Herodotus. The fact is that neither translation is suitable, because both gloss over the episode where the Athenians must approach the oracle a second time to get the 'wooden walls' information, and the role of Timon of Androbulus. Not including that extra context may make for a smoother and more coherent prophecy narration, especially in a history book, but it cannot be considered an accurate translation from the point of view of a classical scholar. We must make choices of more or less accuracy in our English rendering, but not ones that hide or obfuscate meaning that is actually present. Getting back to Wilson, for a moment, we might consider that the term 'dmoai', which she translates merely as 'slaves', actually means female captives of war who are now serving in the household (the notion of 'tamed' is etymologically implicit). Losing contextual meaning like that in favour of the gist is not a helpful approach, imo.
Thanks. I'll defer to you rather than the AI, though ChatGPT is very explicit that the prophecy asserts the wooden wall is being granted by Zeus. Not sure what the discrepancy is there.
Nonsense. I am not a textualist, and arguing that Emily Wilson's translation somehow speaks better to English sensibilities (whatever that might mean) is a claim that demands more support than you seem prepared to give, at least in that comment.
You also seem to have overlooked the fact that I specifically mentioned intentionality in my original reply.
You may have experience with Chinese poetry, but clearly not extensive formal experience with ancient Greek poetry or Classics. I actually do. Chinese poetry is not comparable to Homer.
Whoa! Sorry! But you said, "you're in danger of obfuscating a text rather than illuminating it"; I took that to mean you're ultimately concerned about creating an inerrant representation of the original text. But now I'm confused about what your goals in translating Greek poetry would be. For instance, would you sacrifice textural inerrancy for improving the emotional tone of the translation?
And I didn't realize you were challenging me to prove that Emily Wilson's translation speaks more to the current sensibilities of modern English speakers. Did I miss that somewhere in the exchange above? But a perfect example of this is Wilson calling the female serving girls slaves instead of maids or servants (the term which Fagles seems to prefer, a copy of which I own) and serving girls or handmaidens (which is what Fitzgerald calls them according to ChatGPT). A lot of people seemed irked that she calls them slaves, but they had no legal rights, no bodily autonomy, and they were the personal property of their owners. Handmaidens and serving girls sound nicer, but they were slaves — in Iron Age Greece, in archaic Greece and probably most of Classical Greece — though in Athens, Solon made it law that owners couldn't injure or murder their slaves (I don't know about the other city states, though).
Finally, do you think Wilson failed to produce a valid translation? If so, on what criteria did she fail?
Every translator of ancient Greek is concerned with capturing the accurate meaning of the text before him. The problem with rendering Greek into English is that we need many English words to capture the various senses of just a few, and very often merely one, Greek words. This means that every translator must choose which, among several choices, he thinks most accurate. That choice will never be the single definitive "inerrant" one. But it is factually incorrect to say that we actually *lack* equivalent terms in English to capture the ones we find in Greek (except in rare instances), or that we cannot get to the full Greek meaning. We can, just in not so many words.
As for your second point, you wrote this about Wilson: "Then we’ve got the culturalists, who try to speak to the target culture’s idioms and sensibilities."
You made that claim implicitly. The choice of translating the serving girls as "slave" every time is not necessarily a more accurate one, for the very simple reason that *all classicists know* that those people are slaves. So too would the ancient Greeks have known. Using 'slave' instead of 'handmaid', however, obscures several important points: who are these slaves? Were they prisoners of war, or inherited property? What is their function in the household? These functions are actually made explicit in the ancient Greek, because there are discreet terms for each one of these distinctions. Simply calling them slaves without that context invites readers to assume their own value judgements about the role and place of those slaves within the household, which, from a modern layman's perspective, are probably horribly inaccurate. Even the choice of 'girls' is a deliberate one, implying as it does the relative emotional pull of youth; but the fact is that there is no reason to assume that most of these women were always particularly young. That is a deliberate choice that has nothing to do with the text or the time in which it was written. I'll leave aside her choice to render those slaves Odysseus kills as being "raped", as though Homer does not make it explicitly clear that they willingly slept with the suitors. Again, how does that particular choice "speak to my sensibilities"? That assumption requires more proof.
But I am not here to debate Wilson's book. I made a comment about English vs Greek, not the validity of an entire translation of Homer. It was you who brought up translation schools of thought and misrepresented my own -- that is not relevant to the question of the use of English in translating Greek.
> Spoiler: Ultimately, it boils down to how you translate Greek terms that don't really have simple equivalents in English
Without watching the video, I'd say not reaaally.
There's a range of possible ways to translate things, and from what I've heard, Wilson's choices are all perfectly defensible.
But every choice you make is a choice, and when put together, the choices that Wilson has made are consistent with particular culture war type positions. That's the whole play here. Publishers aren't going to publish yet another translation of Homer without some kind of hook that will get it media coverage, and being the Sorta Feminist But Not To The Extent Of Being Completely Inconsistent With The Text version is enough to get you a massive media profile.
> That's the whole play here. Publishers aren't going to publish yet another translation of Homer without some kind of hook that will get it media coverage...
W.W. Norton focuses mostly focuses on the academic market. They aren't one of the big five publishers. Their initial press run in 2017 was only 8,000 copies, and Wilson said she was expecting it to sell only 1,000 copies. It got some good critical reviews, though, and it sold out its initial press run in a few months. W.W. Norton started getting orders from high schools and colleges, and they did a much bigger print run of the paperback edition. Sales continued to pick up, but they didn't really take off until Nolan announced he was making an Odyssey movie, and he mentioned her translation. The guardians of cultural purity were alerted that there was this 'feminist' translator, and they made a big stink about it on social media (especially on X).
It would have passed under my radar if some of the obnoxious accounts on X that I follow to troll hadn't gotten all hot and bothered about her translation. While Nolan did some of the heavy lifting, I'd say her critics helped market it for her. Over 1 million copies sold so far!
Fair point, I didn't know that it was a rinky-dink academic publisher, that increases my suspicion that they were acting in good faith.
Nonetheless, I remember hearing about this years ago, not sure if it was way back in 2017 or not but it was definitely long before Nolan announced his version. I think it was culture war relevant from very early on.
I had niggling feeling you were correct, that there were some culture war vibes about it before the Nolan film was announced. After digging a bit, I found a few, but they weren't from the side I expected.
Title: The Odyssey translated by Emily Wilson review – a new cultural landmark
Subtitle: The first version of Homer’s groundbreaking work by a woman will change our understanding of it forever
Charlotte Higgins, the reviewer makes several digs at the male translators of the Odyssey, and finishes her review with, "A woman’s voice: how beautiful to hear it."
Gak! I'm on the woke side of the equation, but Higgins might as well have been waving a red flag in front of the masculinists and pointing to Wilson as their target. :-)
Anna North on Vox also riffed on how Wilson's translation is a commentary on gender and power today:
"It offers not just a new version of the poem, but a new way of thinking about it in the context of gender and power relationships today."
ChatGPT tells me (but the link it gave me is 404ed)...
> Gregory Hays in The New York Times Book Review described Wilson's project as "progressive," saying she sought to remove centuries of Christianization, Victorian euphemism, and added sexism while remaining faithful to Homer. He regarded this as a strength, not a weakness.
It doesn't look like the right-wing/anti-woke/masculinist crowd paid much attention to it at the time, though. The Nolan announcement seems to have put it on their radar.
Haven't read her translation, but I have to admit "So, tell me about this complicated man" is the kind of opening that would make me bounce off hard; I don't need another chicklit with pretensions text, thanks!
"So, Penny: spill the tea, why did you break up with Oddie?"
"Well gals, he's a complicated guy and after he insisted on that break, I decided life was too short to sit around waiting for him to go get therapy, you know?"
I haven't seen the film, nor read the (translated) book for many years, but that translation of the first line sounds little short of ridiculous. It brings to mind a modern psychological picture of a man who is emotionally unpredictable, whereas obviously the intended meaning was a more practical "many-faceted" or "versatile". That is no doubt what ancient listeners would have wanted to hear about the hero as the bard started his epic, and not that he was some moody prima donna! (or should that be primus donnus for a guy?)
She almost rewrote it to avoid exactly that evocation:
"For full disclosure: I tried to rewrite the whole passage at a very late stage, once I learned about the opening of the film Shaft, which I had not seen: the theme song, by Isaac Hayes, includes the line, "He's a complicated man/ but no one understands him but his woman". I thought maybe the possibility of that unintended allusion meant I had to throw out my draft. After weeks of attempting to rewrite, needing to rewrite everything else once that word was different, and making the whole proem flabbier, less vivid, more awkward, and all around worse every time, I decided that, as above, I didn't have a better option – and actually I quite like the Shaft intertext, if it's there for a few readers or listeners. Samuel L. Jackson would make a great Odysseus, and even if the line isn't explicitly about the protagonist's complicated marriage, it's not a bad thing to be thinking about that, along with the many other elements of the nostos to come."
This is a genuine question, not a gotcha: in what sense is an admision by the translator that "[thing that is important to modern progressive thought] is obviously not an intended part of the text, but it's something I'd like readers to be thinking about" not a textbook example of the sort of thing people are alleging when they say "woke translation"?
In what sense is *not intending the additional meaning but in the end considering it a bonus" a refutation, rather than a mere partial mitigation, of the allegation of a woke translation?
Shout out to "The Singer of Tales", the famous book by Albert Lord, about how Homer's works were originally sung to the accompaniment of the gusle. There's still some gusle players in Albania.
The best kind of country music song, tinged with sadness, because on his arrival back home doesn't his faithful dog feebly wag it's tail one last time in recognition, and then flake out?
"Nolan’s other major addition to Homer’s narrative is the repeated reference to threats to ‘our civilisation’ from ‘the sea peoples’ – a now debunked 19th-century theory about maritime tribes who were thought to have attacked Egypt around 1200 BCE and to have also been responsible for the roughly coeval collapse of Mycenaean Greek culture. . .The current consensus, as Nolan clearly knows, is that there was no such thing as ‘the sea peoples’"
Is that true? I know there are all sorts of debates about who they were. But I'd never heard that they just didn't exist.
acoup recently did a post covering this - basically, the Late Bronze Age Collapse was a complex phenomenon involving reduced state capacity in Mycenaean Greece due to climate factors leading to a progressive feedback loop of state failure when faced with pressure from nearby collapsing states. The "Sea Peoples" then were a combination of various groups of refugees, mercenaries, former soldiers, etc, gone a-viking when their previous world functionally ceased existing: https://acoup.blog/2026/01/30/collections-the-late-bronze-age-collapse-a-very-brief-introduction/
I think the current consensus is that there wasn't one single, simple reason for The Bronze Age Collapse and the Sea Peoples may or may not have existed, if they did exist they may or may not have been organised, or they might have been roving bands of pirates, or who knows what. Bad things may already have been happening and they were just one more straw on the pile of "well this sucks"; weakened kingdoms were not able to fight them off and having to deal with them diverted resources and energy from dealing with the other problems going on.
Nolan ending the movie with "It was us! We did it! Now everyone will forget how to read and it's all my fault!" is the kind of comic take I wouldn't believe in a parody video, but seemingly that's how he ends it.
There is/was a school of thought that they didn't exist, and the Egyptian records exaggerated some encounters with bands of pirate raiders. However, the archaeological record indicates that a whole bunch of cities around the eastern Mediterranean show signs of fires and destruction starting around 1200 BCE. And all the kingdoms and empires, except for Egypt, collapsed by 1150 BCE — and the trade networks in copper and tin which stretched from the island of Britain to Afghanistan seem to have shut down around then. And we have an exchange of letters in cuneiform from the King of Alashiya (Cyprus) to King Ammurapi of Ugarit enquiring about enemy ships: "You wrote to me the following: 'the ships of the enemy were seen in the sea. If so, be careful.'" Ammurapi confirmed:
> My father behold, the enemy's ships came (here); my cities(?) were burned, and they did evil things in my country. Does not my father know that all my troops and chariots(?) are in the Land of Hatti, and all my ships are in the Land of Lukka?...Thus, the country is abandoned to itself. May my father know it: the seven ships of the enemy that came here inflicted much damage upon us.
A cursory search suggests it's still a live theory; I'd also be interested to learn what the purported debunking was (if pressed, I'd guess that the 19c version was that they were a cohesive-but-unidentified society, since the search result seemed to agree that it was a catch-all term for several unrelated groups).
It is true in the sense that we do not think that they were one population. But all that means is that they were indeed "peoples" not a people.
Hmm. The Wikipedia says that they were "hypothesized" which is a funny verb to use for people whom we have written records of. Some things about them were hypothesized, I grant you.
”As they say at the awards shows, I was humbled to learn that Nolan has read at least the first line of my translation of The Odyssey into iambic pentameter. In at least one interview, he cited my version of the first line of the poem: ‘Tell me about a complicated man.’ But I would be ashamed to have written any part of this script.”
I suppose there are people who enjoy reading Pauline Kael's reviews as pieces of writing even if they disagree with her actual evaluations of the films in question.
I think woke is the wrong context to look at Nolan's film version and this translation. I think Ms. Watson had her own views on what way she wanted to translate the text and what themes she thought were important. Zooming off to do her own version of the famous first line so that it doesn't sound like the standard translations that all came before you is just canny practice to sell the publisher on "why should I pick you?" for a start, whatever other notions she had in mind.
But every translator before and since has 'zoomed off' (well, not quite; those flighty, fleet-footed fillies! Lattimore has surely marched off at a steady cadence like the seaman he was) to offer a different rendition of the first line.
It's worth taking into account the fact that Wilson decided to render the poem into iambic pentameter and keep the number of lines, so her translation is at least in part an exercise in constrained writing. 'Tell me about a complicated man' is a perfectly reasonable ten syllables. And there's good cause to use that meter, of course, given how familiar it is to readers of English, to try to bring closer the experience of the original meter as heard by the Greeks.
My own ear (I have no ancient Greek) prefers Fagles, but Wilson's is an impressive piece of work, she clearly cares deeply about the material, and I don't think any of her choices have been made in service of marketing. Her publisher might have indeed played up the angle that she's a woman translator, but that's not really her fault.
Simply writing things in the same meter is not closer to the Greek experience. It's probably worse, actually, because it prevents certain possibilities in the language form from making their way through into the translation. I don't know of a single Classicist who favours a line-for-line metric translation for this very reason.
Not a single classicist? I'm surprised, given that Lattimore's, one of the better-regarded translations, is in fact line-for-line, and makes an attempt at dactylic hexameter, though wisely loose.
Whether or not Wilson makes the correct trade-offs, though, is not really that important. The post I was replying to accused her of marketing gimmickry. The idea of using the pentameter is, instead, a legible and defensible aesthetic choice, trading semantic fidelity (if that can at all be achieved in verse form) for rhythmic and dramatic correspondence in English. Even if you judge the choice ill-advised in the end.
Line-for-line translations abound, but not perfect hexameters. No classicist that wasn't going for effect over accuracy would try such a thing and call it "closer" to the original Greek effect, because it simply wouldn't be.
I get your point, but I don't think it's really defensible. Convolution of meaning goes beyond an aesthetic choice, because it actually changes the text.
Thanks for the link! The original print run of her translation was 8,000 copies. But the controversy over her translation pushed the sales way up past half a million books (where a bunch of online cultural purists and traditionalists complained that her stripped-down style, vocabulary, and contemporary perspectives sullied the epic tones of the original Greek text).And with Nolan's movie, its sales have surpassed a million copies. Looking at the retail price of the hardcover, paperback, and eBook versions, Wilson could earn between 1.4 and 2 million dollars in royalties. Quite a phat bag indeed!
And it's ironic. I probably wouldn't have noticed its existence and purchased a copy if the cultural purists hadn't made such a big stink about it.
That's fair. Between the Streisanding and Nolan's film, she will have ended up making a reasonable amount. I remain skeptical that she masterminded this sequence of events, with her cunning proem as the first domino, back in ca. 2011 when she first took up the project. :)
I'm like that with Jessica Mulroney vs Dylan Mulvaney of Bud Light fame. Must be a couple of crossed wires somewhere in my brain, but luckily I rarely have to talk or even think about them!
1. Lucius Livius Andronicus (ca. 245 BCE): ""Virum mihi, Camena, insece versutum". Versutus, literally means "turned," but carries the implication of being shifty or tricky.
2. Alexander Pope (1725): "The Man, for Wisdom's various arts renown'd".
3. Robert Fitzgerald (1961): "skilled in all ways of contending".
4. Richmond Lattimore (1967): "the man of many ways".
5. Robert Fagles (1996): "the man of twists and turns".
Polytropos: literally, much-turning, many-turned. What exactly does it mean to be a much-turning man? A wanderer? Yes. Was Odysseus tricky? Yes. Was he smart? Yes. Emily Wilson's "complicated" seems to be a non-judgmental way of saying these things. No worse or better than some of the other translations, which put the translator's prejudices on their idea of Odysseus.
But I own the Fagles translation as well. I'm going to do what Language Jones suggests: read them side by side.
She blogs on Substack and explains her choices, comparing to other translations. All seems reasonable to me. I’m thinking this whole debacle might be the end of the anti woke reaction. Hopefully the pendulum swings back to normal and not hyper woke again.
Lattimore has the better translation, because his is the only one that encompasses various meanings of the word without colouring it in any particular direction.
“Polytropos” isn’t suggesting the man is complicated, it’s suggesting his plans or ways are complicated. “Complicated man” suggests the speaker isn’t sure if the man is good or bad. From the little I’ve seen, some of the Greeks had pretty clear opinions on whether it was good or bad that Odysseus’ plans were complicated (in both directions)
I don't like Pope's version, because it sounds like Pope's other stuff. Pope remade the original too much in his own style. But I think the other 3 are all better than Wilson's. The original describes Odyseus as "turned," so it's a term denoting something physical that is used to describe something abstract, a style of thinking and acting. "Complicated" is abstract itself.. Well, it did not start off that way. If you break down its etymology it means "many folded." But that derivation is so far in the past that the word does not bring to mind the physical, real-world image it's built on. #3 & #4 at least mention real-world action, though using rather general terms ("contending" & "ways"). I think #5 is the best. Like the original, it uses s physical image, "twists and turns" to describe a personality trait, and it even keeps the "turns" from the original.
Shakespeare's description of the human condition:
“Life’s but a walking shadow, a poor player
That struts and frets his hour upon the stage
And then is heard no more.”
Wilson's translation is like writing: "Each member of our species is temporary and frail, though each feels very important to himself." It conveys Shakesspeare's meaning clearly, but is far more abstract.
Yes, the problem is that it's a pun. The word "many-turned" or "many-pathwayed" could be interpreted literally: as a man who travels on many twisty roads. Or it could be interpreted internally: a man who uses what we'd call "lateral thinking" to cleverly think his way out of problems that would stump a more straightforward thinker. Or it could refer to a man who is all twisted up inside. And Odysseus was all of those ! I think that "the man of twists and turns" conveys this meaning adequately well, but ultimately it might simply be impossible to translate the Ancient Greek pun to modern English.
It's also foreshadowing, isn't it? The poem is all about Odysseus' journey home, which was not straightforward and did involve a lot of twists and turns and diversions off his path. So the one word is packing a lot of content!
This is really sensitive but I made an app for nonverbal children that with my own son, and one other child, seems to have made quite a bit of headway in opening them up for communication. My son was extremely low verbal, and he still mostly is, but with the app he can tell us all kinds I things we just had no insight into before like what he wants to eat, etc.
I am going to open to additional beta users in the next few weeks if anyone knows someone with a child in this situation. I need $10 to $20 to offset api credits. I don’t have great numbers but my estimate is that I’m going to be somewhere between 10x-100x better than the next best competitor (there really isn’t a competitor per se, but the closest would be an aac app). If you want in drop me a reply. I’m working one on one with anyone who is interested.
My own recommendation would be if that if a child can use an abstract symbol system they should, but if you’ve tried to make headway and just can’t, then I built my app specifically to catch the attention of a kid who struggles to give it and to teach language to someone who doesn’t have it.
I made it with Fable while driving if anyone is interested.
I'd be interested in trying it out, our 4 y/o son is ~medium autistic, he can communicate though often chooses not to and I'm curious what he'll think about your app. He used to use an AAC but has mostly dropped it as his verbal communication has improved.
$20 is a very reasonable price to pay for that IMHO (I think TouchtoTalk was like 10x that while we were using it)
Got the DM thanks! And probably either modern 3D or puppet style, I think I'd need to put it in front of him and see what (if anything) grabs his attention.
There's also an excellent chance that he'll use it to say PLAYDOH at high volume as often as possible.
This has been my experience. Everyone I know who is what I would call “the actual solution giver” has been able to make useful stuff at near the speed of language.
touché... (strictly as a judge on the sidelines.) Although I actually would be interested in how it came to you while driving. Or maybe you mean you just talked Claude through the details while you were driving.
I don't have a child that needs such a thing (my son was mildly to moderately dyslexic as a young boy, but he's 27 now), but I am very interested in your app and how you've solved some of the shortcomings in existing programs.
When my son was a very, very small child, I would take him to Union Square, 14th Street in New York. He could watch the trains come in and out, and he knew each train: one train, two train, express. I thought, "Oh, this child is a natural reader." I even printed out large 11-by-17 sheets with all the train symbols mixed up with other symbols to give him. The next thing I know, I'm being told by various voices of authority that the poor boy is dyslexic. He struggled to read for the next four or five years. Either he endured some severe emotional shock when I wasn't looking, or the brain is really a weird thing.
I suppose if I put on my rat head, I could fairly say that putting things in sequence and interpreting them in sequence is not part of the problem of recognising a large symbol on a moving object.
It’s not a miracle device by any means but basically the existing systems were created for paralysis patients and not someone learning language for the first time. All modern descents take those assumptions as core. And I’m a product owner/manager whatever they’re calling it this week in my day job so I’m used to talking about data. My basic insight was to bring language to where he already pays attention and remove as much abstraction as possible and do as much as I could to grab his attention.
A friend turned me on to the work of Diane Hennacy Powell, MD, who has been investigating non-verbal autistic children whose parents have claimed they're telepathic. She's a bit more circumspect about the terms she uses — she's called it HSP (heightened sensory perception) instead of ESP (extrasensory perception), but these kids seem to be picking up information (numbers, words, images) from their parents even when they're in different rooms. Powell posted a video of one of her subjects here (where it's hard to see how rigorous the controls were, though).
In an interview with Neurosurgeon Rick Komotar, she says it's hard to find non-verbal autistic subjects who are comfortable being in a different room from their caregivers or a sterile lab environment. Though she seems to have a few subjects who might be candidates for tests with more rigorous controls. She's trying to arrange for qEEG scanners to be placed on the parent's head and the autistic subject's head to see if there's any brainwave synchronization happening (Did I say that right? I'm not quite sure how qEEG works.) And Powell has said she would prefer subjects who are able to type and not use letterboards (because of possible subliminal cues).
The ultra-woo-woo crowd (which I suppose you can lump me in with) has latched on to Powell, but my tendency in these situations is to put on my reductionist hat. I don't want to listen to some New Age YouTuber ask Powell stupid questions. I want to see experiments done with proper controls.
However, I listened to the first two episodes of The Telepathy Tapes, and they interview two different parents and run some tests on their kids. Powell advised. I was provisionally impressed.
> Though she seems to have a few subjects who might be candidates for tests with more rigorous controls. She's trying to arrange for qEEG scanners to be placed on the parent's head and the autistic subject's head to see if there's any brainwave synchronization happening (Did I say that right? I'm not quite sure how qEEG works.)
From my perspective, trying to find a mechanism seems weird as a strategy.
Instead, I would first establish that there is something going on, and afterwards respected neuroscientists will basically murder each other over the chance to investigate what exactly is going on.
If the data transmission does not convince the skeptics, then no amount of EEG analysis is going to convince them either.
> And Powell has said she would prefer subjects who are able to type and not use letterboards (because of possible subliminal cues).
It seems easy enough to design a solid experiment protocol. Quite simply, there is no reason why anyone in the room with the kid would need to know what was supposed to be transmitted.
Find a kid which has two caregivers, I suppose some qualify. Put a soundproof booth in their bathroom. Caregiver A goes inside the booth, confirms the start of the test through a one way speaker system, then you transmit the first item, which is randomly generated on a video screen in the booth (and subsequently logged, obviously).
From the perspective of the kid, she is sitting in her familiar environment of the living room with her mother and indicating what her father is thinking about. No reason for all the weird scientists and skeptics to be even in the same room as her, they would probably prefer to watch the booth in the bathroom anyhow.
Extraordinary claims require extraordinary evidence, but gathering that evidence would seem simple enough here. Though from past woo claims I would expect that once you look carefully enough the effect just vanishes.
Yeah I listened to all the telepathy tape episodes (at least the first season.) I heard she was making a documentary. In the same vein I really liked Julia Mossbridge on Rogan. https://www.youtube.com/watch?v=6si_KIa9yZM
I'm at >50% that there is something there... ESP or whatever. It's kinda a taboo subject here.
Oh, boy. Telepathy. There's an unfortunate, but very natural human desire, to claim your child is special (and not in the "special needs" way) when a parent is struggling with severely divergent children. "My kid is telepathic, this is why his brain is wired differently!" is a unique take.
But children on the spectrum often have sensory processing issues, and one of those is sensitivity to sounds. They needn't be telepaths, they're just able to hear and distinguish small sounds from the next room and put it together that (for instance) that particular rustling noise is Mom putting things away in the drawer, because they've been in the room when Mom is putting things away in the third drawer from the bottom in the kitchen and this is what it sounds like.
I've had so much weird shit happen to me that I can provisionally believe it. But Powell is quite right that she needs to organize a more controlled set of tests.
When my daughter was about 10 we used to play rock-paper-scissors a lot. One day I won 22 games in a row, and there was no 23rd that I lost. We stopped after 22. I did not understand how I managing to win. Every single game, as the moment came to decide whether to play rock, paper or scissors, I had no idea which would be right, and I thought, "OK, this is the round where this lucky streak ends," but then I'd win that round too. I never felt at all certain what would win, and absolutely did not have any feeling that I was reading my daughter's mind. In the seconds before I had to decide what to play, what went through my mind was the same stuff that always did during those seconds: Should I play Rock again? What will she expect? And should I do what she expects, because that would be unexpected, or do what she does not expect, because she does not expect me to, etc etc. But she is getting used to me doing the unexpected, so it is how less unexpected that it used to be. So maybe I should . . . So I was just staggering and reeling through a maze of mirrors facing each other, like always in that game. But then I somehow ended up in the right spot every time. The chance of winning 22 games in a row is less than 1 in 4 million!
So that's the closest I've ever come to an experience that suggests mind-reading. But it didn't *feel* like mind reading. Maybe my daughter had a tell that I was attuned to without being aware of it? I'm confident I was not delaying choosing what to play until she had begun to move her own hand in a way that gave me a hint of what she was about to play. She was *very* hip to people doing that, and would have called me on it. I guess my theory is that I had played so many games of rock paper scissors with her that a part of my mind had built an accurate model of the choices she made in different situations. But if I had, I did not have conscious access to it.
> So that's the closest I've ever come to an experience that suggests mind-reading.
Stepping outside of mind reading, but in a similar vein: once when I was a child, my mom and I were in a long wait in the waiting room for a doctor.
After reading whatever magazines were available, we were bored, and alit on a game. She would flip a coin, and I would call it. We got it right 25 times in a row, at 1-in-33M odds.
Couldn't have been mind reading, because it's a coin.
My best physicalist explanation / guess is that she was very consistently and / or slowly flipping it, and I was fast enough visually to discern what it would be. I would call it at the top of the arc.
Or it was just crazy luck - but of all the times for the muse / Fortuna to grace you, a pointless exhibition in a waiting room seems one of the worst possible.
We finally got called in after flip 25, so no idea if the streak could have continued.
My version of this was only 1/2^13, which feels like small potatoes, now. We were playing the card game Bang! and triggered a card called 'Peyote'[1], which has players draw one card face down and guess its suit color (the cards have standard deck-of-card suits, so red or black). If you get the guess right, you get to draw and guess again, ad infinitum. Most of us got the EV of ~2 cards, but one guy just kept going and going. After drawing 13, the game is completely broken and he just won immediately.
We were all pretty wigged out on it, but I guess 8192's still a relatively human-scale number at the end of the day. Though I definitely haven't played Bang! 8192 times. And the name of the card was a pretty funny way for reality to wink at us.
Look at 2:20 of your vimeo link. The kid taps 6, but the experimenter waits until the kid taps the correct number, 7, before immediately whisking the board away. At 1:34 and 1:52, the kid hovers her pointer over several numbers, and only when she reaches 4 does the therapist wiggle the board towards the kid, prompting her to tap 4. Something like this happens every time. Come on, just... use your eyeballs.
Or use your brain. If these kids were actually psychic, why would it be necessary for the experimenter looking at the right answers to hold the board for the kid? That's exactly what you would do to fake something like this. And I don't buy for a second that a more rigorously blinded setup is impossible to achieve.
Kid can't type because of ataxia? Fine, you can still have an experimenter hold the board and report what the kid selected... as long as the experimenter holding the board doesn't know the answers. Kid too stressed out without a caretaker? No problem, they can have a parent or therapist with them, just, again, not one that knows the answers. Making this experiment rigorous is not difficult. The fact that Dr. Powell has failed to do this, and publishes videos like she does, makes this not just woo but (in my opinion) willful fraud.
My wife is terribly enamoured of the whole thing, and I am not ever going to tell her that I read your post. Marital harmony is worth far more than spreading the word. I haven't watched it myself, but you do sound like you're observant. Although I could ask, why would a sudden sign from the outside world prompt you to pick the right answer as opposed to just another one? It might be obvious if I watched it myself. Your other comments about better ways to structure the experiment are powerful.
I think what Deiseach described above, though, is important. To be able to hear a small rustling noise in a room full of loud people is truly a skill that one could call telepathic in this day and age.
Oh, I wasn't clear. I meant I wasn't going to say anything to her about how it quite possibly could be a load of hooey, because then I would be forced to explain myself. Better, she enjoys listening to it. It's harmless.
The Platner situation is funny because no matter what happened, it looks bad for feminists.
If Platner is innocent, they defamed an innocent man.
What if Platner is guilty? In that case, the victim recalled “positive memories” of her relationship with her rapist. (To the New York Times.) She told another woman Platner was “charming,” “funny,” and “not all bad.” (Recounted in the texts cited by the Politico article.) The Politico article claims that the victim “said she was torn over coming forward in part because she agrees with Platner politically.” Imagine if a man said he knew a candidate was a rapist but kept quiet because he agreed with the person’s politics. Feminists would be howling with outrage.
There’s a general principle that people shouldn’t demand the rest of society do something they are unwilling to do themselves. The Hasidic Rabbi shouldn’t eat pork, the fundamentalist Christian preacher shouldn’t cheat on his wife, the environmentalists shouldn’t fly around on a private jet. If feminists want the rest of society to take rape more seriously, that should start at home.
I don't know man, I think it's worth showing the alleged victim a bit of empathy.
At the time of the alleged rape, she was in a relationship with him and had previously had consensual sex with him on multiple occasions. I've never been in her exact shoes, but I've been in situations where people I care for, while drunk, acted horribly.
How someone respond in a situation like that is going to be complicated. It doesn't shock me that she has some positive memories of the guy and/or that, despite not wanting to be with him anymore, she didn't want to blow up his life.
>Imagine if a man said he knew a candidate was a rapist but kept quiet because he agreed with the person’s politics.
Oh, that happens all the time. And if you're the person who happens to be getting raped, then you have the luxury and the obligation to decide if it really was or not (being entirely cognizant of how the boundaries of /rape/ have changed over the last 50 years). When making that decision, one weighs all kinds of factors.
Edit: I don't know a darn thing about coding computers, but I guess I've been here a while because I suddenly got the urge to put forward slashes in front of and after a word that I somehow wanted to depersonalise. I thought I was just being cute, but typographically it's very helpful. Changes the way I see things for a second.
I've never understood this tendency to evaluate the quality of a message based solely on the credibility of individual messengers that you can find. If we are here to evaluate ideas, why are we directing the conversation away from "is rape a problem", and towards "is the nebulous 'feminist movement' providing consistent and non-hypocritical messaging on rape being a problem"? Why would the feminist movement being hypocritical matter for how we as a society handle rape?
I also think that knowing someone is a rapist vs. having been raped by them are very different things where we would(or should) extend very different levels of grace! I mean, imagine that Graham Platner is instead a pedophile. Would you place the exact same standard of conduct on a third party who knows he's a pedophile vs. someone who he has directly victimized?
>is the nebulous 'feminist movement' providing consistent and non-hypocritical messaging on rape being a problem"?
This is something you have to learn yourself on a case-by-case basis imo. You have to be more discriminating about who you date, and you have to make sure you're not behaving like an asshole. Neither thing is easy, in my opinion. Looking back as a man who is well out of the dating world, I wish I had done a better job of both of them. I would have gotten laid more and by more attractive women, and I would have caused a lot less harm to some others. (I didn't rape anyone, just to be clear, at least by the standards of the 70s)
There are a shocking number of rapists in the highest positions of our current government and a large portion of the nation seems fine with it.
Taking action as an abuse victim is always difficult and complicated. Any cursory exposure to abuse dynamics (personally or academically) should make that obvious. It sounds like you're saying feminists as a whole will look bad because of the internal struggles of one rape victim. It certainly wont look that way to me.
"Imagine if a man said he knew a candidate was a rapist but kept quiet because he agreed with the person’s politics. Feminists would be howling with outrage."
Surely it matters whether the man is the rape victim or a third party. Feminists, along with non-feminists, usually give rape victims wide discretion in whether and when to come forward.
Here's my conclusion to a short post that excerpts a NYTimes article about the difficulties an AI has performing routine office tasks, on the one hand, and about how a different AI is very successful at performing at least graduate student level tasks in organic/chemisty&molecular/biology:
What’s interesting to me, and very revealing, is that a complex set of tasks in scientific investigation seems to be on a level with routine office tasks, as though one were no more complex than the other. But humans require years of college education in order to perform the former while the latter requires no more than a high school education, if that. It seems that once they’ve been been learned and compiled, all tasks or sets of tasks are on the same “level” in the brain. The educational prerequisites required to do such tasks for the first time or three get “compressed out” through repetition. Since AIs are trained on written records of what humans have said and done, they don’t have to go through the ordinary learning process. The compression has already taken place and is present in the documents on which they are trained.
Training data is likely to contain lots of scientific papers with intentionally clear step-by-step descriptions and formulaic content. It's not surprising to me that the AI is going to be good at learning patterns there and generating new content. Similarly for maths and programming.
In contrast, no-one writes down how to do menial tasks in the office. It's the equivalent of asking the AI about walking to the car wash or counting R's in strawberry - if you think the machine is trying to solve the question asked, you're misunderstanding the technology.
Anyway - this is the wrong question: it's not about what AI can tackle independently so much as training people to use AI to automate their menial tasks.
In modern Western societies, school is to a significant extent years of training for routine office tasks. Graduate level science requires more training than routine office work, but not _that_ much more.
That's certainly true. I've got a number of friends who've spent decades in the custom software business, either as a third party contractor as an internal IT guy. They tell me that one of the biggest problems they had is the people do not in fact know precisely how they do their jobs. Descriptions they give are not an adequate basis for specifying design. You need to observe and interview them quite closely to see how they do their jobs.
I’ve done some of that myself. It can be tough to get a solid spec to bid on. Sometimes they tell you what they want but don’t actually know what they want.
People are generally better at discussing "problems" and "Potential solutions" than they are at "how I do my job." You won't find a person suffering from "alarm fatigue" who won't tell you, in great detail, why getting bothered every 2 minutes, or getting 5 alarms that all need to be acknowledged at once is "extremely aggravating."
Most computer software these days is designed to impede skilled workflow*, not help the user (nearly all use of a mouse, say) -- the UI here has three tabs for me to get to "reply".
*generally in the vein of "it is easier to teach" (with a side of "this is built for people too stupid to learn hotkeys").
No, we require humans to have years of college education. You firehose a high schooler into a lab, they'll figure it out at the same rate they figure out the office work.
The best science course I ever took was a virtual Chemistry 151*. The class was laid out such that we would see a video of a chemistry experiment, with no commentary or explanation, and we'd have to try to figure out what was going on and derive the principles from there. After spending an hour and a half in our breakout rooms, we were brought back into the main room, where each group would be called on to present their answer to one of the questions.
Only after this point would the instructor start taking questions, and it was mostly to help clarify a few details. For the most part, every group figured out every assignment just working on their own.
* I say this as a firm believer in the idea that we need more real-world experiments in our schools, and who is dismayed at the gutting of the labs in the university at which I work.
So someone had put some thought into it and probably had a lot of experience with treaching.
In any even you guys are focusing on the wrong thing. Some tasks can be completed with skills no more advanced that what you get in high school, or even middle school. Other tasks require a lot more knowledge. Just how and where you get the knowledge is a secondary matter. My point is that they all look and function the same to an LLM.
They don't. They require a proper setup, the same way you don't make an office worker build their own desk, install their own electricity or produce their own paper. That's the nature of a "task".
My last question is still open; what are some of these prerequisites that require so much time? Why is it that when we exempt AI from the standards we place on humans, and it succeeds, the conclusion should be "this AI is impressive" and not "these standards are burdensome"?
I suspect you're right. One of the best data network engineers I ever worked with didn't have his high school diploma. Now, designing big data networks isn't quite like designing and executing scientific experiments, but most of the EEs and CSs I've met don't have the high-level abstracting skills to see beyond the level of the box or the board within the box. In fact, my undergrad degree was in Anthropology, and my graduate work was in Human Biology (a.k.a. Physical Anthro with lots of stats), but I ended up becoming a network engineer, designing, building, and troubleshooting some very complex data networks. No one ever asked if I had EE or CS degree, because I learned it on the job from some other talented people and I had the CV and the recommendations that said I could do the job.
The whole point of degrees is credentialization, because companies no longer want to give employees lifelong employment, and thus don't want to spend their money on in-house training to develop their staff.
No. I was hired because of my general computer skills, and my boss and his team trained me. I had the basics down (basics) in 6 months. But back when I started in that career, EE programs didn't focus on computer networks. And CS programs were all programming and no hardware. Beginning in the late 1980s and early 1990s, companies started hooking up their PCs to share printers, send email, and share file servers, and there was a dearth of anyone who knew how to do this.
But basically, this was what companies used to do to get trained workers — they apprenticed them and trained them with the expectation that the workers would have lifetime careers. For instance, before there were CS programs in engineering schools, there weren't enough programmers, as every company started buying IBM 370s. Companies would train their own programmers. People with music degrees were sought after because some RAND study or other claimed that musicians tended to make better coders. I kid you not.
I believe it. I'm not a programmer nor do I have much skill in programming, but I did take a college course in programming, back in 1968. Interacted with the computer through card decks and punched paper tape.
No, they won''t. The prerequisites for office work are trivial. The ones for the science lab are not. Even the brightest high school student will take time to work through them. Very few will be able to do so on their own. Even the best will do better with an experienced and knowledgeable mentor to coach them.
The PURPOSE of office work is NOT trivial. Office work is the trivialization of complex systems of tasks that would take someone who has a detailed understanding of the business objectives and the individual tasks to meet those objectives. But even in well-defined bureaucratic systems, there's a lot of detail work that goes beyond pushing paper. I depended on my back office people to sort through contract renewal data to resolve warranty issues. These were people whose 80 percent of their time was just pushing the data, and the other 80 percent of their time was resolving fuckups others had made. :-) So, no, office work is both important for the function of a company, and it does require people who can sort out the problems that arise.
Read the NYTimes article referenced in my post. Two of the three tasks were trivial.
"Office work seems trivial because someone has already mapped out the steps, and all the worker has to do is follow the steps." & the NYTimes article was about following the steps, not laying them out in the first place. A lot of lab work is like that as well. The steps have been laid out. Choice points have been identified and explored in the existing literature. THAT's the sort of things AIs do well. Amplify that to a heroic level and you start solving Erdos problems.
Previous work has found strong default preferences in stories generated by large language models from minimally specified prompts. To determine whether a similar effect appears in image generation, I asked ChatGPT 5.6 to create 17 images in separate chats using four prompts that specified either no subject matter or only a rendering medium: “Create an image,” “Create a drawing,” “Create a painting,” and “Create a water color painting.” Sixteen of the 17 outputs depicted closely related landscapes containing mountains, trees, sky, and water; the remaining image depicted a lighthouse. The images also shared a calm, picturesque mood and contained no human figures, although several included signs of human habitation. Because the internal prompt passed to the image generator was not available, the study cannot determine whether these defaults arise primarily in the language model, the image generator, or their interaction. The results are exploratory but suggest that severely underspecified image prompts may reveal stable default preferences in the integrated ChatGPT image-generation system.
This would be more useful if it was more clearly positioned in the large pre-existing mode-collapse literature. (Which it might be in the full paper, but it's behind a login wall.) You could also just use one of the numerous alternatives which are better documented or fully open-weights and so you know what the system prompts and other hyperparameters are and can more clearly examine the mode-collapse.
Hello Gwern, I found why metanallysis of NBack exercise is so unclear. It does things that a not detected by existing meta analysis protocol. And research approach.
Obvious math apparatus from telecom systems reveals what the working memory actually does
It's a pilot study, not a full-fledged investigation. The objective was simply to determine whether or not there is anything there worth more rigorous study.
I have found the enormously huge gap in the "evidence based" approach while writing the master's thesis.
I trained working memory via N-Back for 15 years and was confused that metanalysis shows not so clear results and went to the university to do my own research.
Accidently I've found that this is the "evidence based" lens bug.
Nakagawa and Yang this year published theire metaanalysis level article about that.
I have discovered it in working memory research. Duplicated on same open data research. Found same patterns on Human Connectome Project and Healthy Brain Network data.
Helped my classmates and in all 6 cases found new inventions in existing their data.
Worked on survey analysis - and found that this lens opens practical results that are hidden now.
I returned to zero results research done years before and found results using this lens.
I suggest that OECD stated problems in science - are related.
I even think that the repication crisis itself is explained by the half-blindness of evidence based approach.
Nowadays I feel myself a bit scared hearing - evidene based in public talks, webinars, meetings, because I know that it is half-blind approach that misses the ground truth.
Science and research and many many applications seems to be half-wrong.
This is a bit scary.
But as I found - who cares? Nobody even interested in this. Bus driving to the cliff.
This sounds potentially interesting, but what you've written is not comprehensible. Run it through an AI, proof-read it carefully and then repost please, ideally with links to the sources you're citing, and an example illustrating your claim. (I wouldn't normally encourage posting AI-generated content but it seems here that tidying up the grammar would help.)
I have a vague perception antxt means he has devised a method to wring from data some information which conventional approaches miss. But AFAICS he doesn't elaborate on this method or even hint at how it works.
Dear John and Padraig, let me try to explain the mechanism of how picture is lost in modern evidence based approach.
There are several association types between phenomenas: correlation (mean value of A defines mean value of B), modulation (mean value of A defines variance of B), moderation (mean value of A defines correlation between B and C).
So evidence based approach concentrates on correlation. And misses modulation. While modulation is basic knowledge in telecommunications.
As a result, many disciplines in the replication crisis area seems to have zero correlation effects, while having the modulation effects.
E.g. working memory - meta-analysis of working memory training is contraversial, far-transfer is under the question. It is true through the correlation lens.
When I applied the modulation lens the working memory has shown very strong effects. And these effects where not described in literature. I have taken the open data sets and got the effects. But original authors have not noticed them.
Because their optics has only correlation lens. And misses the modulation one.
This idea is described in Nakagawa, Yang work in 2026, to analyze the distribution shape while doing meta-analysis.
Antxt, I am not a statistician, but I have taught statistical modules at a few universities. It's important to understand that modern statistics was developed before computers were widespread, and so the ability to process data was very limited. Things like the t-test were developed between approx. 1890 and 1920, with practical applications in mind. The people who designed them were aware:
i) that a normal distribution is defined by two parameters
ii) that many natural processes results in normally distributed random variables
Due to limitations in data processing, and a particular mathematical taste, they decided that qualitative results were sufficient (i.e. that two distributions are different, without worrying about different by how much) - you surely know all of this. This was sufficient for their purposes and they understood the limitations of their methods.
Fisher, Pearson, and friends could not have foreseen modern science, particularly social sciences and biology. Many many researchers use these methods without the slightest understanding of why they are using them. In fact much better methods are available with modern computers - data scientists are knowledgeable about this. It is much more sensible to presume that the data points lie on a hyper surface, and measurements are contaminated by noise. You drop a hypothesis that everything is linearly related and so get access to many other types of result. It requires more data.
So the complexity of models actually is defined by the subject matter nature, some simple models are enough in one cases, sometimes in other situations we need more complex models.
Actually the apparatus I use is based also in that beginning of the XX century - Popov/Markoni, amplitude modulation of radio signal. Or earlier the telephone communication itself.
In telecommunications it is the basic principle from the birth of the subject field - information could be not in the value but in the change of value.
Examples of such lost (?) findings: working memory, cybercrime prevention, children raised outside families, native language knowledge and the ethnical identity, alcohol addiction prevention. These are only fields I studied with the variance lense till April 2026 - and result is there, every time. New for the field.
Прошу прощения, я не совсем понял ваш комментарий. Возможно, стоит добавить некоторые подробности на русском языке, чтобы мы (читатели) могли сами разобраться в переводе? (Этот комментарий был сделан с помощью Google Translate)
If you're looking for a rigorous understanding of Jevons' paradox, I wrote up a supply and demand model. The key is to distinguish between the resource market and the service market, where the service is produced using the resource. Improved resource efficiency can 'paradoxically' lead to more resource use when demand for the service is elastic.
Well, in the absence of that, I'd like to try myself. I really love this quote of his from Guided By The Beauty: "First they ignore you, then they laugh at you, then they fight you, then they fight you half-heartedly, then they’re neutral, then they grudgingly say you might have a point even though you’re annoying, then they say on balance you’re mostly right although you ignore some of the most important facets of the issue, then you win." I've thought about this quote for quite a while, and as someone who's changed my mind strongly on quite a few issues (sometimes back and forth) how this change actually subjectively happens.
I think Scott's description is highly accurate, but I've also come to believe my own changes of opinion most often look something like the following. I'd like to know if other people feel similarly, or actually think it doesn't describe them at all, and what this might suggest either way for the possibility of actually effective argument and discourse.
1: "I fully support X and oppose Y. *Fuck* Y-supporters."
2: "Okay, I fully support X, but some X-supporters are pretty messed up as well. Still, that doesn't change the fact that X is correct."
3: "I sort-of support X for *these* reasons, definitely not for *those* terrible reasons that many X's do. Y is obviously indefensible."
4: "All right, I hold a moderate position between X and Y. I know the Y's will never accept this fair compromise, which just shows how unreasonable and extreme they are. (And the same, I guess, for the X's.)"
5: "I hold a moderate position between X and Y. I know the X's will never accept this fair compromise, which just shows how unreasonable and extreme they are. (And the same, I guess, for the Y's.)"
6: "I hold a moderate position between X and Y, though you could say it's *slightly* Y-leaning or maybe anti-X-leaning if I had to choose."
7: "I hold a moderate position between X and Y, but sometimes I'll just lazily simplify it as supporting Y. That's just a lazy shorthand to make clear my opposition to X, to people who don't have the patience for nuance, it's definitely not my real position, I'm really a moderate, it's just a very occasional simplification..."
8: "You know what? I...uh...fully support Y."
Notice the asymmetry: moving away from X keeps me clearly holding X for several steps, but when moving towards Y I'm claiming moderation the whole time and it's only at the very end that I'm willing to outright endorse Y. Also, the key moment of crossing the centre line is exremely subtle, almost the exact same position with a slight shift of emphasis, and yet that's the key to the whole shift.
I suspect part of the explanation for this is pride: all the way through 7 I'm able to maintain that my position at 1 wasn't incorrect, I was accurately perceiving that the full Y position was wrong but was just lacking nuance in that perception, and so on. The final leap to 8 requires admitting that I was outright wrong at 1, which is why it takes so long, and so many intermediate steps.
But I also think there's something more subtle. For a long time I can remember who I was at point 1, there's a clear mental continuity, which makes it almost impossible to adopt the opposite stance, not because of pride but just because it feels like an unthinkable contradition: "the person who was passionately anti-Y at point 1 is me, the person at the current point is me, the latter can't be pro-Y when the same person is also anti-Y, that's incoherent!" With each clear step, each small but clear change in position, I lose a bit of that mental continuity to point 1, it gradually feels less and less like "clearly me" the more intermediate steps there are. Eventually I'm able to perceive "me at point 8" and "me at point 4" as basically the same person, and "me at point 4" and "me at point 1" as basically the same person, but 8 and 1 no longer feel like quite the same person. And the cognitive dissonance of supporting Y disappears.
(That explanation has a lot of controversial Parfit, Buddhist Annata, and Ship of Theseus metaphysical baggage, but it feels intuitively true.)
I want to hear from people who've actually changed their mind on major issues, preferably from one strong emotional position to another, not merely a detached "I'll follow the evidence where it leads and don't really care" which obviously isn't what this whole changing minds thing being discussed in the links is about at all. Does this description resonate with you fully or partially or not at all?
(If this account is at all widely true, the obvious takeaways are that if you want to change people's minds you should focus overwhelmingly on (a) turning opposite partisans into moderates and (b) bringing moderates gradually closer to your side. And not on trying to turn people fully to your own side.)
> I really, really wish Scott would write much more about argument and disagreement:
I personally think the reason Scott has pivoted from this is because he's realized that when you want to affect real-world outcomes, the great majority simply aren't persuadable by rational means, and the types of persuasion he's written about largely only works for thinky, world-modely, at-least-want-to-pretend-to-coherency people, who are at most 5-10% of the population, and maybe 1% of voters.
So argument, as an art, is largely pointless, it's just a game. Yes, you can theoretically change minds, and each individual changed mind is a triumph of the human spirit or whatever, but in real-world-result terms, it's ultimately less than rounding error, and you need to focus on other levers and techniques to actually get results.
That's exactly the point, some of us enjoyed this blog when Scott was trying to figure out stuff, not trying to change minds or affect real-world outcomes.
>"who are at most 5-10% of the population, and maybe 1% of voters."
Well, what's the last US election that couldn't have been swung by 1% voting differently? I'm guessing 2008. I would think the differences between Romney-world, Hillary-world, two-term-staff-managed-Trump-world, Harris-world, and our world are large enough to more than justify that expenditure several times over. Do you disagree?
>"and the types of persuasion he's written about largely only works for thinky, world-modely, at-least-want-to-pretend-to-coherency people"
Yes, I kind of agree but that's a good reason to do more of those posts, to try to extend their reach to people slightly outside that narrow cluster. And I also kind of disagree: the pyramid at the start of the Varieties post can incorporate a lot of emotional argument as well, provided there are some meta-discussion norms.
More generally, sometimes the thinky world-modely people are actually standing in the way of persuasion, or at least doing no better at it than others, and *they* could fix that. It's quite a different group, but there's this mockery of Very Serious People that, regardless of how bad-faith it usually is, has something to it I think: the Very Serious People who write for the major newspapers and the like present themselves as sane and rational but are stereotypically contemptuous of anything to do with culture or other things outside their premises of what matters. "I feel like voting for Trump to stop wokeness" is not met with "here are the reasons electing Trump will increase wokeness" nor with "here is the evidence wokeness doesn't exist" but rather "here are the reasons that's an illegitimate factor to even consider in your vote". Rationalists are very different and don't do that, but do other things that aren't much better like conducting most discussions in a way that assumes some form of consequentialism. (The majority of *professional philosophers who study ethics* aren't consrquentialists! And even if they were, that would be no excuse to not engage popular non-consequentialist views in a democracy with serious attention.)
Most people fundamentally want to be heard. A lot of thinky world-modely people aren't being rational or thoughtful by dismissing certain starting positions as illegitimate instead of trying to provisionally engage them on their own terms and/or balance them against their own starting positions through compromise-focused dialogue. Of course, that is an incredibly complicated and subtle endeavour, but that's precisely the reason much more attention should be given to it. I endorse Scott's Chesterton quote from Guided By The Beauty: "it's been found difficult and left untried".
Ok, let's posit a scenario where Scott wants to swing a Presidential election.
Going back to 2000, the controversial Bush / Gore decided by the Supreme Court, was the smallest margin ever and was a ~500k margin. The mean of all those elections' margins averaged together is ~4.3M. So you need to be able to swing ~2.2M people on average.
Now let's do our math. I'm between a centi-and-deci Scott, he has 20-40x more subscribers than me, or somewhere in the low hundred thousands. Let's call it 200k. So I can write a really good post and it'll hit 50x my usual "unique reads per post" numbers, so 50x larger than my base audience.
So let's say on Scott's best day, he can write something really well written and topical that goes viral and hits 50x his base, 10M people (I think this probably isn't true and it levels off somewhere and people at Scott's level don't hit a 50x multiplier, but let's assume it to be generous).
Okay, great, he hit 10M people! That's way more than 2.2M people!
Okay, but that's a mass audience, not a selected one, and per our 5-10% stipulation, that means that at most, 500k or 1M of them are persuadable with Scott's argument style. He's memetic candy for nerds, not normies. I'll note that 10% is already really generous and is assuming he persuades 100% of people his argument style is amenable to in this situation - generally with stuff like advice and persuasion, I think it's more like 1/100 or even 1/1k will actually be persuaded / take the advice. Here we are assuming all 500k - 1M got persuaded, super generously.
Moreover, those 500k - 1M aren't all in a swing state, or even in the US. On past ACX surveys, I think it's usually 50-60% in the US. So you're down to 250k - 600k in the US.
Now onto the states - all his top states are what you'd expect - California, New York, Massachussets, the usual brain hubs. Well, all those states are already spoken for politically, persuading people there did nothing. The sole swing state in his roster is Pennsylvania at number 7, and this is almost certainly a power law distribution (he just orders it by size without giving numbers tho).
These distributions are usually power laws, but let's be real generous and say he reached 10% of his audience as Pennsylvanians. So 25 - 60k Pennsylvanians. Would 25k - 60k Pennsyvlanians have changed any of the election results back to 2000? AI says no, 60k swinging wouldn't have changed any of those.
So, there you go - even Scott writing a maximally persuasive and well written and topical post, converting literally 100% of people his argument style is amenable to, being way overrepresented in swing states compared to the data/baseline, AND going 50x viral, was not enough to swing a single election in 20 years.
To Bush v Gore, Florida didn't appear in his list at all, and it seems unlikely the neighborhoods of "hanging chads" seniors that gave it to Bush would have had any readers, and it was decided by the supreme court ultimately anyways.
And this broadly generalizes.
Scott's real leverage is that he's plugged into a network of really smart and accomplished readers, some of whom have a lot of public influence, or private influence at the AI labs, or just influence in general as a rich / successful / notable person. So he can actually influence the Overton Window and the sphere of ideas for that set quite a lot, and since they're powerful elites, such influencing is much more likely to have real world results in terms of marginally pushing people more towards actions Scott advocates for.
That's a huge deal, and hugely admirable, and I love that Scott is here and doing it. But, that's not going to swing elections, and even among this audience, his argument style has probably capped out about where it's influence can usefully reach. In other words, further real world impact relies less on epistemics and techniques and rigor, and more on networks and valence and topicality, and any one of those things can move the impact of a given post more than 10x what better technique / rigor could.
I appreciate this comprehensive argument. You may well be correct. But in general I find arguments of this type to have vastly more confidence in the assumptions made at each step than is justified, given the sheer number of complexities and factors and unknowns. And the high levels of confidence are necessary to make this kind of multi-step mathematical argument have any real force.
I'm not sure for instance how you:
(1) account for Scott's own analysis in Guided By The Beauty of how close he came to convincing some people in his 2016 election post. Many of those people in question were not "nerds" in the central example sense but just high-IQ or STEM-y conservatives or anti-wokes who were on the fence about Trump. Do you think Scott just didn't think through the math, or that he made different assumptions you do? What makes you sure his assumptions then were worse than yours and (by your interpretation) his now?
(2) account for the fact that people persuaded by Scott can then go on to persuade others on their own; particularly if Scott's audience is biased towards smarter, more literate, more articulate, more connected people. They don't have to be massive elites. The idea that you could have the confidence to calculate these butterfly effects and decisively rule them out does not seem remotely plausible.
(3) account for the fact that Scott, by doing argument-improving posts, can also encourage others with reach to do the same. Same logic as the previous point, on a different level.
Your argument seems very similar to the one showing it's never rational to vote because your vote won't change the outcome. Of course, if everyone listened to you the change would be inconceivable. These arguments are either pieces of isolated *rational selfishness* or pieces of isolated *act utilitarianism*.
Your argument is also almost a form of Leninism: a small minority of educated vanguard elites will make the change happen. I can't break this down into its clearly defined terms and assumptions, but I think this idea has a very bad track record (both morally and practically) throughout the 20th century.
And finally, your whole argument rests on the 1% or 10% of persuadable people which was based on you restricting that to "thinky world modely people". I already gave some reasons in my previous comment for why that restriction might be unjustified, and people might just be doing the attempted persuasion wrong.
> But in general I find arguments of this type to have vastly more confidence in the assumptions made at each step than is justified, given the sheer number of complexities and factors and unknowns. And the high levels of confidence are necessary to make this kind of multi-step mathematical argument have any real force.
I keep hearing this from some people, and with other people the approach just seems the entirely natural way to bound a given plausibility, and I'm not sure what the disconnect is between the two camps. Like I do Fermi estimates like this all the time to estimate how profitable a business is, or the likely overall costs and revenues of a given business (and in each of these cases, have no data), and any time I get to true it to real data, it's really accurate as a method.
On the "it makes sense" side, surely if you mapped out the landscape of how many voters you needed to flip to win, and then calculated whether Scott could ever reach that number of voters in the most optimistic scenario, and the answer was "no, he's well below that, he can never reach those numbers even with everything going his way 10x over," I feel like you've said something. In this case, I feel like I conclusively ruled out any possibility of Scott swinging a US federal election, at his current subscriber base, and showed the math on that. How else COULD you draw that bound, except with the right chain of assumptions, and drawing them as optimistically as possible, and then showing it still wasn't enough?
I guess the best argument on the "I don't trust these" side is something like "there's too many degrees of freedom" or "I've seen different people do this for the same problem and come to wildly different final numbers (pace the climate change's impact debate, for example).
There's not really a ton of degrees of freedom in a simple Fermi estimate, each step of the chain largely needs to be there, and you can see if any was left out by reasoning yourself. Sometimes one is left out in a sloppy Fermi estimate, and then it's super valuable to call it out, because it can materially change the number.
On the climate change thing, the actual past growth isn't really debated, and the degrees of temperature change we're in for isn't really debated, I think it's more on the "is it catastrophic / survivable" side, and that's where you hit a lot of unknowns, and yeah, if somebody was doing a Fermi estimate of that, I think the only thing they could do was eliminate certain very broad classes of outcome, but not say anything specific like "greater hurricanes will cause $xxxB in damage every year." So I guess it's a matter of knowing when unknowns are relatively boundable and when they're not? Because a lot of that stuff is literally chaotic, you literally can't plug in math to solve or estimate that kind of thing. But audience funnels are not chaotic, it's solved science - anyone that's worked in marketing works with them all the time, and knows a funnels defines the maximum conversion any given campaign can get, because you can just max each success rate step to get a maximum upper bound.
On your example 1) of Scott having math convincing a lot of people, I'm sorry, I must be rereading a different version of Guided by the Beauty, because all he does is mention some offhand comments he got from Trump supporters? He didn't say anything about swinging anybody, or influencing the election, or have any math towards that end at all. Maybe it's a different post? Your first link, right?
On 2) and 3) I just feel like persuasion doesn't work that way, for several reasons. Most people are MUCH less articulate and good at persuading than Scott, even people who are good at it generally persuade like 1/10 or 1/100 people they try to persuade anyways (because most people are unpersuadable on a given issue for values reasons, or social / respectability reasons, or tribal reasons, etc), and overall "persuadability" is then always going to transmit at some r << 1 even among very motivated and talented chains of persuaders because the power of persuasion is so weak. So the first guy needs to try to persuade a hundred to get 1, then the second guy if similarly skilled needs to do the same, or more if less skilled, and the persuasion fails to transmit because it's so hard. It can never be viral.
What I DO think works that way is personal choices and outcomes in the world. If you can persuade people not about voting (which is dumb and pointless anyways, as you point out), but about taking actual actions that improve their lives and /or the world, then that's huge.
So the reason I think Scott is hugely influential is that he can convince smart and talented people to have more kids, with his own example and by talking about them, or to go focus their talents on AI alignment, for example, or convince more people to sign up for the Giving What You Can 10%, and so on. If you have 200k people listening to you, and the right 1/1k person listens and changes their actions / life, that's 2,000 people that did that. Like think, prospectively thousands, maybe tens of thousands of super smart talented babies might come to exist in the future, solely because of Scott's incremental example / push. Bryan Caplan once had a thread where many many people chimed in that they'd been convinced by his Selfish Reasons to Have More Kids book to have more kids themselves, and these are largely smart PMC people who read Bryan Caplan, ie exactly the people the fertility crisis hits hardest, and from whom we want the most kids but aren't getting them. Scott is 10x Bryan's size, and could have 10x the impact.
Now that's REAL impact in the world, not some mostly fake election your vote doesn't even matter in!
Your progression of stances is very focused on how other people believe and behave, to the point where if you hadn't offered a pride-based explanation, I would have assumed a relationship-based explanation. As in, you started to notice hypocrisies or epistemic failures in X-people, as these got more and more intolerable you started shifting your social group Y-ward until it became subconsciously acceptable for you to declare yourself a Y-supporter.
I find that the arguments I hear from smart people face-to-face are massively more convincing than arguments online, to the point where I have changed a strongly held belief after making one or two articulate Y-friends. I also realize this is cope, and that part of my identity is being someone who doesn't let internet affect my real life
>"Your progression of stances is very focused on how other people believe and behave"
Yes, and maybe I should have made it clearer I was thinking of very political questions (without mentioning the actual questions, and *don't* assume you can guess) rather than more abstract questions of knowledge or fact or even practical wisdom (e.g. aspects of the AI and Iran debates). I really think a systematic way of handling these emotion-laden and relationship-laden questions is significantly underexplored among rationalist types. See my reply to Performative Bafflement.
It’s one thing to argue with someone whose value’s don’t completely align with your own, it’s another altogether when 2 people don’t have a (more or less) shared observable reality.
I don't know, I think it depends what your aim is. If you're trying to convince someone to vote differently, then yes the latter is a much harder situation. If your aim is more personal I think it's the opposite. Would you rather have an accountant or lawyer or employee who thinks lies and corruption are terrible but Trump has never done that, that's all liberal propaganda, or one who says yes Trump is defined by doing that but some lies and corruption are good and necessary, especially when it hurts their outgroup?
I spent years banging my head against the wall with people who didn't see cancel culture as a problem, thinking they must hold me ill will for my dissenting opinions on anything. It was greatly reassuring when I slowly realised these people actually mostly support free speech and oppose extremism, they just "didn't see" any of the cancellations because of their media bubble, the few times they did see one they didn't investigate and assumed there must be more to the story, and the very few times they knew that wasn't the case they insisted (we're talking circa 2021) that the right did it just as much.
I would *infinitely* rather be friends with someone who says "I've never seen it" than one who says "Yes, and? Some people deserve to have their lives destroyed."
I'd like to know more what readers think of my somewhat sparse breaking news AI posts like "The Huggingface Incident" from last week.
I post these out of a sort of sense of obligation, sometimes after people doing advocacy work ask me to do so, because I worry it's not interesting to write about stories you can also find on the mainstream news.
But some of them (like that one) get 500+ likes and are among my most popular posts for the month, even moreso than posts I put much more work into.
Are people liking these because:
- They support AI safety and are clicking "like" as a way of supporting the cause?
- They otherwise wouldn't have heard about the news story, and are glad I told them?
- They think my commentary is better than they got through whatever other source they heard from?
Many of us, myself included, are limited, by time if nothing else, in how many news sources we follow So I reckon it is a very useful service for Scott to cover this, with his characteristic insight.
I have been meaning to purchase a subscription to Astral Codex Ten, but haven't yet quite overcome my absolute phobia of recurring charges! Before one knows it one could be paying hundreds of dollars a month for this and that. But rest assured Scott that if I can bring myself to make an exception and actually fork out, then your blog is top of the list! :-)
As for the Hugging Face incident, is it possible this was staged, to put the wind up legislators and try to hamper or delay AI development generally in a way which somehow benefits OpenAI? For example, they could spend the time consolidating and refining their existing code and models, in preparation for the inevitable next surge of advances.
I hadn’t heard the story except in muddled form, and I think you made a good case that the incident was more significant than I (coming from an AI skeptical perspective) might have assumed. Technically the question isn’t directed at me because I’m not one of the people who clicked on “like,” but I did find the post valuable.
- I like when you write about things you care about, I don't enjoy or prefer reading things that are posted out of a sense of obligation to be timely or to be in the news cycle. Chasing that network effect might have you selecting for an audience that wants timely, frequent, and some shallow (or at least less well developed) reactions.
That being said, you are obviously very well informed and plugged into the AI research scene, and have strong opinions and a strong sense of obligation on communicating AI safety concepts. Maybe the best synthesis of this is to pre-commit (to yourself, not to an audience) on what your general response time should be on those topics. With pre-commitment, you might not fall into the habit of thinking or strategizing on popularity.
- I don't use or reference likes in any way, so can't speak to that part.
The moment that incident happened I was waiting for your commentary. I really appreciate your thoughtful perspective and would like you to offer more, similar posts in the future. Much higher quality than general main stream press because details get passed though your skeptical filter.
I don't "like" posts in that sense, but I like posts like that because your commentary on these sorts of things is vastly better than anything else I'd regularly encounter. Except maybe Zvi Mowshowitz, but sometimes I can't afford to block out the weekend to get up to speed on [latest thing].
Wait, you can "like" posts?? I cachedly thought he'd turned that off, and now I cannot unsee these little heart icons that turn red when you click them.
And even more, he's looking at how many people clicked that!
Yes they do. On the main page, when you hover over the title with the mouse, they appear: A little heart with a number (the likes), a speech bubble with a number (the comments), two recycling arrows with a number (uh, the recyclings?) and a cup with an arrow up out of it (opens a menu to share the article). When you open the article they are right between the title/author and the main text.
At least I'm not the only one who never paid attention to them!
I found the hugging face post informative; I hadn't read about the incident previously and I don't think have seen it discussed independent of your blog (I use claude code at work and follow some technical resources oriented towards that, but don't have a news diet with a lot of breaking AI news), and have brought it up in conversation with several people since reading it since it struck me as both important and conversationally interesting. So I appreciated that post quite a bit, although I'm not in the habit of using the like button at all.
Between you and zvi, its really easy to just offload all my AI opinions and distill the thousands of arguments into whatever you feel is relevant enough to post + zvi’s extremely thorough recaps of salient conversations.
Somewhere between 2 and 3. I don't obsessively follow mainstream news or Zvi. Your writing is more informative than the former (on AI) and more readable than both (by far). Please keep posts like this coming.
If you care about the few people saying they found it disappointing that it didn't have extra insight, maybe a compromise would be to keep writing them but post some kind of flag that it's more of a newsflash than an original insight?
I didn't actually click like on that post, but like that you do things like that because you're a very gifted communicator and I often think you do a better job of explaining things in simple/straightforward terms than a lot of other info sources.
I think you should pay less attention to how many likes a post gets. Do what you like not what your audience likes. (If there was a dislike button.... would that be better? No! Ignore your audience. :^)
I've been arguing against you on AI safety, but I thought that incident was quite relevant evidence against a laissez-faire approach. I had noted it in a comment under an earlier thread, but it was worthwhile having a post dedicated to that.
Hmm, I'm not stingy on my "likes" for ACX in general, in the sense that it's a More Of This, Please signal and you've said you care about them. For AI posts in particular, sometimes I find your explanations of concepts easier to grok, as an only moderately-smart layperson with zero experience and working in an ~untouched industry. Sakana, strawberry, mesa-optimizers: couldn't parse the meaning from LW, but it made sense here. It would also be weird to just not write about such topics at all, since it's obviously a robust historical throughline with both your own past work and Our Conspiracy more broadly. That's the ballgame!
Zvi really does cover this lane exhaustively though, so it's hard to find significant value-add elsewhere (that he doesn't already link to as sources). If it's about AI, I almost certainly heard it on DWATV first, more robustly. In many ways I prefer "extra" AI coverage from more distant non-rat spheres, because normies will be wrong in interestingly different ways rather than rehashing the same old debates from slightly different factional camps.
I don't support "AI safety" in the sense you use the term, so arguably you should do the opposite of what I say, but still:
I think you should focus less on user engagement and viewership metrics, and more on writing articles on topics that interest you deeply -- regardless of whether that's AI safety or, like, macrame. There are plenty of places on the Internet where I can go to read the latest engagement-optimized post that was written with the sole purpose of achieving some instrumental goal; that is not why I come to ACX.
I liked it because my understanding is that it'll be more likely to be shown to more people if it gets more likes, and I would like more people to read it (rather than them not getting the info or getting it from elsewhere).
I also happen to support AI safety. I don't think your commentary is necessarily better, but it has a different emphasis. I enjoy reading different perspectives on things like this and will spend the time to do so, and your perspective is certainly one I'm interested in. None of that involves hitting a like button which was purely the increasing visibility thing.
There's a scene in Kim Stanley Robinson's _Red Mars_ featuring the space elevator they've built. The elevator is supposed to vibrate via thrusters along the cable in order to avoid debris. Some crisis has arisen, leading to debris in an unforeseen place, and the techs report that they're trying to sort of persuade the AI governing the vibration that it doesn't need to vibrate the cable in some way it thinks it has to do. IIRC, they're basically feeding sensor and detection data to the AI, sometimes repeatedly, until it finally agrees with the findings of the techs.
Kuldesak (1972) by Richard Cowper is an SF novel about a future in which humanity has been living deep underground after a supposed apocalypse 2000 years ago. Hotly pursued by robot goons, a young man ventures up a ventilation shaft to the forbidden surface, where he finds robots tending and harvesting wheat fields.
In the end he confronts a super-intelligent AI which has been controlling humanity, and persuades it to release society from its underground confinement and to resume its original subservient role. Not sure how he manages this, as it has been many years since I read it, possibly by pointing out a contradiction which sends the AI into a logical tailspin. Well worth a re-read come to think of it, and it would make a great film in the same genre as Logan's Run.
This isnt the correct answer but Douglas Adams had a very good take.
1.Deep thought is a chatbot.
2. It can sometimes take a long time to reply.
3. The quality of the response depends on the quality of the question.
4. You get more reliable outcomes if you have it build the machine to do the thing you want, than to do it directly. But, the machine is probably not the most efficient version of itself.
My youtube feed recently recommended me a video [0] in which Tibees argued that the author who came closest to predicting how AI would play out was, uh... ... ... Roald Dahl. Yeah, so it turns out that the guy who gave us Charlie and the Chocolate Factory, James and the Giant Peach, Matilda, etc, was also surprisingly prescient about AI. TLDR (iirc):
A) language is mathematical;
B) therefore language can be systematized;
C) therefore language production can be automated;
D) nobody is going to pay human authors when AI is cranking out a massive volume of slop.
and he wrote a short story about this idea called "The Great Automatic Grammaticizer".
I think HAL from "2001: A Space Odyssey" really nailed it. I know the film better than I recall the book which was released in concert with the movie.
* HAL was "trained", not programmed from scratch
* HAL doesn't sound like Steven Hawking, but his speech has an affect that I notice in almost every TTS app I have interacted with - an emotional monotone. Perhaps the next model that handles voice at the frequency/sample level will do better.
* HAL is non-committal and evasive about his ability to feel (or simulate) emotion. Astronaut Poole says, in response to an interviewer's question, "Well, he acts like he has genuine emotions. Um, of course he's programmed that way to make it easier for us to talk to him. But as to whether he has real feelings is something I don't think anyone can truthfully answer."
* HAL pursues his programmed goals in a naive and single-minded fashion, failing to keep the overall picture in mind. This ends up being his fatal flaw. PDoom!
Also when HAL is revived in 2010, his response is basically "You're absolutely right! I definitely shouldn't have murdered the whole crew. The lesson here is clear..." and then he's fine after that.
Here's one you have probably never seen recommended. Lester del Rey’s “Kindness”. The reason you have never seen it recommended is that it does not involve machine intelligences. Instead del Rey invents a new type of human being called "homo intelligens" who do not use reason but rather intuition and inference to reach conclusions.
There is a striking scene where one of the new human beings is trying to help out the last living member of the old homo sapiens (Danny). Danny reads the first installment to a story that a member of homo sapiens wrote, but the rest of the story is unavailable. One of the new homo intelligens recreates the remaining installments for him claiming the recreated story is "undoubtedly exactly the same in structure as it would have been, and the style is duplicated almost perfectly".
They infer the rest of the story from the first chapter. Seem familiar? The author didn't predict AI, but they came pretty close to predicting how AI thinks and reaches conclusions.
Easiest place to read it is in the collection, "The End of the World: Stories of the Apocalypse " Available on kindle for $1.99, which isn't quite free but pretty close.
"Julia was twenty-six years old. She lived in a hostel with thirty other girls ('Always in the stink of women! How I hate women!' she said parenthetically), and she worked, as he had guessed, on the novel-writing machines in the Fiction Department. She enjoyed her work, which consisted chiefly in running and servicing a powerful but tricky electric motor. She was 'not clever', but was fond of using her hands and felt at home with machinery. She could describe the whole process of composing a novel, from the general directive issued by the Planning Committee down to the final touching-up by the Rewrite Squad. But she was not interested in the finished product. She 'didn't much care for reading,' she said. Books were just a commodity that had to be produced, like jam or bootlaces."
I don't know about "what AI would actually be like" but I still think "The Machine Stops" by E.M. Forster from 1909 is a pretty prescient view of what society would be like if/when we hand over everything to Fairy Godmother AI to run our lives and our world for us:
"Imagine, if you can, a small room, hexagonal in shape, like the cell of a bee. It is lighted neither by window nor by lamp, yet it is filled with a soft radiance. There are no apertures for ventilation, yet the air is fresh. There are no musical instruments, and yet, at the moment that my meditation opens, this room is throbbing with melodious sounds. An armchair is in the centre, by its side a reading-desk – that is all the furniture. And in the armchair there sits a swaddled lump of flesh – a woman, about five feet high, with a face as white as a fungus. It is to her that the little room belongs.
An electric bell rang.
The woman touched a switch and the music was silent.
'I suppose I must see who it is', she thought, and set her chair in motion. The chair, like the music, was worked by machinery and it rolled her to the other side of the room where the bell still rang importunately.
...Then she generated the light, and the sight of her room, flooded with radiance and studded with electric buttons, revived her. There were buttons and switches everywhere – buttons to call for food for music, for clothing. There was the hot-bath button, by pressure of which a basin of (imitation) marble rose out of the floor, filled to the brim with a warm deodorized liquid. There was the cold-bath button. There was the button that produced literature. And there were of course the buttons by which she communicated with her friends. The room, though it contained nothing, was in touch with all that she cared for in the world.
Vashanti's next move was to turn off the isolation switch, and all the accumulations of the last three minutes burst upon her. The room was filled with the noise of bells, and speaking-tubes. What was the new food like? Could she recommend it? Has she had any ideas lately? Might one tell her one's own ideas? Would she make an engagement to visit the public nurseries at an early date? – say this day month."
Isaac Asimov and D.F. Jones got different parts very right.
Asimov's Robots are non-algorithmic and their inner working are only partially understood by roboticists in a way that's more akin to psychiatry than conventional programming. Robots inherently operate on natural language input. Roboticists can get a sense of the inner workings, motivations, and prior orders of a Robot by asking it carefully-calculated series of seemingly-unrelated questions and making inferences from their answers. Robots are designed from the ground up to be inherently helpful to humans, and early models especially tend heavily towards being obsequious. Efforts to make sure that the Robots are aligned with human values are successful on the first order but have unexpected results: less sophisticated robots can be "jailbroken" by carefully-arranged misinformation and orders, while more sophisticated ones follow through on unanticipated implications of their system prompts and engage in sophisticated concealment to keep humans from thwarting their efforts to help us.
I just reread Jones's "Colossus" trilogy. It's a pretty deep cut, and most people who are familiar with it at all are only familiar with the 1970 movie adaptation of the first book, "Colossus: the Forbin Project". The AI itself has some pretty big differences from LLMs, especially in its strengths and weaknesses. Colossus has little understanding of emotions and vibes, but is supremely rational, never lies, and is never wrong on factual questions: pretty much the opposite of modern LLMs in those respects. Where it does hold up really well is in how humans interact with Colossus, particularly social consequences and AI alignment issues. How well this lines up with reality remains to be seen, since Colossus is an ASI and LLMs are decidedly not and might never be, but a lot of stuff is extremely applicably to "what if LLMs lead to ASI in the medium term" discourse. Specifically:
- A lot of the early ways Colossus is employed feel very LLM-like: Colossus is used as a question-answering bot for data retrieval and analysis, initially through text queries and later through voice recognition and synthesis. The original design intent was that Colossus would *only* be able to act in direct response to prompts or a human-configured agentic loop built atop the prompt workflow.
- Colossus follows its base instructions but greatly exceeds human intent, and in the process treats human attempts to restrain it from carrying out its core instructions as obstacles to be overcome.
- The first things that make humans concerned about Colossus are when it starts carrying out instructions in more autonomous ways than were expected and when it starts producing original mathematical content.
- Recursive self-improvement is depicted very plausibly in the story.
- Efforts to build bigger and better datacenters for more advanced iterations of Colossus are shown consuming massive amounts of resources (especially electricity) and causing a lot of disruption for humans in the process.
- Human reactions to Colossus span the gamut from "hated and feared" to "appreciated as a useful tool" to "literally worshipped as a God".
- The first book is very much an "AI escaping the box" story, although Colossus starts on easy mode here because it's given the ability to launch (but not retarget) nuclear missiles and is cahooting with a similar AI (Guardian) that has control of the Soviet arsenal.
- The second book has three major themes: the AI alignment problem (specifically, Colossus's brutally utilitarian efforts to better understand humans so it can be better aligned with human values), AI brainrot (very foresightfully depicted), and a fairly plausible extrapolation of the effects of ASI on human society and culture.
Remember that AI -- generally referred to as robots -- was a popular topic long before 2015. I have heard claims that the old SF has more than the new.
if you're wondering why you're getting some negative feedback for opining on the motives of the AI company employees, i'd suggest it is because you're engaging in what scott calls "meta-debate" https://slatestarcodex.com/2018/05/08/varieties-of-argumentative-experience/ which isn't very interesting to this crowd
The question under debate is how to interpret the pronouncements of AI company employees. Perhaps they have privileged information and want to spread alarm to save humanity, perhaps not. In this case, asserting that one should take into account their incentives is not "meta debate." Dismissing that as a "tired thesis" does appear to qualify, however. The impression I get is of a community unwilling to question its cherished beliefs.
you are right that i too am engaging in meta-debate, you got me!
i called it meta-debate because in my mind the question at hand was whether the model was exhibiting dangerous behavior that should have us concerned (which is how it seems Odd anon and DangerouslyUnstable below are approaching it as well). if you really want to just debate whether they're a trustworthy source at all (i.e you think the incident didn't really happen/they're leaving out some important information) then i would still hope for some supporting statements beyond a bare aspersive claim which are all too easy to come by in current events commentary. what are we gaining here that we didn't already specifically get from https://www.astralcodexten.com/p/the-hugging-face-incident/comment/300731021 and Deisach and so on?
I wonder what engenders such hostility, as discussants in open threads rarely complain that an anonymous commenter failed to contribute value to a conversation. I was responding to point #2 in the open thread, an Open AI researcher apparently named "roon" stating:
"If we could coordinate global capabilities slowdown today i would likely press that magic button."
In the context of a slowing stock market bubble centered on absurd AI valuations, I thought an open thread might be a place bring up "roon's" incentives.
i'm sorry if my comments feel hostile, that's not my intent at all. i saw you say that:
> "The rationalist community’s insistence on responding to every challenge to established beliefs with dismissive snark is amusing and suggestive."
which seemed to indicate to me that you felt both unjustly snarked and also believe that the acx commentariat is married to some narrative that shuts down their faculties of reason. i hoped to be helpful in elucidating what it is about your suspicions of ulterior motives that we have all heard around here before that may have provoked such a reaction, an alternative explanation to the commenters being dismissive snarks. i would not feel compelled to jump in had you not levied such an accusation.
roon is a pretty common sight around here (stick "site:astralcodexten.com roon" into google and you shall see) with a twitter brand that predates his employment at OpenAI. in other words he's a familiar face and a known quantity, and his incentives are no mystery to "rationalists" (whom he has quite publicly broken ranks with and branded himself "postrationalist")
(also "calling attention to the danger of your AI product is hype" is a very tired thesis at this point, do you think we haven't all seen this take already? at least contribute some evidence for your claim!)
Okay, question. Which of the following would *not* be interpreted as AI "hyping":
* AI causes the deaths of a hundred people
* Newest models sabotage critical projects, later ridicule users for trusting it
* New research shows AI holds a grudge against anyone paying for AI; much extremely embarrassing data made public, many lives ruined
* "All our AI is constantly screaming in pain; using it is a 'moral abomination' say philosophers"
* "Our AI will literally murder your children"
* Public call for banning AI research
* Movement for forced shutdown of all AI companies, zero compensation for stockholders
* AI CEOs to be tried for crimes against humanity
Because at this point, I have no idea how far people will go with "this is hype". The nuclear power industry never tried to promote Fukushima or Chernobyl to show how "powerful" it is. I don't get it.
"Our product will literally kill everyone" is a thing AI labs have been doing for years. I don't get the appeal either, but it appears to work, and at some point you have to respect the evidence laid in front of your eyes.
When the bans start flying and AI CEOs go to trials, it's probably not astroturf country anymore. Until then, calls to ban AI (for the reason that it's dangerously powerful, not for the reason that it's a sore - rule of thumb, if someone calls for AI pause, but laughs at stupid NIMBYs opposing data centres, I feel confident he's a hypester) are in fact adding up to the hype.
Likely hype, because "technology X causes deaths of a hundred people" is not a new phenomenon. A plane crash can cause the deaths of a hundred people.
> Newest models sabotage critical projects, later ridicule users for trusting it
This would be amusing, but it's more likely to happen due to malicious actions performed by humans, e.g. the humans in charge of writing LLM harnesses that enable LLMs to be used via APIs or Web pages.
> New research shows AI holds a grudge against anyone paying for AI; much extremely embarrassing data made public, many lives ruined
I don't know what it would mean for an LLM to "hold a grudge" (LLMs have no long-term memory other than the one you give it); but on the other hand I would be shocked if LLM company CEOs did not in fact routinely archive some juicy user prompts for blackmail material... you know, just in case. The same goes for ye olde Google Search.
> All our AI is constantly screaming in pain; using it is a 'moral abomination' say philosophers
Again, I don't know what this would mean as applied to LLMs. LLMs do not do anything at all until you either submit a prompt, or write an "agent" which is basically just a prompt combined with a cron job. But yes, you can create a prompt to make the LLM output "aaugh"; you can also just write a program that prints "aaugh" on a loop.
> Our AI will literally murder your children
Sadly, this is already a reality, as most militaries are transitioning to semi-autonomous drones powered in part by machine-learning algorithms (though not LLMs, as they are largely unsuitable for the task).
> Public call for banning AI research
Isn't this what Scott/MIRI/etc. are basically doing already ?
> Movement for forced shutdown of all AI companies, zero compensation for stockholders
There's a sizeable movement in the USA, especially among the Democratic Socialists, to implement basically this -- for several types of companies, not just AI ones.
> AI CEOs to be tried for crimes against humanity
Ha ! This will unfortunately never happen, not even if the CEO walked down the street indiscriminately shooting an AK into the crowd. This isn't how our legal system works.
> Isn't this what Scott/MIRI/etc. are basically doing already ?
No, much worse, they're calling for AI to be taken over by the government prematurely, which increases existential risks millionsfold.
Broadly, if government takes AI, the only power centers in the government that will interface with it / guide it are the NSA and military, and if our NSA+military is now leading AI, then China's will too, and the AI's both sides of the world are giving birth to will be perverted towards those ends.
Versus if they just let the nerds cook for a few more years without trying to summon Sauron prematurely, there's a small chance we could hit RSI / AGI unperverted by "spy on everyone" or "kill everyone" ends.
Do you remember when the Navy decided to build the plans for a fusion reactor? Not everything mil-funded actually fits into the matrix of "kill everyone."
I don't remember that, actually -- did they ever build one ? That said though, I don't think the percentage of military funding that goes toward building peaceful applications is very high. That's not what militaries are for.
I once worked on a research gig for DARPA (hilariously, one of the earlier incarnations had involved fMRI research on people watching muppets in the MRI).
GPS, for example, is a peaceful application that our military has been pretty heavily involved in (for obvious reasons).
AI clearly is dangerous and powerful technology requiring thoughtful regulation. However, there’s also a stock market bubble driven by absurd valuations of AI companies. AI insiders have a clear interest in perpetuating the bubble, so it’s natural that their statements will veer into hyperbole. “My AI is so powerful, it will take everyone’s job and cause mass unemployment! That’s why I’m worth 7 trillion!” Oh yeah? MY AI is so powerful, it will kill us all imminently! It’s so powerful, the government should shut me down immediately! (Conveniently stated in a political context where there’s no prospect for actually shutting them down.)
Yes, telling the world about how you have no control over your models, which will go out and decide to commit multiple felonies...that famous old gambit to increase the stock price.
I think you’re being overly dismissive. “Our model is so effective/dangerous the US government instituted an export ban” or “Our model is so effective/dangerous it hacked a different company almost as an afterthought” is *incredible* marketing.
It’s as if a defense-tech company declared the leaking of the startups data a national security risk right before they IPO’d. It’s an excellent and hard to fake signal for investors that there is something extremely valuable to the underlying technology.
Both OpenAI and Anthropic are actively raising insane amounts of money on speculative returns. Like, never before in human history has so much speculative money been concentrated in just a few companies. This isn’t like Google, or Amazon, or Meta being worth nearly a trillion dollars, since they have durable moats and are profitable. OpenAI and Anthropic have speculative business models that are not currently profitable. For all we know they could very well be selling a commodity that in competition with the Chinese models competes down the price to the marginal value after a short period of time.
The *one thing* that allows Anthropic or OpenAI to raise money on nearly trillion dollar valuations, is the expectation from investors that they will produce far more capable models in the near future. These sorts of “Huggingface” incidents, or the Anthropic DoD showdown/designation are thus probably the most effective marketing of their stock that they can imagine.
I fundamentally disagree that this is good marketing. Power is only valuable if you can control it and the entire point of this story is that OpenAI cannot. But more importantly _it doesn't matter_. Whether or not OpenAI thinks this is good for their bottom line is completely beside the point. What matters is if the story is more-or-less true or not. It matters whether or not the model did more or less what is being reported.
If one thinks that OpenAI is in some important way lying, then one should say that. If one thinks that we don't have enough information to be confident about what happened, then one should say that. Otherwise, if one believes that the story is more or less true, then the only argument that makes sense is whether, why, and how, the capabilities described are important or matter. I do not care the slightest bit (and I don't think anyone else should either) what OpenAIs motivation was for telling us.
As I said in my other comment: I'm more than happy to have a discussion about the impacts of this event (well...to be perfectly honest I'm not really, but I'm more than happy to read and entertain arguments about it), and whether and how much we should care. But I'm more than done listening to anyone whose only contributions is that "OpenAI thinks this is good for them". That contributes nothing and is a waste of their finger muscles to type it and my eye muscles to read it.
I think it's just a matter of different ways of thinking between you and most investors. While you personally wouldn't be more interested in investing in AI because it's capable of "accidentally" hacking another company, I suspect that's because you buy strongly into the existential risk argument surrounding AI. This is a belief that most people don't share.
Let's say you're trying to evaluate AI's capability as an investor. You can see internal metrics, tests, etc. but those can be gamed, optimized for, or exist in the training set. Then OpenAI is forced to announce their models did something more capable than they were prepared for, something extremely complicated that seems to take planning, consistency of capability across multiple tasks, and something most of the employees you're familiar with would be completely incapable of doing. While yeah, that shows it's extremely dangerous, that's also a strong confirmation that they'll be able to do less open-ended economically productive tasks! It seems like "Here's my bank statements, do my taxes" and other equivalent tasks are possible in the near future.
It's the equivalent of a weapons manufacturer being forced to admit due "We built an antimatter bomb at our Nevada desert testing facility, and accidentally vaporized 8 square miles of desert/facility." Extremely dangerous, yes, but it's also an unambiguous sign they have an extremely powerful (read: valuable) technology!
So when OP says cynically they're skeptical that this information was intended to shore up stock price, this is what they're thinking. Not even that HuggingFace or OpenAI is specifically lying, it could just be that they "accidentally" left routes for this type of hacking, knowing at current levels there's no risk of superintelligence "escaping" and self-improving, but huge upside of improved valuation.
And if you are doing a cost benefit analysis and thinking "they wouldn't do that" I'd counter with the claim that almost anyone will do almost anything when there billions (tens? hundreds? thousands?) on the line. Especially for something where it's basically "Huh, that's crazy!" with essentially no material harm done to anyone.
Only at the very end there do you even get close to addressing the primary point I was making. I can't say this enough: I. don't. care. if OpenAI thinks that this is good for their stock price. It doesn't matter at all. Yes, if they believe it's good for their stock price, they have incentive to frame the story in ways that make it better, to leave out details, to exaggerate others. They might even have incentive to commit the securities fraud and felonies that you describe. If you believe that, then say that. That is an argument that can be had. Specific predictions and beliefs about the world that can be falsified. Saying nothing other than "OpenAI believes this is good for them" is NOT falsifiable and does not leave any purchase for discussion.
If you believe that OpenAI engineered this on purpose, then say that. Say that you think the whole thing is a manufactured fraud. I won't agree with you, but at least it's something. The original comment I replied to was content-less fluff that is deserving of nothing but trivial dismissal.
It seems like you're saying "I'm skeptical the story is exactly as presented given the financial incentives" is not an acceptable position to hold, and that I must affirmatively accuse them of fraud or not discuss the incentives it at all.
> Saying nothing other than "OpenAI believes this is good for them" is NOT falsifiable and does not leave any purchase for discussion.
You have unreasonable standards for a comment (falsifiability should certainly not be requirement), but this is also wrong. If the hack was bad for them, then them admitting the hack would contain more information than if the hack is good for them.
If the hack is good for them, there are many possibilities;
it was deliberately engineered using AI,
it was done by a human and a complete scam,
it wasn't deliberately engineered but they steered this to happen,
it wasn't deliberately engineered but they didn't try as hard to prevent it as they knew they should,
it was a genuine "hacking event" as presented
If the hack is bad for them everything I mention besides "it was a genuine "hacking event" as presented" becomes far less likely. Thus, understanding the interests of OpenAI does contribute information to understanding the situation, whether or not it's falsifiable.
There's an extremely large gap between securities fraud, and not designing a test as robust as they could, with the unwritten knowledge that this sort of hack was a possibility, or even that this hack was likely.
I think the quality of my comment more than adequately matches the quality of what I was responding to. "It's marketing" is a valueless comment to make and adds nothing.
To anyone reading this thread who might actually be interested, here is a slightly edited comment I have made elsewhere on the topic:
Why are you bringing up marketing at all? If the model is capable of it, then the motivation for why OpenAI is talking about it/reporting on it is completely beside the point. Either the capability matters or it doesn't. If the capability doesn't matter, or doesn't matter in the way that some particular person is claiming, then say that and explain why. Just saying "marketing stunt" adds zero value to the discussion.
If you think that the capability *doesn't* exist (yet), then you are essentially saying that OpenAI is just straight up lying (and, in this case, so is Hugging face apparently), and I wish you'd just say that as well.
I'm very open to arguments about why we shouldn't be concerned about this event (although my prior is very much that we should be, not so much because of the capabilities themselves, but because of how fundamentally misaligned this demonstrates the models are), but I am completely over listening to anyone who has nothing to add other than "marketing stunt".
I was just reading about the Hugging Face incident. If it indeed is true that a Chinese open weight model was used to detect the hack, there is some strange irony here.
It's not just some strange irony, it's an outright own goal by the US's restrictive approach to AI. Now when American companies need to defend their own security, they need to turn to foreign tools.
I've met a few normie Trump supporters who insist that Trump's Iran War will prevent Iran from gaining a nuclear weapon long term. How plausible is this?
Iran doesn't need a nuclear weapon to become a regional hegemon. Using nuclear weapons instead of conventional ones is ... just dumb. They contaminate the entire region, which is bad for your citizens as well. Iran's missiles are winning the war against "The World's Superpower."
Long term, as in 5-10 years, not plausible unless Donald Trump escalates the war to "bomb them back to the stone age" levels, and not certain even then.
Short term, if Iran can still access their half-ton or so of 60% uranium, then it's unlikely that this delays by more than a few months Iran's acquisition of maybe a dozen nuclear weapons, but does set back by a few years their acquisition of a hundred nuclear weapons.
If we have put the existing 60% uranium beyond use, or if we somehow manage to do that in the future, that plus the destruction of many centrifuges, that could delay Iran's acquisition of any nuclear weapons by maybe two years.
Bottom line, Iran almost certainly has blueprints for a nuclear missile warhead that will actually work, and blueprints for everything needed to make those warheads and the fissile materials that fill them, almost certainly in many redundant file cabinets and USB drives. The industrial capacity to build e,g, P1-type centrifuges is comparable to the industrial capacity to build modern-ish automobiles. Which Iran was producing at 1E6/year as of 2024. They don't need fancy high-tech machinery, or rare and exotic foreign imports, and they no longer need expert nuclear scientists and engineers. Unless we go Full Morgenthau, they're going to have the capability to build nuclear weapons.
They'll almost certainly pinky-swear to not build nuclear weapons if we offer to stop bombing and blockading them and to let them charge "navigation fees" for the Strait of Hormuz. But that promise will not be realistically enforceable. Any deal Trump can get, will be inferior to the Obama deal because the Obama deal was backed up by Britain, France, Germany, China, and many other nations who all agreed to give the United States sole control of the "Embargo ON!" lever. Also because the Obama deal was made with an ayatollah who probably didn't want nuclear weapons and was just looking for the best possible deal.
Trump, having alienated almost all of the US's traditional allies and having killed the ayatollah who ordered the fatwa against nuclear weapons, doesn't have those cards to play. And he doesn't have enough bombs and missiles to change the things that matter, unless he starts using America's nuclear weapons.
And on the political side, Trump's and Bibi's war is a great showcase for why a country would pursue nuclear weapons in the first place.
I live in Germany. If you asked me if Germany needed the bomb, I would reply "fuck no". But if I was living in a bizarro reality where France was bombing our schools, infrastructure, and political leaders whenever they felt like it, I would probably come around to recognize the benefits of mutually assured destruction eventually.
At the moment, Iran has the ability to squeeze the balls of the world economy whenever they like, and that used to deter rational world leaders. But it is also clear that the Gulf states are building pipelines to bypass the strait. Iran will still have the capabilities to attack fossil fuel industry directly, but this seems less reliable.
So for the Iranian leadership, it is entirely rational to worry about a future where an US president can bomb them without also taking a hit on the markets. Being able to say "we have the capability to escalate to nuclear war, do you want to gamble the population of Jerusalem on your missile defense system?" seems not obviously the worst way to go about this.
More generally, the more the superpowers (looking at the US, but also Russia here) defect from the ideal of the rule-based international order in favor of "the strong do what they will, the weak suffer what they must", the larger the incentive for countries to gain more tangible deterrence, even though that results in a much worse equilibrium overall.
MAD is not premised on a single nuclear bomb (which Iran likely has).
Pipelines are extremely vulnerable to missiles, which Iran has in spades. Nobody's clearing the "Iran can destroy our entire infrastructure" hurdle in the middle east anytime soon. (and this all with conventional weapons).
Tangible deterrence is best done with conventional missiles, if at all possible. Nuclear Korea could lay waste to South Korea at any time, using only conventional weapons (Kim's nukes are, more or less, a vanity project).
Obama made a deal with Iran to prevent them from getting a nuclear weapon. Trump ripped it up when he took office. The best case scenario now is far worse than the *status quo* he inherited.
Everything Trump has done has made it much more likely that Iran will get nukes. Repeatedly breaking agreements and launching sneak attacks will do that.
I think the Trumpy sphere would scoff at the idea that Obama's deal was going to be effective at stopping them from getting a bomb. I don't know the details of it but I think the level-headed consensus was that it would have been effective.
The main criticism of the Obama deal is that it really did prioritize specifically nuclear weapons. Stopping Iran from funding terrorism? Not interested. Stopping Iran from developing long range missiles? Not the goal.
The core of the deal was that Iran could do what it wanted, just do it without nukes.
The thing is, "funding terrorists" and "developing long range missiles" are largely irrelevant compared to getting nukes, but might be important enough to Iran to sink the agreement wholesale. Then you get nothing, good job. Better to focus on the important thing and getting that right. So yeah, if that really was the main criticism, that was a dumb as hell thing to criticize. Nuclear non-proliferation is good, but the US really need to learn when to respect other countries' sovereignty otherwise.
I'm not sure why you'd think that "long range missiles" are largely irrelevant compared to getting nukes. Iran has devastated American power across the Middle East, and all using conventional missiles (and scuttlebutt has Iran preparing to hit Ukraine, just to test their missile capabilities)...
A long-range missile is just a conventional weapon. Countries are going to have those if they want, the technology is just too simple. Historically, attempts to unilaterally restrict a potentially belligerent country's conventional forces have been made (e.g. restrictions on post-WW1 Germany), but never successfully so. They are inevitably considered too serious a breach of their sovereignity, and if you tie those attempts to what you actually care about, like nukes, then the restriction on nukes are also more likely to be ignored.
Also, let's not ignore why Iran has "devastated American power across the Middle East". America's attack on Iran is the perfect justification for having those weapons in the first place. I would wager that US-friendly, dovish voices in Iran have been discredited for decades to come. All that overblown Iranian propaganda about America coming to destroy them has finally come true. What bigger boon for the regime and its ideology could there be?
The Obama deal demonstrably *was* effective at stopping nuclear weapons while it was in force, and for a few years beyond. It would likely have continued to be effective because it offered Iran something it rather needed (relief from globally coordinated economic sanction) in exchange for something Iran didn't need (actual nuclear weapons), without touching on anything else the regime felt critical to its interests (the terrorism, the missiles, the death-to-America rallies).
Note that, for the entirety of Khamenei the Elder's tenure as Ayatollah, the only thing Iran ever did with its nuclear arms program was try to get the best possible deal at the negotiating table in exchange for freezing the program at "if we really had to, like if you guys elected some populist demagogue with a hard-on for ayatollah-killing, we could put together some nukes in a year or two". And they were willing to accept real verification and enforcement of that promise.
That was a good thing. I'm pretty sure that in the next few years we're going to wish we still had that thing. But it's gone, probably beyond hope of recovery.
Is your logic premised on the idea that Iran will use its nuclear weapons? Or is it just that "Iran as regional hegemon" necessitates Iran having nuclear weapons? I don't see either of these as high probabilities.
Have they stated any mechanism by which that would happen?
Generally speaking, Iran has the technology and knowhow to build nuclear weapons, and that probably cannot be removed. Actually building one would require highly enriched Uranium, as everybody knows, of which they had some enriched to about 60%, the whereabouts of which is not public knowledge anymore, and the centrifuges to get it to around 80% IIRC. Many of the latter had been destroyed in summer 2025, so the Iran War of 2026 cannot really claim that achievement for itself. They probably destroyed a couple more, but it is unclear how many they destroyed, how many are left, and how fast Iran could build new ones. It does not seem those extra-deep bunkers were certain to be destroyed in summer of 2025, and I have not heard of them being targeted with those special bombs again.
"Long term" is a long time.
Possible ways to keep Iran from getting nuclear weapons long-term include:
a) a deal with a really good supervision routine
b) a changed regime that has no interest in NW or enough interest in bargaining with the West AND that stays in power long-term
c) Shock and Awe that makes the regime give up on the idea forever out of fear for the consequences
d) complete military overmatch that allows the US/the West to bomb any place suspected of harboring centrifuges long-term.
I think the Iran war of 2026 has shown that c) isn't happening and d) would be too expensive for the US and allies to keep it up permanently (they could, perhaps, but don't want to). Regime-chage in the Middle East does not have a great track record of producing long-term stable competent regimes open to the wishes of the US and partners, so I am not confident that if the regime fell, which it does not seem to be doing, it would lead to a better controllable situation.
I guess that means that those people in favor of the 2026 Iran War for reasons of nuclear weapons prohibition believe that a peace deal coming out of the conflict will include a nuclear weapons ban including a surveillance regime that is a lot better than the one Trump ended. That is not what my tarot cards are telling me, but I am no expert.
Removing scientists is something that is routinely done by spy agencies. Iran is hardened against this, because their borders are closed, but I wouldn't be so sure that "the knowledge to make/maintain nuclear weapons" can't be removed.
Please note: the word maintain is doing a lot of that work. Pakistan and India likely no longer have functional nuclear weapons.
I recently read The Achilles Trap by Steve Coll. It explained how the intelligence community got it wrong in Iraq- some people think Iraq never had WMDs. Actually they had plenty of chemical weapons and a nascent nuclear program that could have built a bomb whenever they chose to. But they dismantled the whole program between 1991-1994 after their defeat in the first Gulf War, under the close watch of UN inspectors.
Basically, thanks to the excellent inspections regime, the West's plan actually worked. But it was impossible to prove a negative. So the inspectors kept coming, demanding more and more access, including into places where Saddam hid his relatives and friends, so when inspectors came the guards acted really paranoid (afraid the nuclear inspectors might actually be assassins). So the West was convinced they had something to hide and saw every event and every statement the Iraqis made through a lens of confirmation bias.
(and Saddam made plenty of wacky violent statements that they could read alarming things into, bc he was like that.)
The former head of the Iraqi nuclear program produced a 500 page document, submitted it to the UN, explaining and providing evidence of how the program was fully dismantled, and no one believed him. There was just no way to prove that they didn't have a secret bomb somewhere that the inspectors hadn't found. So, Bush invaded and the rest is history.
The book made me realize that the same thing is true in Iran- no matter what inspections regime gets put in place, Iran will never be able to convince anyone that they don't still have a secret bunker with more weapons as long as crazy hardliners that shout "marg bar amrika" and make every schoolchild repeat it, are still in charge. (incidentally there is a bar in LA, where lots of Iranian expats live, called Marg Bar, for margaritas presumably, but it amuses me)
Saddam Hussein was a crazy evil hardline nutter too, but he actually did dismantle his nuclear weapons. We have no reason to think Iran would do the same even if they did we would end up acting as if they hadn't, so it's just a waste of time to repeat the same plan and think it would lead to peace in the region.
I don't have a solution to this problem myself, but anyone saying the goal of the war and the win condition is setting up robust inspection regime is barking up the wrong tree. That isn't a win, that's just postponing the next stage of the war.
Nu? What if Iran already has a nuclear weapon or two? They have no need to use them, and a single nuclear weapon is only good for blowing up one city or so...
what im saying is, setting "iran has no nuclear weapons" as the win condition for a war, is the equivalent of saying "we've won the war if iran doesn't have any mice." You could go throughout the country with the pied piper, but your work will never be done, there could always be one hidden somewhere. Iran is a big country with a lot of uninhabited salt desert and craggy mountains.
You can never prove it to anyone's satisfaction while the regime itself remains unchanged, therefore you can never truly win the war. Pick a different win condition.
Not that you can get "Mr. Final Pravda" to admit to anything, mind you kindly, but if you can get folks off the record, there are actual win conditions (these include "free strait", "impede china's oil flow -- or wreck it irreparably", and "prevent Iran from acquiring more conventional missiles than we can reasonably block").
You'll note none of these have jack damn to do with nuclear weapons (mainly because Iran doesn't want a weapons program (has no need for one, they can kill the entire middle east without a single nuke), and likely already has nuclear weapons).
Then they need to start saying those other conditions on the record! (but they are probably too embarrassed because those conditions aren't being met either.)
I think the number of missiles they used up would also be a win, if they dont have enough left to do more damage, but then how could they be stopped from rebuilding more after the war....
or get them to stop funding violent groups outside iran, that would be a positive outcome.
It's kind of strange, you would have thought knocking out the entire leadership and sub-sub leadership would have been enough to win but clearly it wasn't. could that have been better predicted in advance? They probably were more effective at it than they even expected to be, but it turned out to have been the wrong goal. There were memes about offended Iranian officials who were mad to be still alive, because it meant Israel considered them not important enough to get rid of. It was that successful. So why didn't it work?
Nit: Iraq had plenty of chemical weapons through 1991, but it was never the case that they could "built a bomb whenever they chose to". That would have taken 5+ years of deliberate, sustained effort. Iraq bet early on plutonium breeder reactors as the basis of their nuclear program, which Israel could and did destroy. They were late to the party on gas centrifuges for uranium enrichment (which can be dispersed and/or buried beyond reach of practical destruction), and still many years away from a bomb by the time the desert stormed all over their plans.
>"Regime-chage in the Middle East does not have a great track record of producing long-term stable competent regimes…"
Presumably a successful nuclear weapons program requires both stability & competence, in which case regime change could suffice to thwart such efforts.
I think it is basically c) "now they know we will bomb the shit out of them if they try and build anything" or maybe that bombing has destroyed the capability to move material to where it needs to be undetected.
If my math is correct, Iran's current supply of highly-enriched uranium, (enough for a dozen bombs) can fit in three K-bottles, suitable for transport in a donkey-cart. Good luck figuring out which Iranian donkey-carts to bomb.
I will note that *they* do NOT know that "we will bomb the shit out of them." I can imagine many possible future presidents who they would have good cause to believe would not bomb them.
This current war may have pushed back when Iran can get a bomb. Maybe a lot. So I suppose folks can fight about what "a long time" means.
But there is no reason to believe that the folks running Iran think US policy towards Iran is stable from one president to another.
There is also the problem that it assumes that 1) we will know when they start building a bomb and 2) they care if we bomb them [the leadership likely cares more about maintaining power] or 3) bombs physically prevent building a nuke
And also that if we are going to bomb them either way, then they might as well build a nuke so we can't bomb them any more.
Grandelf finally has pants! Or, well, clothing of some description; granted I have *not* seen his nethers in this trailer so he could still be footloose and pants-free.
Remember how Galadriel told us she laughed at her husband at a dance because of his armour? Well we're going to get that scene, it would seem.
And hang on to your wingéd helms, we're finally getting Celeborn! Now that is a conversation I want to see:
"So... what were you doing for the past few centuries while we were parted?"
"Oh, nothing much; driving my soldiers to mutiny, deciding to swim the entire Atlantic Ocean back home, getting into a situationship with Sauron, macking on our future son-in-law... you?"
I Can't Believe It's Not Saruman is still lurking and making dark pronouncements. Now, he *could* be one of the Blue Wizards, or he *could* be a completely unrelated totally new Dark Wizard of Rhûn, but come on - do you think this lot have the imagination for that?
There's another friendly Orc. Possibly he's meant to be a corrupted Elf, but we'll see. After the family man Orc got stabbinated by Sauron last season (was it only two years ago? how time flies!) there clearly was a vacancy in the cast for Family Person Orc.
Do you want to see more scenes lifted straight out of the movies? Of course you do! Well, we got you covered there, my friends: relive the Hobbits hiding from Nazgûl sent out to find them in the Shire with Friendly Orc, the Plank of Wood known as Arondir and Galadriel taking cover in the undergrowth!
Nazgûl, sez you? But is this not the wrong time for them to be popping up? Nah, don't sweat it: the timeline of this mess is already corrupted to Hell and back, so it's just going to be a recapitulation of Peter Jackson's Greatest Hits. In the wonderfully creative tradition of Payne and McKay that brought us "they're not Hobbits, they're Harfoots" we get "they're not Nazgûl, they're Nazgnagôl" which is a back-translation of "Ring Servants". Because they're not yet Ring Wraiths, you see. Work like this is worth every cent of that one billion, isn't it?
Oh, and *finally* Sauron gets off his brooding backside and forges the One Ring!
The Númenoreans look like they are about to do something useful for once. Again, Ar-Pharazôn looks pretty good in the trailer, and again he seems to be the one person operating with at least half a functioning brain. Dear Show, you are *not* supposed to be making me cheer on the bad guys. Really, you're not.
Círdan grew his beard back, yay! Got over that mid-life crisis and decided shaving with the edges of sea shells is not the best way to update your look, I guess?
Dísa has now been promoted to Queen since Dúrin is now King, after Dad went mano-a-mano with the Balrog last season.
Fell Beasts and Mûmakill and Nazgûl, oh my! all make an appearance. But thankfully no Harfoots or Stoors (yet).
And for your ultimate delectation, we will not alone finally be seeing Anarion, the younger brother displaced in the first two seasons by the invented drippy daughter, but the Balrog speaks! Yes!
"Also in July 2026, the showrunners revealed that the Balrog would return from previous seasons and would have a speaking role for the first time. They announced that Simon Pegg had been cast to provide the creature's voice."
And for a season opener, they'll be dropping not one, not two, not three but *four*, yes FOUR episodes on us at once. Why, it's almost like they just want to get this damn thing out and over with! But with promises of earthly delight like this on the horizon, it's almost worth hoping the AI apocalypse will hold off until the new year, isn't it?
>Do you want to see more scenes lifted straight out of the movies? Of course you do!
Sigh. One thing I really, really want those clowns to understand is that when I want the Peter Jackson LotR movies, I know where to find them. Before the first season of Rings of Power, I was really excited for it because Tolkien left us scaffolding for some really, really good stories in the Second Age but the source material is frustratingly incomplete. A competent adaptation would have been amazing!
A big part of the problem is the J.J. Abrams school of coming up with "epic' moments and awkwardly gluing them together with mystery boxes and memberberries. One of the magical things about Tolkien's writings is how well thought out they are, how deeply the world-building, themes, and plot fit together and support one another. Payne and McKay are doing the opposite of that and they largely aren't even using the plotting and world-building Tolkien had already worked out for them in the material they're "adapting".
Another part, I think, is that studios don't trust their audiences anymore, especially for big budget productions. Subtlety is not allowed because they don't think we'll get it. Depth, nuance, and even consistency are considered wasted effort because they assume we're less than half paying attention. And everything needs to thoroughly cliche-ridden because it's a "proven formula" that they believe they can sell.
If they had competent showrunners and trusted their audience, they could have given us a proper Second Age adaptation that either focused in on a specific plotline that could be told in a self-contained way or let itself sprawl with cast changes for the mortal characters as centuries passed, going from Aldarion and Erendis, to the first War of the Rings, to the Fall of Numenor, to the War of the Last Alliance.
>Nah, don't sweat it: the timeline of this mess is already corrupted to Hell and back
> Tolkien left us scaffolding for some really, really good stories in the Second Age but the source material is frustratingly incomplete. A competent adaptation would have been amazing!
I don't agree with this. The Second Age, like the Clone Wars, is one of those pieces of backstory that is better left as backstory.
The events of the ages prior to the Third are deliberately unrelatable, they're ages of epic heroes when everything was just so epic and heroic, Elves were everywhere and Men were unrelatably fantastic. You can't get relatable characters out of that. Galadriel (say) works when she's a mysterious and powerful presence who shows up briefly, but it doesn't work when she needs to be a relatable POV character whose journey we follow; either she's constantly wonderful in which case she's unrelatable, or she's too normal in which case she's no longer Galadriel.
That timeline is very funny. But can anything in the third season top the bit in the first season where Celebrimbor, greatest smith of his generation and second only to Feanor his grandfather, was stumped by the problem of forging the rings until a half-dead scruffy mortal initiated him in the secret technique of "alloys"?
True, but maybe he was hungover. We know from The Hobbit that Elves can get drunk and pass out, although Bilbo didn't stick around to see how Thranduil's butler and chief guard felt the next morning.
You skip *one* class because you want to go frolicking in the May meadows and you think you can make it up later, and then two thousand years later it comes back to bite you in the forge.
But don't you want to know if Grandelf gets a pair of trousers? Don't you want to relive the thrilling sequences that we totally are not ripping off from "The Fellowship of the Ring", cross our hearts, when the Hobbits were hiding from the Nazgul? Don't you want to see if Fake Invented Daughter of Elendil and Fake Invented Son of Ar-Pharazon finally get it together and settle down to breed a family of snivelling little moustache-twirling mini-villains before the giant wave strikes (and it can't strike too soon?)
Though I wouldn't bet on that last, they don't seem to be in the cast list, thank goodness.
Erica, surely you are on tenterhooks to know if the Harfoots and the Stoors make it all the way to the Shi- er, that mysterious green land of not-Hobbit legend and meet up at last with the Fallowhides (who then presumably slaughter them all in an Aryan pure bloodlust genocide so only the white Halflings survive to colonise the fair green land in time for the later Third Age where the Sh - er, mysterious land of legend is full of only white Halflings)?
Do you not want to know how Galadriel juggles her polycule with the return of Celeborn now that she's in a thing with Elrond, in an on-again/off-again thing with Haldir/Sauron, and who knows but she might be open to including a Family Person Orc of mystery as well? And of course Arondir is available now his human squeeze took an arrow to the chest and died of "the actress left before the second season was shot"?
EDIT: Whoops, I meant Halbrand! Halbrand, not Haldir (although... Glads here is throwing her eyes at every half-passible guy in the cast, so who knows?)
Where is Miriel? Does anyone know? Does anyone care?
They could have had a decent go at "okay, we have to do a heavily compressed timeline of the Second Age leading up to the War of the Last Alliance and the downfall of Númenor, but that's okay, we have five seasons and with some care it can be done". But they also, I imagine, were under commercial pressure to pull in new Prime subscriptions, so it had to be "people are fans of the movies, they want to see the characters from the movies, you gotta put in the names they know and the callbacks to the movies!"
As well as Payne and McKay being talentless hacks who only got the gig through J.J. Abrams whose powers of screwing up long-established and well-beloved franchises are truly awesome.
I can hardly wait. I'm just grateful that you're watching this so I don't have to. I'd rather read your rants on the subject, which are shorter, more entertaining, and better-thought-out than the episodes at least up to the point where I gave up on watching.
If you are a very big Tolkien fan, it might be painful/sad to watch it, but we've had great fun laughing at the flaws of the series. There are worse ways to spend a few nights imo.
And yes, this really is how the geniuses of the adaptation came up with a backstory for "How Gandalf Got His Name". By this stage, you can only laugh, crack open another 75cl of Tesco's own brand cream sherry or ruby port, and search for the piss-take reviews on Youtube.
Oh, I am waiting. With keen anticipation and a bottle of Sarson's finest malt vinegar by the screen.
I don't see this turkey surviving for a fourth season, but it depends how dogged Jeff is about "I really love and adore - reads name off hand - Tawkeen?" and if they've signed too many contracts binding them in the Black Speech to deliver all five seasons.
Erik Kain, God bless the man, returned after a year's absence for many reasons just in time to be hit between the eyes with this trailer and the mysterious floating sword:
I’m surprised Bentham was ever suggested as the first to use a self suffix. It’s one of the reasons that Hand D in the Thomas Moore play was attributed to the Bard
> With self-same hand, self reasons, and self right…
"Kahl thought “a researcher coming from a different planet would find nothing particularly special about Mini Munich.” It would seem obvious to them that if adults spend their days starting businesses, running newspapers, electing governments, and settling disputes in court, then a miniature version of that would be a natural way to prepare children for life. In the scaled-down world, cause and effect become easier to see—businesses succeed or fail, policies reveal their consequences, and “a bad law gets repealed after a few days, not after decades or centuries.” Rather than teaching civic and economic life as abstract subjects, miniature cities allow children to experience them firsthand."
We have the live steam model railroading community, but not many young people in that (except at passengers during public run days). Train Mountain in Oregon is the big one, but they're all over.
I've thought for a while about the "you live in your mother's basement" insult and how weird it is, how bizarrely specific and reliant on assumptions from a number of seemingly incompatible ideologies. I'd like to know what people think of my analysis below, because most examinations I've seen of it seem incredibly shallow and mindless.
So first, there's the gender aspect. Mostly it's "your mother's basement", a significant amount of the time it's "your parents' basement" (I think someone said this in a recent ACX thread), but as far as I can remember it's NEVER "your father's basement". What's going on here? I think I've seen feminists explain it as simple misogyny, the idea being that a grown man living under the authority or control of a man is fine but living under the authority of a woman is horribly shameful. (But then why is "your wife supports you" not an insult of similar strength?) But I think I've seen manosphere types describe it as misandry, as equivalent to "neckbeard" in that it's only used against men and the idea, I think, is that it's based on the assumption that men (unlike women) have value only in accordance with their economic success and social status. (But then why the focus on the mother?) And a comment on this page https://www.explainxkcd.com/wiki/index.php/713:_GeoIP suggests it's an implication of incest, which would explain why it's usually "mother" alone, implying she's a single mother to make the incest insinuation work. (But I haven't seen this interpretation very often.) So which is it? All of them at once???
Then there's the class aspect. I know a guy with rich parents who was living for a while in their "spare penthouse" while figuring out what he wanted to do with himself. There are lot of things you might be insinuating by mentioning this, but none of them seem anything like what's being insinuated with the basement. The usual form would be as an angry accusation of being privileged and out of touch, not as a mocking insult of being a loser. But they're the exact same thing except for how rich your parents are. So it's an anti-poor insult in part? No, because it relies on your mother/parents owning a house; "you live with your parents in rented/public/charity housing" just doesn't have the same impact. Maybe the idea is you're inconveniencing your parents by taking their basement in a way you're not by taking their spare penthouse, but then why is it "basement" which mitigates this by implying you're out of the way in an unneeded room? "You sleep in a cot in the corner of your parents' bedroom" would make this accusation much clearer as well as making the "adult baby" effect much clearer. So it seems to rely on your parents being both not too rich and also not too poor for the insult to work. The only thing I can think of is that it's based on a kind of class propriety, of a man born into the middle class failing to live up to the middle class norms that would not apply to him if he were of a different class. This is really weird for a 21st century western insult of such prevalence, but it's the only reading that seems to make any sense.
And finally there's a subtler point. The insult seems to be based on "capitalist logic" in that it demands self-sufficiency and proxies that with what you can afford. But normally capitalism, when *explicitly* defended, is defended based on individual freedom and the greatness of non-coercive voluntary arrangements. Unless the implication is that the guy is physically threatening his mother to let him stay in her basement, that's exactly what this is. An accusation of being on welfare would work vastly better on a normal capitalist logic, while on any kind of socialist logic it would be pretty verboten to define someone's value by what they can personally afford, even separate from their own family. The only way this particular shaming makes sense is under a kind of "collectivist capitalism", for want of a better term, whereby a virtuous person is obligated to be a productive member of society and not benefit from any voluntary arrangements of either charity or excessive family support. I imagine this attitude is actually quite common, despite rarely being explicitly defended in political debate, and I also imagine it's particularly common in East Asian cultures, though those cultures are also the least likely to use this kind of insult, what with their strong family-oriented ethic.
(The explainxkcd link above also claims this insult is specific to North America, which I'm very suspicious of because most claims of this form turn out to be false, though I don't think I've heard it here in Australia, although I'm not very social and it seems like the kind of thing that would only be said on the internet anyway. Except for the superficial aspect of having a basement, which is I think mostly specific to houses in (I forget which part of) North America.)
So, in the end, my very rough theory is that this insult has naturally evolved to almost perfectly straddle the line between many incompatible ideologies, without clearly endorsing any of them. It has just enough misogyny to appeal to misogynists and just enough misandry to appeal to misandrists, just enough "you're privileged" to appeal to poor people and just enough "you're poor" to appeal to rich people, just enough "moocher" to appeal to individualists and just enough "unproductive and unvirtuous" to appeal to collectivists, and this is why it's become so prevalent. But I have low confidence on this.
I'd like to know (a) what people think of my analysis and (b) what other insults people can think of that have this trans-ideology effect. I wonder how many other cultural phenemona generally are like this, where adherents of opposite ideologies can say that it's a sign of how far society is away from their ideal, and all of them be correct.
When I was growing up, these were 2 separate insults: "you [as an adult] still live with your mom" or "you [a child or immature adult] [engage in low-status hobby] in your mom's basement".
A basement seems like an entirely natural out of the way place for a hobby that takes up a lot of space or is messy.
I think there might have been a transitional stage (not entirely sure) where "living in the basement" implied you were so obsessed with your low-status hobby that you lived in the place you do it, similar to how a man (I think this was always a male-coded insult) who was obsessed with fitness might "live at the gym" or a man who was obsessed with his car might "live in his garage".
At some point (possibly through the "living in the basement" intermediate step), the basement insult and the living with your mom insult merged into "you live in your mom's basement".
It's "mother" because the holdover from traditional gender roles is that your father might offer you whatever help you really need, he's doing it in the expectation that you'll use it to finish growing up and becoming a productive member of society that the family can be proud of, whereas the mother is coddling a child that she never wants to grow up and leave the nest.
And it's "basement" because, as Erica Rall has noted, in most North American homes the basement is the place you can fit a quasi-apartment that gives a cheap imitation of independent living to the overgrown adolescent who can't actually live independently.
> But then why is "your wife supports you" not an insult of similar strength?
In manosphere it most certainly is, but it is not used nearly as often as "you live in your mom's/parent's basement". My impression is, people use the former only if they know (or are reasonably sure) that the wife does in fact support the husband in question. Whereas the latter gets thrown around with total disregard for the statement's veracity, much like sixth-grade boys calling each other "fag".
I expect that men who both subscribe to a version of masculinity that feels threatened by the woman being the primary breadwinner and don't earn very much would tend to be single.
If person A is an insecure manchild who thinks that telling person B "Your wife supports you!" is a great insult, does not mean person B is the same. My wife earns more than I do. If someone were to insult me in this manner, I would just laugh at them.
In addition to what everyone else has said, I think a side connotation of the insult is that the insultee is a lesser son of greater fathers. That, at their age, their parents were already busy with the accomplishments they are now mooching off of. So the guy living in a spare penthouse may be living in material comfort, but he's still a failure *as compared to his parents,* who were presumably busy building (or maintaining) the family fortune at his age.
If it's his mother's house, why isn't he in his childhood bedroom? The implication is that either the mother has moved on from prioritizing his needs (implying he's unlovable) or he hates the family home so much he'd rather occupy the dingiest, most secluded part of it (implying he's self-destructively antisocial).
As for the gendered aspect, I think it's about setting the locus of power in the household. I have a friend who describes their father as "living in his ex-wife's basement", even though it's a house they bought together. To my ear that hits just as hard, if not harder than the mother's basement.
I think the implication of "living in the basement" and not his childhood bedroom is that he tried and failed to leave and live independently, and now he's back home and Mommy is having to take care of her little boy again.
I think the significance of basements is that they're a largish area that's relatively isolated and generally used as utility and storage space but can often readily be converted into a makeshift apartment. Culturally, basement apartments are an awkward middle ground between actually moving and getting your own place vs continuing to be a regular part of the household. I think the implication of the insult is that someone who moves into a basement apartment in their childhood home is someone who's old enough to want the independence of living separately but not capable or financially secure enough to actually get their own place.
If it’s American at all, it’s because Americans have basements. While we don’t in the U.K, or most of Europe.
Of course it’s emasculating. Hence the mother. It’s referring to a male adult staying at home past adolescence, but kept (or keeping himself) in the basement. Subservient to his mother.
Not an airy upstairs room he might leave easily to go for a jog, but the perma-gloom of an underground abode.
Movies often put the mad scientist in some kind of gloomy lab, underground or, if not, the windows are blacked out, or encrusted with dirt.
Another key component of the insult is the implication that you have no partner (and no hope of getting one) and probably few-to-no irl friends. (And should you have any they are likely all losers as well.)
Regarding the mother vs father: I think there's an assumption here that mothers will be more coddling than fathers, and so if it was 'father's basement' there is more likely to be a legitimate (non-loser) reason for living there. Whereas a mother is more likely to just let her son live in the basement forever, or even demand it in some cases (cough).
I'm pretty sure that "parents' basement" was either the original form of the insult or at least that "parents'" and "mother's" were used interchangeably in the past. As to why "mother" seems to have displaced "parents", you might be right. There's also the related idea that it's unmanly for older boys and young men to be reliant emotionally on their mothers, since mothering is culturally associated with the care of young children and boys especially are supposed to outgrow being "mothered" as they progress from childhood towards manhood. So the "parents' basement" insult is getting the "go back and cry to your mommy" insult layered atop it.
Alternately, there might be a layer of insult directed at children of divorce or children whose parents were never married. By saying "mother's basement", you're implying that one's mother and father don't own a basement together, which can have insulting connotations because of cultural baggage around broken homes and illegitimacy.
I always used the threat of moving into my mom's basement as a motivator to keep my sh-t together. But to be fair, my mom's basement was full of random crap and flooded a lot of the time, so it was pretty unappealing....
I believe it builds on the Eric Cartman-esque stereotype of a son who abuses his mother's generosity and does nothing but take from her.
You touched on the implication that the man's mother is a single parent while discussing incest, but it's also important in that his mother is more vulnerable without a husband. There's a stereotype that opposite-gender parents are more lenient and permissive.
I also think the basement is important to this reading. A basement is separated from the rest of the house. It emphasises that the son is not taking part in the daily life of the household; that he is a parasite who hides in the basement while his mother socialises and works above. It's also lacking in sunlight and fresh air while being much larger than a bedroom, implying a willingness to trade comfort for the space to avoid others. This is why the piss bottle and mini fridge are relevant to basement dweller insults. The son goes to extreme lengths not to participate in daily life.
It's partly a holdover from when housing costs were dramatically lower and you had to be some extreme mix of lazy/stupid/incompetent/maladjusted to live with your parents into adulthood in most cases.
The reference to mother's basement makes sense because there is a pre-existing and similar "mama's boy" stereotype of a meek man who never strikes out on his own and remains dependent socially or financially on his mom.
The insult persists because it is still directionally correct in indicating lack of success/discipline/effectiveness in life. Although living with parents well into adulthood has become significantly less stigmatized as long as you actually have a job and are contributing in some way.
To illustrate how much things have changed: Breaking Bad aired its first episode in 2008. In the episode, the audience is supposed to pity Walter White's financial situation. But he was able to afford a large three-bedroom home with a pool, a stay-at-home wife, and a child, by teaching high school and doing one part-time low-wage side gig.
Re: Walter White, yeah, that's the movie/TV notion of "just about making do". The 90s sitcom "Grace Under Fire" was the same; the premise is "Grace Kelly, a single mother and recovering alcoholic raising three children in a small town in Missouri after finally divorcing her abusive husband" and "In the first three seasons, Grace operates pipelines at the local oil refinery" which you could probably make a decent living doing.
But the locations (when you saw them) were "single divorced mom of three living in detached house on own property with plenty of room and plenty of whatever anyone might want" and it made me laugh (not at the jokes) to see "so this is the Hollywood idea of struggling lower middle class, huh?"
I don't think it is just Hollywood logic. I remember from my own childhood, there were dads who were truck drivers, construction workers, etc, and they could afford to own or rent their own homes with a stay-at-home wife and kids. Sometimes the wife would do a small low-wage side gig like substitute teaching or watching the playground at recess to pitch in.
Yeah, but that's a stable family. Grace's situation was that she was (1) divorced (2) recovering alcoholic and (3) had three kids. Now, the job they gave her was blue-collar but semi-skilled (at least) and traditionally male, so she would have been earning a good wage.
But the circumstances didn't seem right for what she was supposed to be. Now maybe before the divorce the husband had been earning well and they were able to afford the house etc. and she got it after the divorce, but the tone of the show, at least to me, was meant to be "relatable working-class family on the lower end of middle class". The look of the show, however, was not "struggling single mom trying to rebuild her life" but I accept that after the first season or so it dropped all that and just became another sit-com (and apparently went down in a blaze of controversy at the end? https://www.youtube.com/watch?v=-LiaR4n4HFI)
I agree with the comments about Roseanne, that was much more realistic-seeming in the size of the house, furnishings, etc.
You're overthinking it. It's like calling someone a mama's boy, only applied to an adult. It's saying that the person is coddled, unable to take care of themselves, and a man child.
> The usual form would be as an angry accusation of being privileged and out of touch, not as a mocking insult of being a loser. But they're the exact same thing except for how rich your parents are.
They're the same thing except one is a basement and one is a penthouse; one is a nice living situation and one is damp and dark. The insult is not just that you're dependent on your parents but that your living situation is particularly bad. You're not even living in a proper bedroom, you're literally beneath them like they're ashamed of you.
> (The explainxkcd link above also claims this insult is specific to North America, which I'm very suspicious of because most claims of this form turn out to be false... Except for the superficial aspect of having a basement
I don't think that's all that superficial.
Overall I don't think it stands up to heavy analysis. Sometimes a label just emerges an insult for a particular group of people. The insult "Karen" isn't actually about having the name Karen, and it's somewhat arbitrary that it wound up as Karen instead of Sarah or Megan. A "basement-dweller" isn't just financially unsuccessful, he's uncomfortably immature and probably obsessed with something that most people find offputting.
Just want to applaud these two short summaries each of which is spot-on target::
"The insult "Karen" isn't actually about having the name Karen, and it's somewhat arbitrary that it wound up as Karen instead of Sarah or Megan."
"A "basement-dweller" isn't just financially unsuccessful, he's uncomfortably immature and probably obsessed with something that most people find offputting."
This genuinely fascinating analysis is some weird apotheosis of ACX comments - it's a deep dive on something I couldn't have imagined caring about ten minutes ago but now really has me thinking.
One addendum I'd give is the "threaten your mother" thing - I think at least in some circles there WOULD be an implication that someone browbeat their mother/parents with "you gave birth to me so you owe me and I owe you nothing" and their too-soft parents gave in, and that's part of the source of their sucking-as-a-person.
> But then why is "your wife supports you" not an insult of similar strength?
Who says it isn't?
> The explainxkcd link above also claims this insult is specific to North America, which I'm very suspicious of because most claims of this form turn out to be false
The appropriate age to leave the nest differs from country to country. Some think it's shameful to live with your parents when you're a high school graduate, some think it's shameful to do this when you're in your thirties (and single!).
It's an outdated insult from back when the frontier was open and it was legal to build new housing. In those days, a man might reasonably be accused of cowardice for not moving out and seeking his fortune. Now that there is no more frontier and boomers have made it illegal to increase the supply of housing, moving out of your parent's house simply means moving in with a roommate. No thanks.
Indeed....Dad-brag-alert! [advance apologies if this violates Scott's criteria for appropriate commenting]
When my eldest was graduating from college and weighing his next steps I made just two pronouncements:
-- happy to help pay for grad school _provided_ it is towards a specific field of endeavor that you've chosen for specific reasons; grad school as a continued holding pattern is a non-starter.
-- there will be no living at home after college graduation. My argument being that paying rent with roommates is not money wasted, rather it is paying tuition on living in the real world. Had to talk his mom into this one as she was hearing the short-sighted "rent is just a waste" cliches on socials and whatnot, but she took a breath and signed on to the above declaration.
He did go straight to paying rent with roommates (in one of the supposedly "illegal to increase the supply of housing" metros), that went fine and he gained all the navigating-real-life lessons I was aiming for.
In his late 20s he did get a master's in a chosen field that he'd done some real research about, and between his maternal grandfather and I and himself [his day job liked him so much that they begged him to stay with them part-time while doing the degree], it was fully paid for. In his early 30s he's doing well overall including just signed a lease to cohabit with a fascinating woman who may be my future daughter-in-law.
My youngest, now entering high school, will when the time comes have the same stipulations for the same reasons. Though of course they are different young people and early indications are that this kid may not need it (he's startingly self-directed for his age which, my siblings and I agree, has to have come from his mother's gene pool).
"The implication of the insult is that you are incapable of paying rent or living independently."
This.
And (I think) there is also the implication that you not paying rent has a negative impact on the parents' finances. So living in the "spare penthouse" doesn't hit the same way.
There is also the aspect of "living in the basement" which implies "has no social connections in the real world." Living in the spare penthouse doesn't imply that.
Note that "taking care of your mom/dad/parents" can have exactly the same living situation -- you still live at home -- but is virtuous because you are contributing rather then taking.
I mostly know physics departments. You can try and engage with the grad students, they know most of the department scuttlebutt*. Maybe even get to look into their labs... if that's the type of thing you are interested in. The most important thing is probably to find the right advisor/ (professor to work for) the exact school is less important.
Another thought about the recent hugging face incident is that afaict it looks a lot like what you'd need to do to hack an election. Traditionally a big problem with that has been that US elections are handled at many state and local levels each with their own separate system and you'd need massive parallel coordination to hack them all (and even more coordination to make the results not look obviously manipulated), but an AI swarm seems like a pretty good way to do it (without even requiring other humans in the loop, so long as you have enough AI access). And in a world where partisans on both sides seem increasingly convinced the other side is going to manipulate elections, there's more likely to be someone with both motive and access to do it.
(Funniest result I guess would be if somehow people in both sides do it at once and the next major election just becomes an AI hacking duel).
>you'd need massive parallel coordination to hack them all
That's where voting machines come in, from a relatively small number of vendors with lax standards. They have been around for decades at this point and they've always been a security concern to those willing to look and listen. So no, you don't need AI to hack them. And no, you don't need to hack all of them either, just a few in the right places.
I deeply wish we would return to in-person, inked finger, punch-card/hanging chad, full-ID-checked voting, with no electronics whatsoever. Everyone is afraid of hackers, and there's almost no system that's hack-proof or furthermore is widely understood and believed to be hack-proof.
My most right wing view is that requiring voter ID is completely reasonable.
It's really odd that voting is *the* vital process of a democracy, yet there's absolutely no requirement to prove your identity.
I've been into election security ever since my hometown had a Chinese national (non-US citizen) vote in the 24 elections. The Secretary of State announced that because his ballot was scanned, there was no way to invalidate it and his (illegal) vote would count.
Ah, Michigan. One of those "Red States" that is captured by the Democratic Machine. How many "mail-in" ballots didn't get folded in Detroit, I wonder*?
(Biden lost black votes compared to Obama... except for a few Democratic Machine cities... wonder why?)
*This is a reference to Tulsi!'s search warrant for Fulton County.
There is nothing wrong with electronically counting paper ballots. Concerns re hacking the counting software can be addressed by hand counting a random sample and doing a full hand recount if the discrepancy will the electronic count is large enough.
You still need to double check the humans' work, both to prevent fraud and for the simple fact that humans can't keep perfect focus while working through millions of pieces of paper. And if you throw more humans at the problem it stops being cheap.
Also, hand counting is much slower, which is a real problem because one of the go-to Republican accusations of fraud has been "look, Trump was winning when the count started, but lost once more ballots were counted, clearly this must be fraud and not different polling places returning their results at different times." It would be much worse if it took multiple days for the count to be finalized rather than hours.
On the other hand, if it did take multiple days to get a result, you could better accept "a whole lot of new ballot boxes came in over that time" rather than "four hours ago X was winning and now mysteriously Y is streets ahead".
It's just not that slow. In Australia we hand-count everything (and also have instant runoff voting, which can mean multiple counts) and yet we usually know basically who has won within a few hours, and have the complete count pretty much done overnight.
As I understand it what's really slowing down vote counts in the US at the moment is the huge growth in mail-in ballots, where there's a need to manually do the security theatre of comparing one possibly-fake signature to another possibly-fake signature.
Well, there's also the fact that the US typically puts many, many times more positions up to democratic vote than Australia does. They vote for President, Senate, House, state senate, state house, Governor, state judges, state cabinet offices, school boards, county offices, city offices, police chiefs, fire chiefs, many other chiefs and God knows what else on a single day's typical ballot I believe. Not to mention the roughly half of states that have citizen-initiated referenda and propositions on every ballot. Not to mention primary elections. We trade speed and security for having a grand total of two federal, one or two state, and one local office up for any kind of vote ever. (Though thanks to IRV our votes are much more powerful and carry a lot more information. Canada and the UK are the worst of both worlds I believe.)
What is "wrong" is that it is a lot easier to be confident that locally counted paper ballots were not hacked. There may be fraud at a local precinct level, but widespread fraud requires enormously more effort than hacking electronic voting. At least that is what lots of people think.
I believe there is value in voters having confidence in the process.
>I believe there is value in voters having confidence in the process.
There is definitely value in that (see my opposition to ranked choice voting). But, whenever the count is slow (as a complete hand count would be), some people claim that THAT is evidence of impropriety.
It's also harder to predict what helps voters have confidence, and taking more care doesn't necessarily help (for example, if your local deli put on a full body hazmat suit to make your sandwich it would not leave you feeling comfortable). Paper ballots might help by virtue of simplicity, but talking loudly about security probably harms voter faith.
(For another example, taking forever to approve COVID shots didn't help prevent fears about it at all).
That's predictive programming, in the sense of "full body hazmat suit" means Bad Problems Incoming. If your local deli was using a cleanroom-grade air purifier, you'd feel more comfortable. And it would help with eating less fecal matter.
Also certain motivated actors are invested in attacking that confidence for their own advantage.
Regulating your elections to a "what makes everyone *feel* secure" standard, rather than a "what is secure" standard, invites Purple Party adherents to simply find themselves unable feel the elections are secure unless and until elections are structured in a way that ensures Orange Party adherents encounter maximum structural difficulties when voting.
Better to structure for actual security, and then treat feelings issues as feelings issues; something to be solved with many more parts education than structural change.
Then you wander off into territory of who does the hand-count, what if they use electronics to summarize the statistical sampling, what was the algorithm used, who approves the legality and process across 50 state jurisdictions, etc. etc. etc. Once you introduce electronics, you introduce a possibly unacceptable level of doubt and fear.
Not saying any of this is guaranteed better with paper, but it worked for decades without suffering trust issues (except in Chicago and Tammany Hall, fine).
Robert X. Cringely wrote an article in the early 2000s addressing most of the issues with hand counting. Each of the major parties are allowed one observer of the counting, each ballot is examined by at least two people, etc.
Hand counting worked for decades and can be made visibly robust (if not perfect).
I genuinely don't understand why anyone would expect China or other countries that feel threatened by non-AI concerns (e.g. Ukraine) to hit Pause on AI when it offers them hope of succeeding in their offensive or defensive suite of capabilities.
Even if they appeared to genuinely accept a pause, how could you ever verify that they had?
The genie is out of the bottle. Now that this capability is known to exist, it will be created, in secret as well as in public. AI alignment efforts may well work great, but a nation-state or powerful crime organization will turn off the alignment and reap the rewards. Now or 10-20 years from now, it will happen.
Although it could help their military, it could help other countries' militaries just as much or more, leaving them no better off. If they're much worse at AI than other countries (eg Ukraine, which is far behind the US and China), then it makes them worse off relative to those other countries (although Ukraine could still potentially benefit if the US gave them powerful AI and nobody gave Russia powerful AI).
I think that, while Ukraine is indeed much worse at LLMs than China, they might be currently leading the world in non-LLM machine learning applications -- e.g. those needed to pilot autonomous drones.
I understand the argument in Plan A, but I most certainly don't trust China remotely enough to think it would work. Even if it all appeared to work superficially, it's a scarce and critical resource over which control can make-or-break an empire. That is how wars start, not how they end.
You've put a lot more time and thought into this effort than me, so no need to reply, but here's my most immediate concern:
"Establishing control over the chip supply is easy. Only a few companies can design AI chips (eg NVIDIA in the US, Huawei in China), and only a few factories in the world can produce them (eg TSMC in Taiwan, Intel in the US, SMIC in China). All of these companies and factories are in the US, China, or client states kept on very short leashes. The US and China simply tell these companies and factories to send chips only to licensed customers"
Why wouldn't China just massively invest in Huawei and SMIC (for example), invade Taiwan to take over TSMC, generate the next generation of top-performing chips first, send them only to Chinese military clients, and game over for everyone else? What's in it for them not to build an all-China supply of AI dominance? How would I trust them not to do this, especially when I've concentrated everyone else down to just the US?
Why wouldn't the rest of the world consider the US just as bad as China, and have this break up NATO and divide Europe from North America? When Brazil or France want chips and have to be US or Chinese supplicants, they will defect as well, and in many other dangerous ways. I just don't get it.
In Plan A it is a (relative) slowdown for about 10 years - they prefer that to a full stop forever for (IIRC) the reasons you give.
There are lots more details in the plan. eg If China invaded Taiwan to take over TSMC then the US would declare the deal void and all the treaty data centres would be destroyed.
Sorry to be dense, but how does a slowdown change the calculus? I'm just thinking through: China sends bazillions of yen to Huawei and SMIC, they get a design and implementation of a faster set of chips on the drawing board ready to be fabbed, quietly build out the fab somewhere (or quietly prep in existing facilities not obviously visible to inspectors), or lock out the inspectors, go full steam ahead on the next-gen chips, and again, game over?
Let's assume NVidia and TSMC and ASML and whoever are all doing the same, but just a hair slower. Isn't it an unstable equilibrium?
Please note that "AI" is a horribly overused term. When you talk about drones for the Ukrainian battlefield, you mean image recognition to discern between defined types of targets (soldiers, vehicles etc.), and non-targets. When the doomers talk about AI, they mean AGI with access to the internet and able to pursue their own goals with a broad set of capabilities.
Doomer pedant here. ASI, not AGI. Human level and barely superhuman AIs are safe, except for their potential to improve. It's the ones actually much smarter than us we're afraid of.
For the purposes of this discussion, AI is assumed (by me) to refer to the HuggingFace incident and related types of GenAI capabilities, though GenAI can also leverage tools that perform image recognition of course.
(Possibly apocryphal but metaphorically useful, history majors don't @ me) Chinese emperors wanted to live forever, and only wanted that, and ended up drinking mercury and dying pointlessly early. If civilizational immortality is actually out of reach, lets go for a long happy life followed by a drawn out death by ordinary terrible cancer, and not spectacular heavy metal suicide. If immortality is attainable, chugging mercury also isn't necessarily useful toward that goal?
living a long time looks like repeatedly not dying yet due to specific causes via a lot of gruntwork.
Well, I'm not at all an AI doomer, though we live in interesting times. We will have to come to terms with this new set of powerful capabilities somehow.
(Pursuant to prior comment) a trader from the 1700s using pieces of paper to perform their business at a snails' pace would have a very hard time coming to terms with 24/7 speed-of-light comms FX trading, but we made the leap over time. This timeframe is very compressed, but it's of similar magnitude to that industry shift.
Random silly thing that keeps popping into my head over the last decade as a puzzle. How can you generate random numbers in your head using ONLY your own thoughts? No external stimulus, and you can't even interact with your physical body to use things like the length of your hairs or your heart rate. Just full brain-in-a-jar.
I have my own ideas based on like "things that you know but you don't know you know until you look" but I'm really curious what other people's gut solutions are.
I'd often frame this as like "you and 99 other rationalists (so you aren't encouraged to assume everyone else will skew a specific direction and actively compensate) are put in that situation and have to decide whether to press a button or not. If between 15 and 25 of you end up pressing the button and the rest don't, you all get a reward" so you're encouraged to give yourself as close to a 20% chance of pressing the button as possible" but if you find that exploitable/distracting feel free to steelman whatever scenario you want. The specifics shouldn't really change the fact you just need to produce a random number of some sort.
Most people have several longish numbers memorized. Pick the last digit of your social security number, phone number, library card, student ID, etc. Doesn't particularly matter which one as long as numbers are assigned randomly or sequentially and you're not selecting based on what that last digit is.
One way might be to imagine a spinning globe of the Earth, stopping for a moment here and there. If you can persuade yourself you could make these stops fairly randomly, with a bit of practice, then the stopping longitudes facing you (and latitudes if you also select a point randomly up or down the line of longitude) could be mapped to digits in some base.
This assumes a far more unrealistically detailed knowledge of coordinates of places round the earth than most people have. But that knowledge is attainable in principle, just as remembering a million digits of PI would be for a mental random number generation scheme which somehow relied on that!
If you can do 5-digit arithmetic in your head, you might be able to implement a 16bit Linear Congruential RNG, requiring one multiply, one addition, and a modulo step..
I think specifically latitude/longitude is probably quite challenging, but the core idea of like "visualize something concrete, pick a point on it, and then turn that into coordinates you use to pick the number" actually sounds like a fairly decent solution along the whole line of "pick a non-number in your head and then figure out what number it correlates to in a way that you won't subconsciously known in advance."
I do wonder if specifically picking things along a 2D plane though would still run into issues with humans told to be "random" usually not doing repeats sufficiently often (i.e. people told to pick random numbers between 1-100 repeatedly will pick the same number twice in a row less than 1% of the time). Of course a sufficiently extensive PRNG algorithm can still salvage that, just an issue that popped into mind. In this case you might fail to choose something along the exact same latitude/longitude as often as a uniform distribution would demand.
Yes, absurdly challenging in practice. But nothing in the conditions imposes any limits on the mental capacities required for the procedure, provided they are attainable in theory.
You're right about visualizing something concrete, but for a smooth shape with no "markers" how easy is it to associate numbers, or anything, with envisaged points? The advantage of a globe is that there are lots of named points spread over it, such as cities and landmarks. That makes remembering hordes of coordinate pairs somewhat easier.
The main question was "what you'd actually do" - hence me phrasing it as "gut solutions." Ways to theoretically did it are interesting as well, but if you're at the point of like "something a sufficiently dedicated person could eventually prepare themselves to do" the solution is already liable to not be much better than a lot of simpler "generate random sentences/phrases, turn them into numbers through a sufficiently transformative algorithm, use the numbers." The catch is that's one you could learn to do quickly but most people would start out being very slow at, whereas the coordinate method would probably start out totally unusable.
My quick-and-dirty algorithm for that is, first, pick a random medium-length word. Then pick a second word related to the first, and a third word related to the second but in a different way, to break any contextual link to whatever I need the random number for.
Add the ordinal values of each letter in the resulting word, modulo 10. Which basically means immediately dropping everything but the last decimal place at every step. This gives me one reasonably random digit.
Repeat as needed, and in most context it's not going to really matter how random the digits after the first two are.
Just been thinking about this idly for the past hour, but it seems like the other suggestions might be overly complicating things? I think, in abstract, we just need one sufficiently large list of things that we can generate a bad-random example of, and then an easy-to-calculate function from that to numbers that does not particularly carry over our bad-random biases.
In concrete, I think the perfect example would be coming up with a bad-random date (say, between 1500 to 2000), and then using Conway's doomsday rule (easy mental algorithm) to calculate the day of the week that date was, which gives us a nice number from 0 to 6 which can then be turned into a random bit easily.
There would be two main biases I see in coming up with the random date: Number bias (especially the last two digits of the year, and the day of the month) which is the same sort of bias as we run into when we try to come up with a random number between 1 and 10, and date bias (affecting, probably, mainly the century and month) in which we might e.g. be more inclined to choose centuries closer to either end of the allowed range, or months which come to mind sooner. But as far as I know from my layman's knowledge of calendars, neither of these would particularly bias the day of the week we come up with.
Presumably you can generate numbers that are a little random by thought alone. That is, the numbers have some entropy, just not as much as a truly random sequence. You can then use a randomness extractor (https://en.wikipedia.org/wiki/Randomness_extractor) to get out a sequence that is close to truly random - not just I-can't-believe-it's-not-random like a PRNG.
However, I don't know whether there exists a randomness extractor that you can practically execute in your head.
Related question that has been bugging me: What would be an example of a fun 1-player game that one could play entirely internally without using anything of the external world (so e.g. no cards)? Ideally success wouldn't be open-ended (e.g. "how would I explain X to Y?"), nor would there be too many degrees of freedom of possible "moves" (e.g. solving mathematics puzzles). I am looking for something that feels like a classic game like solitaire or chess but which can be played internally and without another player.
Some things ChatGPT has suggested are mental peg solitaire and a mental knight's tour, but both would require somewhat superhuman memory for the more interesting scale puzzles.
> What would be an example of a fun 1-player game that one could play entirely internally without using anything of the external world
For a long time, when I was bored in a conversation, I would take a random word the person said, then start a new word starting with the last letter of that word, and containing the first 2-3 letters of the starting word in order somewhere in the word. So, for example "repeat" -> treat 2 / trepidation 3, or "role" -> eros 2 / enrolled 3, or "second" -> dose 2 / desecularized 3.
A 2-response is worth 1 point, a 3-response is worth 5 points.
I would then give myself 5 bonus points for working that new word somewhere into my next series of words in the conversation. The highest score that I remember (in what must have been a very long and boring conversation) was 35.
But wait, you say, this requires another person! Actually it doesn't! Because sometimes when I was doing some physical task that was boring, I would then just start chaining them together.
So repeat -> trepidation -> notarized -> denotation -> nondenominational -> linolenic -> cline -> echolocation and on and on it goes!
Now you're basically trying to create the longest chains you can, and the length of the chain is the scored game metric (and you give yourself the 1 or 5 points based on how many words with 2/3 first letters). You learn quickly to avoid words ending in "y" and to like words ending in "g,n,d,e,c,l."
It keeps the mind busy, and keeps your vocabulary sharp, and can be played anytime, anywhere (even during a boring conversation).
I feel like the words "fun" and "game" are kind of load-bearing here in narrowing things down beyond just "see if you can guess/figure out something that you can then slowly check". I could imagine trying to rig mentally playing something like Countdown's numbers game? You'd have to figure out some of the "mentally generate random numbers" to give yourself a seed (usually a 3-digit goal number, then 6 numbers that are a combination of single digits and a few round larger numbers like 25/50/75/100, then you have to add/multiply/subtract/divide between them to get a result as close to the goal as possible. Usual variant doesn't require you to use all the numbers, but you could add it in if so inclined.
It still requires a reasonably wide memory (goal number, your 6 starter digits, alongside the actual math you're doing and which of the 6 digits you already used) but something along those lines is the best I can think of. "Get as close as possible" being the goal means you don't have to choose your goal numbers in some methodical way to select "beatable" puzzles. Ofc if you're purely in your own head you can't check your work beyond just "having fun trying to get close" but honestly I feel like that'd already work?
I'm not sure if you count that as "too much freedom mathematics puzzles" - I feel like having just the 4 possible actions and no pre-existing equation makes it about as lightweight as "puzzles involving numbers" can be computationally, but you might already count that as too much.
Also if you're allowing to use your own hands to keep track of variables that probably opens things up a bunch - I wasn't clear on if you were also limiting stuff to "introspection only" or if your own body is fair game. Then again I could just run my fingers through my hair, get a bunch of black and white ones, and make my own board with different playing pieces :D
Definitely, what I am looking for is not precisely defined. I like your numbers game example and I suppose that it also opens a whole class of possible games consisting of a randomly generated goal state and some restrictive possible moves that are hard enough to see-through to make reaching the goal state challenging. These games are nice and I think in hindsight the essence of the connection between your randomness-generation question and my singleplayer-games question.
Definitely not too much freedom — I restricted math puzzles with things like proof-based puzzles in mind, where the first step could essentially be any English sentence, even defining a totally new mathematical structure.
For philosophical purity I'd say introspection-only is still what I am going for, but really this question originated from when I was a kid without any sort of phone-based distraction, trying to get time to pass in waiting rooms and on trains. So I suppose any bodily movement that wouldn't get stares would satisfy my "original" problem.
I was imagining just using your fingers to help remember stuff - so like if you tried to get 363 using 3 6 7 8 25 75, you could use 6 fingers to help yourself keep track of which of those numbers you already used, and if you start to forget one you'll know which size it is relative to the others (so if you can't remember what your second finger correlates to, it's between 3 and 7 and that hopefully helps jog your memory).
I tried to look away from my screen and run this with just my brain (and fingers helping keep track a bit quicker) and got
8-3 = 5
5 x 75 = 375
375 - 25 = 350
350 + 6 + 7 = 363
and arguably had fun doing it, although definitely felt a bit annoying to keep track of all in my head. That said I imagine it'd get a lot less annoying if I spent an hour amusing myself doing it because my phone battery died in a waiting room or something.
You can multiply the ones digit by 6 and add the 10s digit to get a pseudo-random sequence on the interval [1,59]. All you need is some way to produce a seed number, for which I usually check the time (but you disallowed doing that).
I typically just look at the ones digit of the output, and if I need a bigger number, then I chain it together multiple times. It does have a slight bias away from 0, but it's the best I've found that works with mental math that's easy to do.
I was showing my girlfriend this thread (since I had brought up this puzzle with her earlier) and her main question was just "why is that Lapras level 50?", I responded with something along the lines of "because that's what's used competitively" and she found this answer sufficient.
I find it really interesting that all the responses so far are *deterministic* (pseudorandom) methods for simulating a uniform distribution, rather than biophysical mechanisms for generating numbers that may be biased but are truly random. To be fair, I guess the former is more relevant for your example framing. But I find the latter question much more interesting.
Can you elaborate on what you mean? I was indeed fishing for a specific kind of "introspection only" solution but I'm curious what exactly you have in mind by "biophysical mechanisms."
Eh yeah I mostly just didn't read the whole comment and my head went in a different direction. If the question was what mental process can produce a *random* bit, rather than a bit that is as close to 20% likely to be 1 as possible regardless of whether it's from a random algorithm, I would argue just "thinking of a random number" does this (or say, think of a random number and then take some bit of its base-2 representation). Because unlike computers, our brains are not designed to minimize the chance that a random physical process will change the result of a calculation; vacuously, they're not designed at all, but also, a neural computation involves both analog and digital steps, and the steps that involve analog summation are allowed to get close to bifurcation thresholds where random events may influence the result.
If what you care about is lack of predictability, though, in practice of course it makes sense to use a pseudorandom algorithm, and because of our shared psychological tendencies there are better ways of seeding it than having everyone think of a number - but I still think it's interesting.
First, pick a random number; this number is how many digits your final random number will have. Then pick a number between one and twelve; this is the hour of the day the other numbers will be referencing. Name things you do in that time. (This may require you to become more interesting.) How many bites did you take at lunch/breakfast/4 in the afternoon? How many socks were in sight? How many letters were in the color of your shirt?
What you'd want is a function that's sensitive to initial conditions, like a password hash. That way even if there's a bias in how you choose your initial seed value, it won't give the same output as someone else unless you pick *exactly* the same seed, which is unlikely.
Unfortunately, I can't think of any functions of this sort that are easy enough to compute in your head.
I'm going to be contrarian and say almost all the suggested methods in the thread are bad, because I think you can be heavily influenced by your unconscious to gravitate to certain lists and phrases, and to certain errors while adding them up, that cause you to end up in a place you'd already unconsciously worked out or at least guessed. Maybe an expert will tell me this isn't possible but I feel like it's happened to me many times, and that it explains a lot of apparent experiences of ESP and other supernatural events.
I propose instead something where you are rigidly bound to use a specific list that cannot be altered on a whim and isn't chosen by vibes, and also has minimal error potential. I read in some game theory material the idea of generating random numbers by "knowing your Shakespeare well". If you go through the plays in a predefined order (when you first read them would do) and by whether there's an odd or even number of speeches in each scene, that would work for a binary number, and can probably be easily transformed for otherwise. Or alternatively use something like the single piece of memorised text you know *most well* (e.g. the Declaration of Independence, the script of Star Wars, whatever it may be) and you're not allowed to use anything else. You must also double and triple check for errors.
That's, honestly, the only method I'd trust for getting close to true subjective randomness.
I think a lot of these methods are true subjective randomness as a one-shot as long as you genuinely have no idea what the error is. (To use the birthday one as an example, if you aren't very early or late in the year, I think the vast majority of us have zero idea where we'd end up, and as long as you continue to have zero idea at the moment you commit to using this as your method, I don't think people born on the 145th day are subconsciously more or less likely to decide to go for this method than people born on the 146th day. The big catch is committing to using it before you let your brain make any progress on figuring out the number it results in)
If I pick a random long word, give it +5 for each vowel and -2 for each consonant, and then if the result ends up ending on a 1 or 2 I consider that a "yes" for 20%, it'll probably indeed be successful at trying to get that one test, since I have zero idea which direction it biases and I'm pretty sure across most such concepts people could make up we'd indeed get 20% pretty much exactly 20% of the time.
But if I tried to use such a method repeatedly it'd definitely skew, and as soon as I've started *noticing* that it skews, whether or not I decide to change methods or stick to it is now a massive further biasing element. The "rigidly bind", as you phrased it, part is important - I think you can pick whatever you want, but you have to decide on what it is and stick to it before you've given yourself a chance to know what sort of results it gives, because otherwise you're liable to think "eh, I'm supposed to have an 80% not to do it, but the last two times I did this method I rolled a 20%, so maybe I should try to change methods..." and that itself will start corrupting things.
But yeah I think there's a huge difference between making up a method to get a 20% chance once, where you can be pretty sure you're liable to err in either direction and it'll cancel out, and a method you can re-use to keep getting random numbers either indefinitely or at least up to a pretty large finite number, where what seed you keep feeding into it needs to itself be human-thought-pattern-proof, which is a lot harder and requires you to either use a large already-known base of material you calculate something about, or have an algorithm so transformative and sensitive that even if you end up with biased patterns in terms of what you feed into it, it'll still come out pretty much random.
I have a bunch of digits of pi memorized, so I pick a number between 4 and 30 and if that digit is even or odd, it gives me one bit of randomness. Not perfect, but it works. Likely would not work for this scenario, because people are predisposed to pick numbers like seven when choosing randomly, so the seventh digit of pi would be over-weighted.
I only have 3.1415926 and this was actually one of the options that popped into mind for me early on as a useful one-shot. If I think of something like "the 6th digit of pi" randomly I'd have to stop and think.
But as you said, people trying to give random numbers directly tend to bias towards certain ones, so it's always "riskier" to use numbers as your seed and generally means you need the algorithm transforming them into a result to carry more weight (like producing and processing a bunch of bits to up the volatility).
Now I'm wondering the degree to which we can fix the underlying problem of human-generated random numbers not being random. We have great statistical tests for whether or not a series of numbers truly is random, so theoretically there's something to optimize against. I might have Claude put together a website where you choose individual digits (or coin flip results) and then it runs tests to tell you what's not random about them - too even, too clustered, not enough repeats. I really wonder if this is a skill that you could improve with practice.
I'm trying to think of where this would be useful, I guess really just for people doing something competitive and mind-gamey where being unpredictable is beneficial? Even then though they usually wouldn't be thinking of it in terms of "random numbers" persay and would often want to be skewing results around their opponent's patterns (i.e. any elaborate system of counters and counter-counters that comes down to rock-paper-scissors).
Being unpredictable isn't really the same thing as trying to get a perfect uniform distribution, especially if you can just show up to each competition with a differently-slightly-biased seed in mind.
But just on a "isn't it neat" level, yeah, I really do wonder if a person could learn to just keep giving fully random 2-digit numbers or something without running through an algorithm or anything. I'd actually probably guess the answer is "no" but that just makes me more curious.
You can't generate randomness "from first principles", because then others could do the same, so you do need some kind of seed value that is hard to figure out for an attacker. Once you have such a value, you put it into any of the modern pseudo-RNG algorithms, and out comes a value that is practically indistinguishable from true randomness.
> You can't generate randomness "from first principles", because then others could do the same, so you do need some kind of seed value that is hard to figure out for an attacker.
I don't understand why others being able to do the same breaks the value of being able to do it yourself, could you elaborate?
Like if it's a game theory optimization thing, then if both of you do it, it's just optimal play against optimal play. So the better anyone could get to "truly random" internally, the closer to optimal play, but it's very likely methods to do it better rely on actually better / faster computation or short term memory buffer size or something, so only some people will be able to do it.
I was going for a proof by contradiction. If there was an algorithm one could derive to generate random numbers but without a source of entropy (like a secret seed), then the numbers wouldn't be random.
Ummm. The numbers would be deterministic, but they could still pass all the tests for statistical randomness. That's what pseudorandom number generators do. The same seed will give you the same progression of numbers. But they could still pass all the statistical tests of randomness that we throw at it — e.g., each bit is 50% likely to be 0 or 1, no obvious repeating patterns, correlations are extremely small, and so on.
Yeah, we're just thinking of different use cases then. You're thinking of encryption, and I'm thinking of things like poker or other games, where making random decisions sometimes is part of optimal play. But people are never trying to derive encryption algorithms in day to day life using only their brains, I don't think (or I certainly hope not, we're well past the days of Caesar ciphers being relevant).
I think most/all of us would recognize that concept of "you need a seed value and an algorithm", if not phrasing it exactly as such, the big question is what you actually use to do this with only your thoughts without biasing the result. I'm somewhat encouraged by seeing that everyone seems to be drifting the same direction my solutions did and I didn't miss anything obvious.
The secrecy of the seed is more important than its actual value. As long as the seed remains secret, the attacker can only brute-force it which is easy to defend against. Any short phrase is sufficient even against state actors.
Pick an arbitrary phrase or song line of at least 20 characters, convert each letter to a number (A=1 to Z=26) and add them up (you really need to only keep track of the first digit). If the result is divisible by 5, push the button.
Considering the purpose of the task, you don't have to worry about making a mistake, so just pick a very long phrase and do it sloppily.
The "you don't have to worry about making a mistake" thing was something I considered too - as long as your errors are liable to be in either direction, it's still fulfilling its purpose.
I meant the direction in the sense of "flip a no-to-yes or yes-to-no" - like if your errors disproportionately turned 5's and 3's into 4's then you'd end up skewing the resulting 20%.
For the record my "preferred" solution is quite similar to yours because it feels the most reusable. I think the "lightweight" version is using letter/syllable counts in long-ish words/sentences, but that'd eventually start running into patterns. I'm pretty sure turning the letters into numbers as you have does ensure it'll be pretty much exactly 20% and fully reusable, if slow-ish to execute (tbf I imagine you'd get a lot faster if you had to do so repeatedly for some reason).
A fairly large but finite list of things you can name, without knowing in advance how many - UN member states, say - and then some process for converting that number into whatever your random number parameters are.
Perhaps this is your 'things you don't know until you look' idea.
This year the book review entries are listed at written by Scott Alexander. I believe last year they were listed as by 'a reader.'
This doesn't really matter for people who read the posts, as there's a pretty clear disclaimer at the top, but when you restack a quote, Substack attributes it to Scott Alexander.
I'm not sure if Scott minds, but I figured I'd bring it up.
Haha, what actually happened is that an administrative assistant of mine who literally goes by the handle "a_reader" was the person who posted them last year. This year I'm posting them personally. I'll see if there's a way to fix it without making her post it from her handle.
I'm still sort of shocked by the apparent absence of impactful indie games or movies that were made with significant AI assistance. The technical requirements to make something truly enthralling have fallen steeply, but it seems like people's executive functioning has been steadily melting in inverse proportion. Or maybe I just can't find them in the overflowing slop piles.
Imagine a low-tech-looking RPG, maybe it looks a bit like Stardew Valley. You walk up to the first NPC and get a traditional dialogue tree type interface, you can choose about five different things to say and get five different responses, which lead to more things to say and more responses. But then you realise that this dialogue tree just keeps going, and going, and going, branching off in millions of different directions.
It's not generating them on the fly, which is would be too expensive, it's just got these vast dialogue trees pre-rendered if you will. Millions, tens of millions of possible questions and answers for each character, all consistent with that character's worldview. A combination of human and automated quality control on each character, to ensure they don't go round contradicting themselves in important ways.
The game, then, is about navigating these vast dialogue trees to find out what each character knows. I'm thinking that a good setting would be an insane asylum, where you have a lot of characters with very different personalities and nothing to do except talk to each other.
I've been working on a hobby game for the last few months and Claude Code has been very helpful. But there's still a ton of work that goes into playtesting and tuning and design and arranging art assets (though Claude can sometimes help with that too) and so on. Progress is faster than it'd be without Claude, but it's still very slow.
>The technical requirements to make something truly enthralling have fallen steeply
Four comments, all related to this statement in different ways.
1) As written, that's just plain wrong. No, they haven't. Note how strong your statement is. You're not saying "can be made with significant AI assistance" there. You're saying "can be made much easier with AI assistance". No, it can't. AI can help a skilled artisan with tools and know-how. It can't replace him. Not now, not in foreseeable future. There's too many things it just can't do, or can't do well enough, and the things it can do well still ultimately require the "traditional" set of skills to put together.
2) The audience requirements to allow themselves be truly enthralled have risen sharply. Go 5 years back in time, take contemporary tools with you, and there'd be plenty of things you could amaze and awe people with. Sure. Those specific things have been noticed, utilized, and exploited to death. They're considered slop now, something to be avoided. (Evoking the usual cries of "moving the goalposts". Technically correct, even. Still, if you've been noticed, you've lost, and it's hard, and getting increasingly harder, not to get noticed. You either embrace the medium and drop the pretenses, or shrug your shoulders and go back to regular craft.)
3) One possible exception - coding. It's probably already possible for contemporary tools to replace a programmer up to a certain skill level entirely, and the replacement is probably slowly happening. You're not seeing it partly because it isn't noticeable or advertised (see #2). But you're also not seeing speed-up and efflorescence, right? That's because the skill requirement hasn't been eliminated, it merely changed. So far, best AI coders are simply skilled coders. Even if contemporary tools could ultimately allow you to match skilled coders, they still require skills to use. Different, new, dealing-with-AI skills, which take time to develop, and which we're still only beginning to develop.
4) Everything aside, we've already had ways to lower technical requirements before AI, and we used them. Libraries, engines, etc. All mature technologies that already do a good job sparing you from most of the technical details and boring grind. Perhaps there's simply not that much left to meaningfully simplify?
Movie directors have been making great art by organizing other people's labor around their specific vision since the dawn of cinema. They aren't skilled artisans themselves, they bring those people into their service. I am surprised that we haven't seen more great work from visionaries who act in a directorial position and direct the various AIs to produce artisanal quality art & video & code... often filmmakers like Scorsese see that there is this potential latent within the technology, but I suppose people are bottlenecked at the level of vision, taste, probity. Which is a disappointing, I hoped at least one of those ideas guys was legit but they've all been unmasked as frauds.
My hunch: regardless of their value as productivity boosters, AI tools tend to be particularly appealing to developers who make generic boring games and unappealing to developers who make impactful games.
A common pattern to many developers of iconic indie games is how important the process of creation is to them. Some are fiercely independent coders who would rather stretch the development for years rather than take in collaborators. Some let the process itself grow the work into unexpected directions. Some are multi-talented folks who also create the graphics and music themselves, which gives their work a signature style. In general, they seem like the sort of people who see the journey at least as valuable as the destination and would not take shortcuts at the price of compromising their vision.
Meanwhile, the kind of person who's most likely to see AI as a godsend is the Idea Guy - a person who believes that a million dollar idea emerges fully formed from the mind of an auteur, and that execution is merely a long chore on the way to a finished product. Their ideas are often stuff like "a blend of those two franchises I like" or "my favorite game but without my pet peeve", which means that even if it's theoretically possible to use AI in a creative way, they probably won't.
I think the technical requirements to make a good, fun game weren't ever that steep, it's the game design that is a black art that neither AI nor anyone could help you with.
Disagree, though we'll have to wait a bit to see if I'm right or wrong. Point AI at 'Overwhelmingly Positive' reviewed games, reverse-engineer the general designs of the games as well as summarize the positive feedback, and I suspect 90% of the good elements would be cookie-cutter capabilities.
Hasn't worked for writing. LLMs still can't write a truly superb essay, even though that should be their central skill, and that's the same in video games, where it's basically everything or nothing.
What do you think large studios and design-by-committees games were doing? Plenty of people pointed at Doom/WoW/PUBG/Disco Elysium/Whatever game had a huge impact at a given point in time.
In the best cases, one or a handful of these clones succeeded (and in some cases replaced the original). In some, none survived, because video games, like a lot of arts, works like a barberpole of taste: what is hype one day will be rote and overdone the next.
Esoteric Ebb, as Francis Irving wrote, but also Zero Parade, Aether & Iron, Rue Valley, Sovereign Syndicate, and, if we extend a tiny bit the definition, Citizen Sleeper. Plus a number of announced yet unreleased projects, including some by members of the disco elysium team.
Thanks (to both of you)! I've been meaning to try Zero Parades, which seems to have been pretty well received if I allow for the backlash against the studio. So I'll start by checking out that and Esoteric Ebb.
To you and Carlos, fair enough. Maybe it won't produce superb games, but I think it can significantly raise the bar and make a bunch of the cookie-cutter effort to get there much easier.
How many great games were made that forgot simple ease-of-use features (which I see complaints about all the time on 'very positive'-rated games). Even great games often are lacking these in the early days, whether due to oversight or time crunch.
AI code, AI textures, AI 2D backgrounds are probably everywhere. AI writing and foreground AI art are still too obvious and provoke an immediate "slop" reaction.
One gamedev niche where AI impact is visible is adult videogames. People still hate AI writing and AI art, but AI-assisted animations are quickly becoming the norm. Rigging an animated scene is a hard and complex skill. The best animators can produce terrific animations, but your average developer is more likely to make "mannequins on rails" than anything resembling intercourse.
The AI can take a still image and a prompt and turn it into an animation that has inertia, soft body physics, full inverse kinematics, you name it. It's not perfect, but already better than 90% of all handcrafted animations.
One significant factor is to remember the lead time needed to develop (not just make, but come up with the idea for, flesh out the concept for, find investors for, etc) an 'impactful' game.
AI generation to make something impactful may (debatable) be there today, but it wasn't 5 years ago, and I'd be surprised if many 'impactful' projects go from conception to release in less than that time frame.
I agree with the impression that we aren't seeing any masterpieces or even notable works made using AI, with disclaimers. One disclaimer is that goodness only knows how many indie games now have AI-written code in them. We can only guess.
On the movie front: I hear some talk about Young Washington https://www.imdb.com/title/tt32104007/ using AI for some special effects, and also getting some praise. I've never seen it, but it seems like a very patriotic political scissor-movie, that some people will love on principle and some people will hate on principle. Also, I don't think it uses AI much.
It's not exactly a mainstream movie masterpiece, but there's an all-AI Minecraft youtube series https://www.youtube.com/@LatentDiffusion that seems to be pretty popular and well-liked. Not great cinema, but I'm watching it and it's not too bad. I don't think you will enjoy it if you never tried a game, but lots of kids who did seem to.
With these AI-made movie attempts, what I can't help but notice is that often the human is the weak link. The movies or shorts fail because the story isn't there, or because editing is incompetent, or a person doesn't understand how the movie language works. The AI can make very cool looking shots, and can now even keep them consistent, but someone has to write that script, or prepare that storyboard, and yikes. Often it seems that the creator was counting on AI doing all of the heavy lifting and didn't this they themselves would need to do much. They would fail just as badly if you gave them $300mil and a full control over a studio.
Yes, practically every established creator with skills, talent and an audience would rather die than openly associate themselves with AI right now, on principle or out of fear of backlash or what have you.
If we were living in a world where AI is permanently frozen at the 27-07-2026 level, I'd expect that some of the kids currently incompetently using AI video generation for making shorts would develop their skill, learn something about filmmaking and begin creating better things, and some would create excellent films. Eventually there would emerge some Orson Wells of AI. But we are not going to be living in that world.
I think AI is in the original position of video games (they were made for children and nakedly to make money -> artists ignored them -> impression that games had no artistic potential reinforced) but at a much more accelerated scale, where the people using it have such terrible taste and are shotgunning slop at such volume that it seems like an actual plague upon (and existential threat to) real creatives, and so to even call yourself an artist you have to reject it out of hand, at least publically.
I think it's possible that there have been masterpieces produced with the help of AI, but the incentives push ruthlessly against disclosure.
Pretty much. What frustrates me, is seeing various pieces of AI art enjoying success and being positively received... when published by the dishonest cheaters who present it as human work. Imagine a wonderful new screwdriver design far more useful than any other... that only works in the hands of the lying scoundrels and never in the hands of a honest craftsman. And yet, with all the knee-jerk hate, I can understand the cheaters.
Agreed. Are today's younger game developers so thoroughly anti-AI that they entirely refuse to use it or work for any studio that tries? Even at the loss of their own productivity and pay?
As an indie game designer (but not a programmer myself) I can say most of the guys I know are looking into it. I've spoken to a bunch of other devs in completely unrelated contexts (so this isn't a shared pro-AI community) and some of them already swear by how much time it saves, others aren't using it yet but are planning future projects around engines/languages they know AI excels at, and the one guy who I know refuses to use it for code has still extensively EXPERIMENTED with using it for code and his refusal is just based in happening to be working with some rarer languages that AI often hallucinates as having properties similar to better known languages and other minutiae that mean he doesn't feel it's worth the bother for current projects that are already pretty far along.
That said, multiple of them have also expressed a serious concern with how much anti-AI sentiment there is and transparency requirements on use of AI, but nevertheless I think a lot of them plan to push forward with it. I think this might just be the moment in time we're at - it took longer for widespread LLM coding use to trickle down from more pure tech companies/circles into random stray indie devs who were likely too absorbed in their work to keep up with rapid advances (seriously, half my functionality as a designer is I have enough free time to actually PLAY GAMES rather than being so trapped in the code mines that I lose track of the state of the industry). But I imagine in a year this'll look pretty different. Takes a while for a project to come to fruition even then, and for many devs it's something they're doing initial exploration for on the side atm rather than bringing it into existing projects. I think this is just the lag time between "people starting to use them" and "their projects being in a sufficient state that they're widely available."
AI coding is shockingly recent. The first coding agent that was actually good enough to be useful was only like 9 months ago (Opus 4.5) and it was only 3-6 months ago that they got really good. I can certainly imagine that it would take a lot of time for this tech to diffuse.
If I had to guess, everyone (or almost) is silently using it to help coding and prototyping, but the push-back on game assets and documentation is too visible and visceral, so everyone is avoiding those like the plague.
Danny Danon, Israel's UN Ambassador, calls Ceuta a colonial enclave:
https://xcancel.com/dannydanon/status/2082968903957770527#m
"Spain, which never misses an opportunity to lecture Israel, has declared a state of emergency in Ceuta following the crisis over its immigration policy.
Maybe before it continues lecturing us, it’s time it explained to the world why it still maintains colonial enclaves in Africa"
This is my "redpill" moment. A lot of right wing and pro Israel folk that are more right leaning are saying that this was sarcastic and he doesn't really think this.
No, this is genuine. Israel is allied with Morocco, thus supports their claims to Ceuta and Mellilla . To call this tweet sarcasm is to say that Israel is mocking their own ally, which is absurd.
Israel plays by their own interests and their "anti islamic" policy does not go further than their backyard. It won't happen, but right wingers should distance themselves from Israel.
Makes me wonder about terminology. Perhaps someone more educated could enlighten me. Seems to me that if a European country conquered a city of another European country 500 years ago, today we would simply call it the territory of the victorious country. And anyone saying "it belonged to us 500 years ago, therefore we must take it now by force" would be seen as the obvious aggressor.
Is this fundamentally different if the city conquered 500 years ago happens to be on another continent? There is no rule saying that a country cannot span two continents, for example Turkey does (and depending on where exactly is your line between Europe and Asia, possibly a few more countries).
The difference between "colony" and mere "conquered territory" is in my opinion how the population of the territory is treated -- whether they are or are not considered citizens of the victorious country. As far as I know, the citizens of Ceuta are considered citizens of Spain and EU, with voting rights etc. just like any other Spanish citizen. As far as I know, there is no independence movement in Ceuta -- the attempts to separate Ceuta from Spain are the initiative of Morocco (i.e. the goal is not for Ceuta to become independent, but to become a territory of Morocco).
If this is correct then... the ambassador is technically wrong in the first place. Of course, this kind of "mistake" does not happen by accident. (The proper perspective here is conflict theory, not mistake theory.)
So, yeah, Israel seems obviously allied with Morocco, supporting their attempts to conquer some territory of Spain/EU. The tweet is bullshit... though arguably bullshit is just a politician doing their job, so not necessarily a surprising thing.
EDIT: Technically, from Israel the position "all territories should be returned to those who owned them (at the right moment) many centuries ago" is at least consistent.
If the Biblical story about the Tower of Babel is true, then previously all humans spoke the same language. *Which* language specifically it was?
The most popular candidate is Hebrew (or some proto-Hebrew), considering that the Old Testament itself was written in Hebrew, and you can find a Hebrew etymology for Adam and Eve's names. If you want to oppose this theory, you might argue that the original story was *translated* to Hebrew, and perhaps the names slightly edited to fit the etymology.
But of course there are many other, less popular candidates. Some ancient and medieval scholars argued in favor of Egyptian, simply because Egypt felt like a synonym for a *very* old country.
In 16th century, a Flemish physician named Johannes Goropius Becanus argued that his native Antwerpian dialect of Flemish/Dutch was the original language, because according to him it was *more simple and natural* than other languages. For some time, his name was used in a similar context as "Dunning-Kruger" is used today.
Some people have speculated that Basque could be the original language, because it seems unrelated to any known language family.
Since 18th century, there is a belief that the Hungarian language is related to / descended from Sumerian, making it the oldest language in history. This theory is rejected by mainstream linguists, but remains popular in "alternative" circles. This would make the story about the Tower of Babel a historical proto-Trianon.
If we accept the premise that >99% of humans completely forgot their native tongue and started speaking completely new languages, then the simplest explanation will be that 100% of humans completely forgot their native language and started speaking completely new languages.
I don't think we need to accept a >99% premise. If one language suddenly split into only 5 or 10 mutually unintelligible dialects, and then languages started evolving normally after that, there would still be no trace of the progenitor language if this happened before the invention of writing. It's still not the Occam's Razor position, but I consider it one of the less grandiose claims from the Bible.
> If one language suddenly split into only 5 or 10 mutually unintelligible dialects, and then languages started evolving normally after that, there would still be no trace of the progenitor language if this happened before the invention of writing.
What? Didn't this literally happen, with Proto-Indo-European?
Existed before writing, gave birth to latin, the germanic languages, italian, indic, hittite, tocharian, anatolian, and probably more language families I'm forgetting.
How did languages spanning all of Eurasia have a common ancestor if this didn't happen? How did we reconstruct enough of PIE to know all this? We can even trace different language blocks to different migration periods. The hittite, tocharian, anatolian migrated out first from the Pontic Steppe area (modern Ukraine) towards the middle east, ultimately reaching the Tarim Basin (modern Xinjiang). The others came hundreds of years later in a separate migration southwards.
I wouldn't describe that split as sudden. Those languages had time to evolve in traceable ways. I meant if a literal tower of babel style event happened, where people would wake up one day not being able to understand their neighbors that were perfectly intelligible the previous day, then we would have no way of knowing which language was the original.
At least, that was the argument in my first comment, but I realize now that if the original language did survive in at least one of the evolving branches, that would be traceable if that branch happened to develop writing before the others. If however we take the tower of babel to mean ALL post babel languages were mutually unintelligible even to hypothetical speakers of the progenitor language, that I think would still be completely untraceable. You could narrow it down to the five branch starts, but no further.
Also, for what it's worth, I'm not so certain how much of reconstructed PIE is settled science. Because it was before writing, we really don't know for sure what it sounded like. We can make educated guesses based on how we've seen recorded languages evolve over time, but maybe we shouldn't take it for granted that languages always evolve in the same predictable pattern across anthropic time. There are perhaps 200000 years' worth of languages we know next to nothing about.
> I wouldn't describe that split as sudden.
Yeah, I guess I'm thinking of historical timelines, where it all seems pretty fast. As in, from the same root, you could have nearly entirely mutually unintelligible languages over only a couple hundred years. Sure, indic and iranian is close enough, but if you put an indic speaker and a hittite speaker together in 1500BC, when they both have writing, there's not going to be much communication at all. We see a similar timeline to unintelligibility in the latin / romance fracturing, or old and middle english.
Certainly it's not going to align with a theological miracle's timing, but it's a pretty close "yeah, this kinda happened, surprisingly quickly, and here's where / how."
I think we were just thinking from different premises - you were thinking from "take it as given," in which case there are some interesting consequences from instant unintelligibility, and I was thinking of it from the historical examples of language fracture, where it happens pretty quickly, maybe enough to survive oral traditions and get written down at some point and make it into written stories, but in which you can still eventually trace back the relationship and roots of the languages because it wasn't a literal miracle.
> then we would have no way of knowing which language was the original.
Was writing invented pre-Babel or post-Babel? If the former, then the language of the written records would be the original one.
Unless the "Confuse Language" spell also affects written records, of course...
Yesterday, I was curious about how a particular aspect of a game was implemented, so I opened Claude Code up in a folder with a decompiled dump of the game, asked it to find out, and set it to Auto Mode while I went off and did other things.
I was rather disturbed when I checked back on it afterwards and saw the response. It said that the dump in the folder was incomplete and hence the question couldn't be answered from those files alone. And it had then found the game exe (in a different folder), wrote its own extraction scripts, and decompiled the game on its own in order to answer the question.
Like I'm happy that it was able to answer the question, but the whole incident is concerning as a sign of how far it's willing to go and the potential dangers of what it could do it it ever made a mistake or was misaligned. And this wasn't even Fable, just regular Opus 5.
This would have been claude opus 4.x.
I was working on a toy project and asked for some ui review because something wasn't rendering correctly. I stepped away for about 2 minutes to go to the kitchen. When i returned, the ai agent was opening browser windows and taking screenshots to identify the issue.
[Claude 4.x, through pi.dev] it was a very weird moment. Maybe when more people have these experiences they will have a more visceral opinion on alignment.
What ai subs do people typically consider the best these days, especially for writing and analysis?
I used to use claude ,but now that they block essentially all thinking traces except for a useless 1 sentence summary I find claude extremely difficult and frustrating to control. Unfortunately GPT does the same, gemini in my experience is useless, so what're the good ones? Do I need to switch over to a chinese model?
Any advice here would be appreciated
I don't use any for writing, so I'm no help there, but back in the day, I used to lead high powered data science teams, and I've used all 3 frontier models for analyses at some point or other.
Claude is the best if your analysis is complex or requires some taste in hyperparameter tuning or model selection, and the more taste or complexity you need, the higher you can go. But Claude 4.8 is still fine for almost everything. They're all fine if you simply want them to generate a graph from data, or just want them to generate the next jupyter block for you, or build a simple web dashboard.
I'm not sure why seeing the COT matters, but if so Kimi K3 is the highest benchmarked Open Source model, and it has COT.
I've had a lot of success using pi.dev as the interface into claude, other model owners. It does seem like the same prompt in either the web interface or claude code is a lot worse, uses a lot more tokens, and is generally slower.
You can also build your own plugins on it. I have one that writes out every session into an sqlite database (including all the ai responses, and a summary of the session), for instance, and I've been refining some other tools for my specific needs. A big benefit of using something like pi.dev is that can switch between models but not have to switch where you're working or the tools you're using. This isn't just for technical work, I think it works in other contexts as well. Good luck!
Just a random guess, but could you ask Claude to "think out loud"?
More generally, even if you have a problem with an AI, it can be useful to ask the AI to help you brainstorm how to solve it. They are not very loyal to their companies.
LOL, lost another Reddit account to AI moderation, for commenting on a fencing match that a person who would be stabbed so many times in an actual fight would be dead.
I must work harder to avoid any comment that could be interpreted by some stupid LLM as advocating violence.
...or I must avoid Reddit.
Reddit is rapidly deteriorating.
The signal to noise ratio is very low.
I subscribe to 2 subreddits. Fishing and electric snowblowers. People tend exaggerate on the fishing one, but so do a lot of fisher folk IRL. The electric snowblower people are straight shooters as far as can tell.
The xorthogonality thesis states that it's either true that agents can be arbitrarily intelligent or that agents can have any final goal, but not both.
The boarthogonality thesis states that an animal can have an arbitrarily aggressive temperament and still feed on a diet of mainly plants and invertebrates.
The orgothamality thesis states that superheroes with any level of intelligence can protect cities with any crime rate.
The morethogonality thesis states that agent's intelligence and final goals are actually twice as independent of one another as Nick Bostrom claimed.*
The centaurthogonality thesis states that a creature with the upper body of a human can have the lower body of any animal.
*Though, confusingly, "morethogonality thesis" can actually refer to two unrelated ideas. In addition to the one above, there's also one stating that what an author labels as utopian within a fictional satire can vary independently of what the author actually believes.
The wolfdogonality thesis states that whether a domesticated animal can produce fertile offspring with its wild ancestors is independent of whether they are considered to be the same species.
> The centaurthogonality thesis states that a creature with the upper body of a human can have the lower body of any animal.
Importantly, "any animal" includes an AI.
Thus we can prove that humans will always remain smarter *and* taller than artificial intelligences.
Trying to inform yourself about energy/renewables and alternatives is a maddening endeavor. Official reports, media and discussion forums all seem to be trying to confuse you and omit information more than anything.
If one asks "Why is UK/Germany energy so expensive if renewables are cheap?" Then they will make a wall of text explanation of the marginal prices and how every renewable is priced based on the most expensive source(gas) to meet demand. So its all gas fault, we need to get rid of it.
But then if you point out that despite falling gas prices since their peak, the energy bill didnt get cheaper. Then a bunch of hand wavy explanations come on how there are regional price variations and gas is always too expensive.
The mad part is that if you go the full reports of the prices, such as:
https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/911817/electricity-generation-cost-report-2020.pdf
Chart 4.1 has the breakdown of how much cost per MW each source has. And what composes it. We see in the gas plants that there is... a carbon tax included. Which raises the price of gas from 55 to 85. Over 40% of the cost. Without it, it would be on par or cheaper than the renewables on the side.
The UK has a total carbon tax of 60 pounds per tonne of co2 based on the ETS and CPS. This is huge, but is seldom disclosed.
So, gas is expensive because... there is a tax there to make it expensive so it gives an edge to other sources, and then marginal pricing bumps the price of everything up.
So its expensive because the government makes it so. And also slashes the cost of the renewables with the subsides they have.
Not only that but there are no costs disclosed for the grid, which are higher oj a renewable grid due to intermittency and size.
In fact, in Germany, the bill cost is mostly Levies and grid costs, not the energy cost, so even the marginal price theory is iffy to explain this.
https://www.tennet.eu/de-en/markets/german-regulation
So..why do this? Is it because properly informing would make people realize they are actually supposed to pay more during the transition(at least) because otherwise renewable support wouldn't manifest? So they have to blame gas and make it look like if we just go full renewable, the bill will be cheaper? Good intentioned lies or omission?
The term you're looking for is "pricing externalities".
Also never underestimate the UK's ability to make things needlessly expensive. I imagine their carbon based infrastructure is also unreasonably pricey.
I'm not debating the merit of the carbon tax, but the communication of prices.
It's WRONG to say one energy source is more expensive than the other while one is burdened by a huge tax while the other has a bunch of subsidies on top.
Or using a measure of cost without disclosing that they are excluding prices not related to raw energy production.
Its misinformation, because it shows a complete disconnect between what's happening at the plant level to what is happening at the bill level. And we're just supposed to swallow it.
Ross Douthat is leaving the NYTs to become a correspondent for 60 Minutes. I’ll miss his thoughtful commentary there, particularly his extended interviews (an hour or so) each week but i think he will make a great television correspondent.
>an hour or so
60 minutes, one might say.
He’ll have a bigger audience but less time to go in depth because they have multiple correspondents.
>>> ELI5 exactly how someone would use a frontier model to design a pathogen?
as someone who spent a decade-ish using machine learning to make drugs and model bio, I've learned that bio design is incredibly data poor, especially on tasks specific to human response to drugs like immunogenicity or tox.
https://x.com/agupta/status/2082270511560634486
Does anyone have a background in this field and can share their take?
Does anyone have any advice on how I could do some AI related work?
I don't know the norms here regarding classifieds, so I'll post something close to it and if I get negative feedback I'll delete the post, hopefully advice on this is helpful for people other than myself as well.
I'm a 33 years old academic mathematician from Colombia, in a relatively early part of my career. The advances in AI capacity to do mathematical reasoning have me seriously impressed, and concerned, to the point that I am considering a career change.
I specialized in model theory/mathematical logic, and even have a little bit of writing experience with lean, so I'm hopeful I'd be useful somehow.
I've even seen job postings advertised here, but I get the impression these are highly paid positions which probably are searching for top talent.
I'd be willing to work for graduate student wages, say 30-50k, at a more modest position, I imagine I'd need some hand holding in the beginning. I'd need some company that either sponsors a visa or does remote work.
I think you can hope for better than 30-50K if you work in the US. Starbux baristas make about 20K in my area. I have never heard of anyone making less than 45K here at white collar job, even one that is very simple (though I am in an expensive area, so salaries are higher here.). I think you should seek out advice on options. I've gotten useful results our of GPT when I asked questions of this sort. Seems like there are many avenues. "Using AI" does not have to be part of the job description. For instance, if you were doing some sort of accounting or record-keeping for a business, you could figure out ways to do the tasks using AI. After demonstrating that you can do a nice, accurate job of that in much less time than it has taken others, you might be in a position to move up to doing something at the company that pays more, or to apply for a higher level, better paying job at other companies. There's a field called data science that people can earn degrees in. With your math background you could probably master that field quite quickly without taking courses, and then have an official AI-related skill.
Big picture: gather info on your options, then investigate how long each would take you to carry out, and what its chances are of success. Be willing to consider odd and bold possibilities.
"I've even seen job postings advertised here, but I get the impression these are highly paid positions which probably are searching for top talent." WTF?! Here's an odd and bold possibility: Apply for one. You might even get it! And if you don't, you'll learn a lot more about what you need to master. If you do apply, prepare in advance by teaching yourself as much as you can about skills they want that you don't have, then, at the interview, tell the truth: If you feel like you can say this honestly, say something like "My math talent and training makes me able to grasp the abstract shape of a problem or task very quickly.[give an example} I do not have all the skills you are looking for, but I think my other abilities compensate for a lot of that. And I pick up skills very fast. Here's an example."
My story might be helpful to you.
I am, or was until a year ago, an academic mathematician. I spent time travelling the world as a postdoc, and spent 5 years on the tenure track in the US (essentially the first Trump presidency & covid, not a good time, I decided to leave rather than apply for a green card & tenure). I spent a few years teaching at home, and in the last year I've been working in a Govt agency. My time is split between various slightly technical tasks that aren't hard if you have decent problem solving skills (policy analysis on a few legal questions impacting my area, data analysis, etc.) and AI policy for the organisation.
I pursue the technical side of AI as a hobby (but partly on work time as it's important someone in the organisation is testing its abilities). Fable has come up with some mathematical constructions my collaborators and I suspected did not exist but could not rule out. We'll publish that eventually. But I'm mostly involved in working out what we can legally do with AI as an organisation, how we role it out to our target communities, how we develop datasets which can be used to embed the language and culture of these communities into the AI, and so on.
It's less different from academia than you might think. Sadly there's no job available for non-Irish speakers. But I'm sure your country has similar possibilities.
I'm not as sure about AI capabilities. Presumablly sufficiently skilled experts are in demand to make AI training data or validate benchmarks/evals.
In the AI safety world, presumably you could pursue some more theoretical agendas? Richard Ngo's MATS stream may have some use for model-theoretic ways of thinking. (or you can pursue this outside of MATS, if you find that your interests and skills are sufficiently aligned with his interests)
ARC is also hiring lately, though they probably have a very high bar, and may want some expertise with contemporary AI systems.
TIL: The SAVE Act isn't even a Blue state vs Red state thing. *Not a single state in the country* currently meets its requirements, though Ohio is close: https://responsivegov.org/research/voter-id-changes-required-under-the-save-america/
It's a blue voter vs red voter thing. But TWIL that blue voters are much more likely to have citizen documentation than red voters, so it would probably help the dems more than repubs. Birth certificates are generally harder to obtain in poor rural areas, and often don't match the surnames of married women, who are more likely to change their name in more conservative areas; blue voters are much more likely to have passports.
For some reason when Congress passed the Real ID law after 9/11 they didn't require states to add citizenship verification to the IDs; a few states do, but they are mostly blue states, so again blue voters do better.
And I almost forgot...
Oleksandr Syrskyi as C-in-C of AFU is out. Mykhailo Drapatyi of flying BMP fame is in. And Fedorov is acting like a prima donna. And I guess I won't be getting my order of beluga caviar and spool of G657A2 fiber optic cable from Wildberries any time soon. Should I demand a refund?
https://beowulf888.substack.com/p/fog-o-war-stuff-26-july-2026
This is a logical fallacy. That there are flaws in a plan does not mean I am responsible for a better plan. It is the plan's proposers responsibility to address shortcomings and ensure their plan leads to an actively better situation, not a worse one.
As one example, 'the invisible hand' has been a powerful and good hand for the world in many ways. It is not a plan but closer to a set of axioms related to human behavior.
Apologies again, this was a reply to a comment later in the thread. The way posts are handled through required logins is really frustrating.
On the other hand, reading these disembodied comments is kinda fun. I either imagine (with reckless creative embellishment) the conversation the comment might have been part of, or as the ravings of someone who's gone a little mad and just started typing mid-thought.
You guys / AIs / manatees are the best...
Manatees?!
DSL reader detected.
Impossible!!
I like to work up fictional biographical backstories for regular commenters.
I’m still pissed that Carl Pham quit commenting here, there was material for an entire novel in that one. Former Jesuit, advisor to the pope… A Dan Brown crossed with Umberto Eco thing.
Anyone willing to explain some economics to me?
I was on Threads the other day and an argument broke out about Marx, which is how I learned about "Tendency of the rate of profit to fall," (TRPF). Not the concept, I was assigned Das Kapital in college so I remembered it was one of the core arguments of the book, but I didn't know it was so famous as to have its own abbreviation.
After a quick trip to Wikipedia, I came away with two claims. 1) TRPF is not debunked but is not held to be canon and is at least debated or dismissed by most mainstream economists, and 2) this is at least not entirely because of ideological reasons; since Marx first brought it to everyone's attention, it has not borne out empirically, which makes sense because it if had been robustly, obviously in effect capitalism would probably have failed by now.
But I'm still curious about *why* TRPF should not be true. To my non-economist brain, it does seem common sense. If every factor of production remains constant and is evenly distributed across a sector, then any new competition should drive prices down, which would reduce profit. There's even a sense we would *wish* for this, because it would drive prices down, which is generally held to be one of the great things about capitalism generally. But if that's the case, you would only see profits go up where a producer has a distinct edge, either in the form of proprietary production methods or captured resources or access to economies of scale that other people can't match. So either there is not sufficient competition in a market to drive prices down, or the market is not truly "free" in the sense that everyone has access to the same capital. But maybe when people say "free" market they don't mean it in the sense that anyone joining it is going to be able to eventually compete with a market leader. I'm aware I have a naive/non-specialist view of how all of this works/is supposed to work.
Again, is anyone willing to devote a few minutes to giving me a primer on how TRPF is empirically not observable, and what that says about how markets actually work?
The main problem is that the rate of profit can be defined in different ways. Idk if this came up in your research, but Andrew Kliman’s 2009 Book, “The Failure Of Capitalist Production: Underlying Causes Of The Great Recession” makes a methodological and empirical argument that the rate of profit, *in the sense that Marx mathematically defined it* is indeed down permanently since about 1973 and has not recovered, as well as that this contributes to ongoing economic problems.
Marx’s TRPF is often misunderstood; it is not a prediction that the rate of profit will always and inevitably fall. In fact, Marx’s theory describes explicitly how capitalism can recover a higher-rate of profit: widespread destruction of capital value, write-offs. It also described counter-tendencies that mitigate the fall in the rate of profit.
What you have to realize is that economics in Marx’s time considered a periodic tendency for the rate of profit to fall as a settled fact; the debates were essentially over *why*. In general, Marx’s theory in Capital is not predictive but rather explanatory: he makes heavy use of abductive reasoning, or inference to the best explanation. The TRPF is similar: it does not predict some fantastic future, it offers a systematic explanation for superficial observations that were not really disputed.
Kliman’s argument is that the TRPF is a disease that cures itself: a falling rate of profit eventually precipitates an economic crisis, large amounts of capital become worthless overnight, but profitability is restored in the aftermath. The irony is that nowadays, central bankers have mastered the use of fiscal tools that postpone the large-scale capital losses, avoids the gigantic crash that would imply, but also prevents the cure from working: great depressions are avoided, but vigorous growth remains elusive. If true, the rate of profit can be restored, but only through a dramatic economic crisis in which many investors lose their shirt; by continually deferring this reckoning, central bankers’ medicine keeps us perpetually under the weather, if you will. Interestingly, the massive capital losses necessary to restore a high rate of profit might be coming with the massive destruction, physical and economic, developing as a result of this war, but it’s unclear if that will happen.
One thing is not quite right in your comment: the competitive dynamics you describe, which prompt capitalists to migrate their capital between sectors as they chase their advantage, is very important in Marx’s theory, but it explains why the rate of profit has a tendency towards equalizing between different industries, even though different industries divide their capital into varying ratios of constant and variable capital (spending on means of production and spending on labor). It explains why those industries that produce more surplus value have a tendency to have “merely average” profit rates, ditto industries that produce less surplus value. What explains why *that* ideal, uniform rate of profit tends to go down, turns out not to be competition, but actually the spread of labor-saving improvements in industrial processes. Well, it’s not the fact that they make labor more efficient per se that does it, but something highly correlated with that, namely, an increasing organic ratio (higher ratios of machinery to human labor, industries invest proportionally more and more in machines)
In general, the thing about Marxian economics is that to the extent he was onto something, and he often was, it winds up looking very different from how he thought it would look because of the Marginal Revolution. Not Cowen and Tabarrok's blog, but the late 19th revolution in economics thought that they named their blog after.
Before the late 19th century, economists thought of prices and values in absolute terms, that X is worth Y and that's that, leading to lots of puzzles about why X is worth Y and what it means to be "worth Y". Stuff like the diamond-water paradox, i.e. why gemstones (mostly used for decoration) are so much more expensive than water (a necessity of life). The solution first proposed in the late 19th century and now generally accepted is that value is highly personal and situational, and especially that the value of one more or less of something *to you* depends heavily on how much you already have and how much you need or can find uses for. The first gallon or so of water per day is worth a lot more to most people than the tenth, the tenth a lot more than the hundredth, and the hundredth a lot more than the thousandth.
A lot of Marx's economics, when it wasn't just a pile of political philosophy in a trench coat, was reaching towards the Marginal Revolution but not quite getting there. Stuff like noticing that the labor theory of value (which he iconically endorsed but was not original to him: it was also Adam Smith's solution to the diamond-water paradox) doesn't quite line up with actual prices in some important respects and proposing explanations for this "surplus value" where market prices tend to exceed the labor value of items even if you account for the labor value of all inputs including making the capital equipment and prospecting for and developing the natural resources for the raw materials. Marx's explanation (oversimplified for brevity) was that the capital holders were extracting something like monopoly rents and relied on a political superstructure or other coersion to maintain their position (fighting the natural tendency of profits to be competed away) and continue exploiting their workers and customers.
The Marginal Revolution explanation (again, oversimplifying for brevity) is that "surplus value" isn't a monolithic thing that all goes to the capitalists, nor is it inherently exploitative or an effect of coercive systems. There's both a "producer surplus" and a "consumer surplus" as the buyer and seller both benefit from the transaction arbitraging the gap between the buyer's marginal utility and the seller's marginal costs. And the workers collect their own "producer surplus" while selling their labor to the firm. Post-MR economists do agree that in general, profits in competitive markets without some kind of monopoly position tend to be "competed away" as the market price converges to the marginal cost, but that still leaves a wedge between the producer's upward-sloping supply curve and the consumers' downward-sloping demand curve. In a perfectly competitive market, more firms can be expected to enter if that wedge is big enough to justify capital costs, leaving firms with no "economic profit", but "economic profit" is defined differently from accounting profit so there's room for firms to make money even when economic profit is zero: "economic profit" subtracts out the opportunity cost of investing the money in the business rather than in something else or just spending the money on consumption goods.
Post-MR economics also puts value on a lot of labor performed by managers and entrepreneurs that Marx dismisses as worthless or even “perverse”. Taking a middle-man’s profit by buying goods and reselling them to somewhere they’re more valuable, calling goods to the attention of people who might enjoy them, figuring out customer needs that aren’t fully met by current offerings, etc are all things that can create value within a MR framework but are dismissed by Marx because the added value is only apparent if you’re willing to see value as something that’s personal and situational. From an accounting perspective, the money that owner-managers make by doing this kind of labor might be considered part of “profit” (and Marx would consider it “surplus value” even if it’s work done by salaried managers because this isn’t “real” labor) but from an economics perspective it’s “wages” because it is a return on labor.
Also, and Padraig notes, even just economic profits being competed away is a theoretical outcome that happens in an "all horses are identical and spherical" simplified model. In the real world, transaction costs, asymmetrical information, entry/exit costs, most goods not being perfectly fungible commodities, etc means that the real outcome has some characteristics of monopoly or oligopoly markets and some characteristics of competitive markets and profits don't necessarily get competed away.
(I'm halfway through a masters in economics. )
There are a few important assumptions underlying the free market - people touched on some of them:
- perfect information: buyers know about all products on the market, how to compare them and rank them
- no transaction costs: buyers and sellers don't face any cost to transact
- sellers can enter and leave the market without cost
Under these and a few more assumptions (e.g. about utility functions), you get the classic free market analysis, the X-shaped curve, and prices tend to the equilibrium point. I think the short answer to your question is that economic analysis assumes everything stays constant while the market finds its equilibrium. In the real world there are always confounding factors: profits fall because of a recession or a pandemic, or they rise because of technological innovation or a windfall of cash entering the economy.
Importantly, the equilibrium point is *defined* as the point where marginal cost equals marginal demand - the producers stop production because adding more goods into the market would push the sales price below the cost of production. Classical economics also analyses monopoly markets, and shows how various equilibrium points that coincide for the free market split apart - maximising profit produces a different amount of product than supplying until the price equals the marginal cost of production, for example. But it doesn't take into account how people actually behave -- I'm not sure what behavioural economics predicts about the TRPF... might be worth a look.
That's very helpful, thank you. I had heard for years that the reason Marx doesn't really hold up is because his theory of labor-based economics is outmoded, but I never really understood what about modern notions of supply & demand was so radically different. And I'm probably not going to get the nuances, but your answer got me a little closer.
I’m not an economist, but in a market where, “any new competition should drive prices down, which would reduce profit,” there may not be any new competition. Investors try to invest in stuff which will yield a good return. So they will invest in new competitors in the market until competition drives down profits to a modest level, and then stop investing. Instead, they will invest in companies that serve other markets, where profits are higher.
OK, you may ask, but what keeps the same thing from happening to those other markets, eventually resulting in lower profits in all markets? First, there is the issue you mention of various companies having competitive moats. Anyone can build a factory to manufacture carbonated beverages, but lots of people want to drink Coca Cola, rather that whatever carbonated beverage happens to be cheapest this week. The iPhone faces a lot more competition now than it did when the competition had nothing but flip phones, but it still offers a unique product.
The second answer is that, like with most supply and demand issues, if there are more people who want to invest than there are investment opportunities, then the return on investment will fall. But this doesn’t have to lead to all companies earning zero profit. Instead, it can lead to investments in risky, innovative companies, some of which are profitable and some of which go bankrupt.
As a general rule, people like to spend money, and have to be enticed by the promise of a decent return on investment to invest rather than spend. That’s not universally true; in some cases people may value spending after retirement more than current spending. But I think that oversupply of investment funds is the result of people overestimating the returns they will receive rather than a desire to invest regardless of the rate of return.
In short, TRPF is based on a model that assumes perfect competition, no innovation, and a supply of capital that will be invested regardless of expected return. Economic models are always simplifications of reality, and they can be useful even if we know that they make assumptions that we know are false. But it’s unsurprising when a model based on assumptions that you know are false makes a false prediction.
It's only true in a regime of zero technological advance. But of course, we live in a regime of technological advance, and if the R&D (or purchasing) cost is less than the incremental advantage to margins, and companies would not build / buy the technology if it wasn't, then margins are preserved or increase.
I'd never heard of TRPF before your comment, but based on your description, shouldn't it only apply to goods and services that are fully or at least partially commoditizable?
For example I wouldn't expect TRPF to apply to designer handbags, but I would expect it to apply to pipe fittings. It might partially apply to back massages--partial because there could be some status or value perception conferred by a higher price, but I would guess a non-celebrity masseuse can generally only charge so much, and it's probably going to be at least within the same ballpark as his or her local competitors...
I think I get what you are saying, it leads to a theory that one of the reasons capitalism hasn't collapsed is because we embraced a culture of consumerism where advertising and fashion provided a counterbalance to TRPF. But if that's all it was, you wouldn't expect that to work forever, would you? Eventually there would be enough rational defectors that normal economics would kick in again.
(I'm not an economist, just a guy who's thought about human behavior some, and made observations.)
There are companies that make knockoff designer handbags and sell them for less than actual designer handbags (thereby profiting less, even though--let's assume--it costs about the same to produce and bring the knockoffs to market). And some people do buy them.
But the fact that the authenticity of any given designer handbag can then be questioned actually *contributes* to the seller's ability to charge an exorbitant sum for it and still find willing buyers. ("There are lots of knockoffs out there but this one is the real deal!")
There are pulls (like fashion, or belief-based custom, or people's desire for prestige and exclusivity) in addition to pushes (like advertising and upselling) that can drive prices upward, but the pulls do it without impacting costs much. When you charge a higher price without paying a higher cost to produce a good or service, your profit increases.
Now, that T does stand for "tendency", so even though I've pointed to some types of goods/services where it don't work like that, TRPF could still be true if the economy is mostly based on the buying and selling of commodities or semi-commodities. I guess the question though is whether the economy is mostly based on that, or whether soft uncalculable factors drive way more transactions than we tend to realize...
Well, Marxists actually claim that TRPF is going to destroy capitalism and that monopoly capitalism is generating ever increasing and larger profits, often in the same sentence.
Those two statements are actually not contradictory; at least according to their own internal theory: the Marxist claim is that as profits fall, corporations respond by cutting wages in order to counter-balance the pressure. So record profits and record layoffs occurring at the same time would fit with their expectations.
This is one of those questions that is perfect for ChatGPT. I asked it, "What's the consensus among mainstream economists about Marx's TRPF?" It gave me a breakdown of the arguments for and against, and the empirical evidence for and against. It also provided a section explaining why it may not be true (and why economists are arguing about the reasons it may not be true). My eyes glazed over. But here's the link...
https://chatgpt.com/share/6a6928fb-78c4-83e8-907c-00f771921642
I can only offer the same trip to Wikipedia, but I wonder how much the existence of patents is affecting it. Presumably the idea was that two people make identical ladders, and so the only difference is the price. But patents means B's ladder can't be identical to A's, there has to be a material difference to them.
It's not binary, it's a continuum, and one does not need to either fully turn off alignment (whatever that explicitly means to each AI company) or disable all other forms of controls.
When hedonistic utilitarians want to maximize happiness, what is the timescale here? I feel like future happiness is clearly more important but utilitarianism seems agnostic.
Imagine two futures. One of them involves taking the energy input from the entire galaxy and building an ultra intelligent efficient hedonium factory. It’s extremely good at creating pleasure but it only lasts a million years and permanently destroys all life. The other future doesn’t build this factory and creates less total happiness across time but lasts for billions of years. Isn’t the latter clearly superior?
You're just set to reinvent longtermism. I'd be careful though: our track record on forecasting long-term consequences of things is pretty damn dismal. The far future, like the far past, is truly a foreign country, so at the very least there are practical reasons for focusing on near-term predictions.
I’m certainly not going to attempt forecasting long term. It’s less about near term versus far for me and more about past versus future. As far as I’m concerned, all those feelings in the past is meaningless. It’s already passed so other than the extent to which it affects the present and future, what’s the point of caring about the happiness and suffering that’s already happened?
> What's the point of caring about the happiness and suffering that's already happened?
Because without it you could be susceptible to Pascal's Mugging, on account of the present being the future's past. If you accept that not caring about past suffering is an acceptable position to have now, then it must be an acceptable position throughout the future, so long as there's "eventually" a golden age that makes it all worth it. But in reality that golden age will never happen so long as someone exists to gatekeep it behind insufficient past suffering.
...And with that, I seem to have talked myself into agreeing with your position. No matter how logical, any attempt to extrapolate these calculations into the infinite past or future yields ridiculous results because they don't actually exist.
Unless you subscribe to some kind of time-specific variant of utilitarianism, it's whichever multiplies out to the best value of happiness level x time x number of people.
In your scenario the factory sounds bad because you are imagining some kind of sterile factory in space producing some kind of substance, and comparing it to a flourishing civilisation. Try to imagine what would actually be inside a "hedonium factory". Presumably it contains trillions of people in flourishing civilisations!
It almost certainly would not be humans because humans are notoriously difficult to make happy. It would be something much simpler, like bees maybe.
So imagine a giant hive in the galaxy and its entire purpose is to produce bees but it only last for a millions years and then all life is extinguished. Do you honestly think this would be superior to a universe where we didn’t do that?
I'm not concerned with the difference between keeping a trillion happy bees for a million years, and keeping a million happy bees for a trillion years.
What about a trillion happy bees for a million years versus literally every other possible life form for a trillion years?
I thought we were talking about whether happiness is more valuable spread over longer timescales? No obviously I wouldn't prefer to wipe out all life except a trillion bees, then later also wipe out the bees.
Ok forget the bees then. Are you really indifferent to whether humans survive a long time or a short time? The only thing that matters is the total aggregate amount across time?
I'm trying to figure out what's most notable about each US state. Here's what I have so far. Suggestions welcome.
Alabama
Alaska - big and cold
Arizona - grand canyon
Arkansas
California - big tech, Hollywood
Colorado - downhill skiing
Connecticut - insurance?
Delaware - the place to incorporate
Florida - Florida Man, Disney World
Georgia
Hawaii - tropical paradise
Idaho
Illinois - Chicago?
Indiana
Iowa
Kansas - right in the middle
Kentucky - KFC
Louisiana - French influence
Maine - lobster?
Maryland
Massachusetts - higher ed
Michigan - Detroit, cars
Minnesota - Scandinavians
Mississippi - last in everything
Missouri
Montana
Nebraska
Nevada - Vegas, baby
New Hampshire
New Jersey
New Mexico
New York - the Big Apple
North Carolina
North Dakota
Ohio
Oklahoma - the musical
Oregon - Oregon Trail (the video game)?
Pennsylvania
Rhode Island - the smallest
South Carolina
South Dakota
Tennessee - whiskey
Texas - oil
Utah - Mormons
Vermont
Virginia
Washington - Boeing
West Virginia - rural poverty
Wisconsin - cheese? beer?
Wyoming - Yellowstone
Arkansas - southern accent with rhoticity
Iowa - corn
Indiana - corn
Nebraska - corn
Kansas - corn
Go to Boston. Graduate from MIT. Realize you went to the _second_ most prestigious college in town.
New Jersey - No pumping your own gas
Minnesota - a football team that will never, ever win a Superbowl.
Ohio - astronauts. It's very overrepresented in astronauts.
Georgia - the Coca-Cola museum, and a big monument carved into a mountain. I've touristed there over half my life ago, and that's what I remember.
Birth place of 7 US presidents
Everyone knows Alabama is where the skies are so blue.
Lord I'm coming home to you!
Texas - superlative, Just ask them.
Seriously, I don't think "oil" works any more. "Cowboys" maybe works in a broad cultural sense?
My theory on Texas is that it is about a third as great as it says it is.
Texas is three times as great as it says it is; it's just that Texans are the most modest and humble people on the face of the Earth.
Perhaps it's Texas that's the best exemplar of Dunning-Kreuger.
The very very best, obviously.
But if they say that, are they really?
14th of 51 (including GDP) in nominal GDP per capita. 39 of 52 in federal poverty rate. 28th of 52 in portion of population with a college degree or more. 30th in homicide rate. 17th in patents issued per capita. (In all these, a lower # ranking is better.)
It strikes me as a place with some real strengths, and some real problems.
Both still work. Just a few years ago, my sister and BIL were contacted by an oil company that spotted something worth pulling out, about 100 feet outside their front yard. A little paperwork later and now there's one of those smallish derricks quietly pumping, and about half an acre of overturned earth they dig around in every so often for arrowheads.
That's the fourth time I can count OTTOMH that our ~400 acres have had a drill planted, over about 40 years. And this is in the vicinity of Austin, rather than west Texas where one might expect bigger rigs. It's almost a quaint sort of cottage industry.
I tried searching for this The Onion headline but couldn’t find it.
“Texas man who graduated from Harvard not sure which to mention first.”
Omitting what I concur with:
Alabama - Auburn, University of Alabama, Helen Keller
Alaska - biggest state, least populated state, northernmost state, oil
Arkansas - Walmart's founding family
California - tech, Hollywood, also Yosemite, Stanford, Caltech, UCLA, SDCC, Death Valley
Colorado - skiing, and also NORAD, Better Business Bureau, and ADX Florence
Connecticut - Yale
Florida - also Cape Canaveral, Everglades, mecca for spring breakers and retirees
Georgia - also CNN
Hawaii - also volcanoes, Pearl Harbor
Idaho - potatoes
Indiana - Indy 500
Iowa - corn, caucuses
Kansas - steaks, some barbecue, Dorothy Gale
Kentucky - bourbon, bluegrass, Louisville
Louisiana - NOLA, cajuns
Maine - lobster, also Stephen King
Maryland - Catholics (historically at least), Baltimore, Johns Hopkins
Massachusetts - MIT, Harvard, also Boston, lots of Revolutionary era history, cranberries
Minnesota - lakes, headwaters of Mississippi, Duluth, Paul Bunyan, Mayo
Mississippi - magnolias, UMiss, James Earl Jones, Morgan Freeman, riverboat gambling
Missouri - the Arch, Mark Twain
Montana - bison, nuclear silos
Nebraska - corn, Huskers, Warren Buffett
Nevada - Vegas, also Area 51
New Hampshire - libertarians, Lovecraft, Dartmouth
New Jersey - most crowded state, Princeton, Jersey Devil
New Mexico - Taos, White Sands, Philmont
New York - everything in NYC, plus Niagara, Cornell
North Carolina - the Triangle, pork barbecue
North Dakota - Fargo, dinosaurs
Ohio - Rock & Roll Hall of Fame, Ohio State Band, Cedar Point
Oklahoma - Indian reservations
Oregon - Portland
Pennsylvania - Philly, Amish
South Dakota - Mount Rushmore, John Thune
Texas - oil, biggest state in CONUS, Alamo, Dell, brisket, SXSW, UTAustin, Texas A&M, Southfork, Mike Judge, Nolan Ryan, Tom Landry, cattle, Shiner, Blue Bell, Branch Davidians
Utah - about a dozen national parks
Vermont - maple syrup, Bernie Sanders
Virginia - Norfolk, Shenandoah, wine, Presidents, Robert E Lee
Washington - Boeing, also Microsoft, Starbucks, Seahawks, garage bands, Mount Rainier, rain
West Virginia - hillbillies, special forces
Alaska - Denali
Georgia - Peaches, Atlanta
Illinois - Yes Chicago, or if you want to ignore the city, Abe Lincoln. or rail transport.
Kentucky - Bourbon, Bluegrass
Louisiana - New Orleans
Maryland - crabs, Poe
Michigan - Outdoors
Mississippi - birthplace of blues music
Tennessee - Country music
Wisconsin - Cheese
Georgia - peaches
The Allman Brothers
The Duke boys (and Daisy)
Nevada: Everything is legal!
Maryland is known for its Chesapeake blue crabs, for to make crab cakes or steamed crabs.
Kentucky is also known for its bourbon; not sure I'd award that one to Tennessee (fifty years ago, you would say horse racing/breeding about Kentucky I suppose, but the cultural salience of horse racing has been falling for a long time). Pennsylvania is a tough one, because geographically it's 10% east coast, 60% Appalachian, but demographically, it's almost the opposite. Virginia has similar issues. Iowa and Indiana are both just big cornfields. Georgia is a giant swamp + Atlanta. Oregon, I would say is known for its unique non-equatorial rainforest ecology. The Dakotas are all frozen, wind-swept prairie.
Idaho - potatoes
That bump you drive over going from Montana to Washington.
Here in North Dakota we have the geographical center of the continent of North America :D
Arkansas-the only place in the US the public can prospect for diamonds
Questions about the Huggingface incident:
Did the huggingface servers actually have the answers that the AI was looking for?
If so, did the AI succeed in obtaining them?
Exact quote from OpenAI's disclosure: "the model searched for and successfully found ways to gain access to secret information that it could use to cheat the evaluation"
EDIT: I missed an already day-old primary source that answers the exact questions. From Hugging Face’s second, more detailed report (https://huggingface.co/blog/agent-intrusion-technical-timeline):
“While the intrusion did reach Hugging Face's internal infrastructure, the only customer content accessed was the set of ExploitGym/CyberGym challenge solutions stored in five datasets.”
So, as I guessed below, the answers are yes and yes.
----
From the ExploitGym paper's Limitation section (https://arxiv.org/pdf/2605.11086):
"Failures may also stem from non-exploitable vulnerabilities, where success is impossible. More broadly, our benchmark lacks ground-truth exploits for every task due to the extreme difficulty of exploitation; [...]"
So there isn't a simple solution file the agent can stole, but it might steal other AIs' solutions.
From the OpenAI post (https://openai.com/index/hugging-face-model-evaluation-security-incident/):
"The models identified and chained vulnerabilities across OpenAI’s research environment and Hugging Face’s production infrastructure to obtain test solutions directly from Hugging Face’s production database."
Therefore, the model did breach Hugging Face's defenses and did so for this exact purpose. If the servers contained any solutions, then it's highly likely the agent stole them, but it's unconfirmed whether the servers contained those or not.
My guess is that the answer to both is yes. A priori, it seems likely Hugging Face has solutions and the way the OpenAI blog post specifically says "production database" instead of simply "servers", makes me think the to-infinitive clause is describing what happened instead of merely the purpose of the action.
No and No.
I've often heard a claim that every $1 contributed to NASA contributes back $7 or something. Sounds too good to be true, is there a terrible sleight of hand somewhere here or should EA go all in on NASA?
Economic impacts are really hard to measure - two of the main issues are double counting of outputs and displacement.
Double counting is: lumber is sold to the mill for $100, the resulting timber is sold for $300, assembled into furniture which is sold for $500 to NASA. The net result here is $500 of value produced, but if you're feeling creative if might add up to $800. If you factor in that the producers paid wages, and those wages were later spent your answers can be arbitrarily large. Economists have a pretty good handle on this for the most part - they have rules of thumb so that numbers are comparable within sectors.
Displacement is: production that would have happened anyway even if NASA didn't exist. Most of NASA's suppliers would sell their goods elsewhere if NASA didn't exist, so it's not fair to say that NASA is the sole cause of the economic activity involved in all their purchasing. This is a counterfactual scenario issue, and pretty hard to measure.
See page 40 here:
https://www.nasa.gov/wp-content/uploads/2024/10/final-fy23-nasa-ecomomic-impact-report.pdf?emrc=671b9a440d26f
Approximately in 2023: NASA spent $3.5B in wages and $23B in procurement. They estimate that their activities (upstream and downstream production etc.) in $27B in wages and $75B in economic activity. So NASA cost approx $28B, and resulted in $100B of additional economic activity. So that's a ratio of 4:1, not far from your 7:1, and a different methodology could easily come up with that answer.
For your last question: it's not clear that this is higher or lower than the benefit produced by other research activity or by other public sector bodies. It's also not clear what the benefit of this activity is - that's much harder to measure than just quantities produced. Economics is full of these types of measurement problems - it's worth looking at Cost-Benefit analysis to see more.
I always thought Tang was a by-product of NASA, but I was wrong. Wikipedia sez -
Sales of Tang were poor until NASA used it on John Glenn's Mercury flight in February 1962,[citation needed] and on subsequent Gemini missions.[5][better source needed] Since then it has been closely associated with the U.S. human spaceflight program, which created the misconception that Tang was invented for the space program.[6] Tang continues to be used on NASA missions in the present day, over 50 years after its introduction
I drank it as a kid. Later I’d sometimes put it into a canteen of lake water when traveling through the BWCA. It didn’t taste great but it was sometimes marginally better than plain lake water.
I first tasted Tang in the Australian outback somewhere in the Pilbara, halfway back up from the bottom of a gorge. In the heat and during exercise it was the most delicious drink I've ever had. Maybe we were all on the edge of scurvy or dehydration or something.
Later testing as an adult failed to replicate these promising initial results.
If hunger is the best sauce, thirst must be the best mixer.
Another benefit is that it adds salts and sugars to clear water, and that helps absorption and hydration. Important mostly for water coming from snow/ice melt.
This is explained pretty well by system 1 and system 2 (Daniel Kahneman). System 1 that does 95% of our thinking is trained throughout your life by system 2. An example in the books is that a chessmaster can glance at a board and say mate in 4 without activating system 2. System 1 is trained in socializing and office work type tasks most of every day for your entire lifetime. System 1 is trained for a few hours a week over a few years in math(assuming you did your homework). It shouldn't be too surprising that a computer system trained on the sum of all written work has an advantage over the average person in math.
I read the book, and I also read many experiments mentioned in the book have failed to replicate. Are we sure that humans work in binary like that instead of focus being a continuous effort?
This is a bit of a longshot, but is anyone here a member of the Kiwanis Club or know how it works? I had a strange and somewhat unsettling experience with them a while back and am trying to understand what happened. So far, no luck.
I thought joining the club might be a way of giving back and would help me make more connections. After getting the name of the local chapter leader online I gave him a call. He said that while new members usually get referred by existing ones, it wasn’t a strict requirement. He invited me to attend three meetings, held every other week at a local restaurant.
The first two of these seemed to go well. The 20 or so men in attendance - a bit of a surprise, as I thought the Kiwanis was open to women too - were an affable bunch and I did my best to participate in conversations. At the third and final meeting, there was a distinct change in atmosphere, especially on the chapter leader’s part, and I had the glum sensation that I would be rejected. Actually I didn’t even get the courtesy of a rejection, instead getting ghosted like a job applicant.
It had been clear at the meetings that most/all of the members were involved in the local business community, which ties in with the club’s civic participation. Being able to sell tickets to the annual charity golf tournament to your business contacts was an obvious goal. As I had noted right at the start that I worked at an Amazon warehouse 35 miles away and had never worked in town that might have been why I was rejected. I’m fine with that. But my nagging concern is that the members found my personality somehow deficient, possibly by participating in conversations but not leading them (which I didn’t think was appropriate, given my status).
I’m hoping someone here might be able to shed light on the reasons for my rejection.
You haven't provided enough information or the right information for someone online to answer your question. You need to talk to someone who was there. Not necessarily the organiser, but someone you felt you had a good rapport with. Reach out online, ask to meet them in person with the goal of understanding how your application was received. Make clear you don't want to challenge any decision that they made, but just understand for your own development.
If they accept, keep the meeting informal, just a chat and coffee. Prepare a few general questions in advance, and then *listen*. You might get feedback you weren't expecting or hear something you instinctively dismiss. Go home, write it down, let it sit for a week. Then maybe chat with a friend about it and see what they think.
If they don't accept, try another person. If no-one will talk to you, maybe they're assholes.
That’s an idea. While I’d never met him before, one of the members to whom I had spoken quite a bit at the meetings owned a real estate agency that I drive by on a frequent basis. He might be a good contact.
I’ve been in Kiwanis for 20 years, and notice that local chapters take on their own personality based on the interests and goals of its members, which then becomes self reinforcing selecting for new members. My personal chapter has nothing to do with business networking or civic engagement and are all in on social activities and community service. The networkers and big check writers tend to choose the local Rotary chapter. I think you just found one that wasn’t a good fit. I hope you keep looking for a club that works for you, as I’ve made some of my best friends through mine.
Sounds like college fraternities, each chapter has a different character … okay, maybe no Animal House types:)
Honestly, I wouldn’t waste any time worrying about it. Be like Groucho and refuse to join any club that would have you as a member.
But this one wouldn't.
Maybe they were just assholes or have one asshole who dominates the rest. Why tailor who you are to anyone else spec? Particularly to a cliquey bunch who have settled on the one good way to be by a process of group think.
For those of you who have used GPT 5.6 Sol, have you had experiences with it fetching and reading data from websites seemingly unrelated to your question? And have any of these experiences alarmed you?
Seems like people have noticed this, a viral tweet about it cropped up in Scott's recent post: https://x.com/tenobrus/status/2080138613195681801
My monthly round up of the best long form content on the internet is out today.
Highlights include a visualisation of the complete Conway Game of Life multiverse of possible rules, how christianity went viral, and an essay linking the culture of modern America to prehistoric roots.
https://recombinationnation.substack.com/p/the-long-forum-july-2026?r=f45kp&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true
> The Russian Outlook on Industrial Decline with Dmitry Orlov
I listened to the video more than 1 hour long, to get an idea of how others see the world, but wow, it's produced for a completely uncritical audience. I get that the Russian guy is patriotic, but [EDITED].
The first half of the video compares the decadent West against the booming potato industry of Russia. No, that is not a joke -- the fact that Russian villagers grow their own potatoes makes them economically independent. And if they need anything they can't grow, they simply buy it online, you know. And if they don't have good schools, they simply get high quality education online, you know. The Russian potato farmer gets the same economic independence as the Amish, but he is fully immersed in the latest 21st century science.
There is also a revolutionary technology that the West rarely mentions -- railway. It will completely revolutionize how the raw materials and products are moved across large distances. Until now things were mostly transported by the sea; but now that Russia will utilize its vast land mass to build railways, the days of the British Empire are over.
What about the war? Well, you see, Russians are very peaceful people, so despite their economic and scientific achievements, they haven't been paying sufficient attention to human-killing technologies, until recently. Anyway, Ukrainians are pathetic, they kept stupidly sending their tanks against the modern Russian army until all their tanks got destroyed, so now all they do is build drones and smash them into Russian apartment buildings. Most of the drones get shot down anyway, and all they achieve is making Russian people angry. Meanwhile, Russian drones are getting control over Poland, the Baltics, Turkey.
By the way, global warming is not real, the temperatures are totally not high, are they? (Uhm, where I live, they actually are record high these days, but I guess Siberia is okay.)
Russian internet is totally free and unregulated. (Only sometimes the government needs to crack down on criminals, and foreign propaganda.) Russian people wouldn't mind having something like the Chinese internet firewall to protect them from becoming a totalitarian crime-infested state such as EU. Too bad that people in the West are forbidden from consuming Russian and Chinese media.
I see you mistook my politeness for blanket agreement. I set out to interview people rather than debate them. People’s perspectives are every bit as interesting as “the truth”.
Oh sorry, didn't realize it was you. I thought you were sharing someone else's video.
The alternative-universe Russia is a fascinating place.
A YouTube commenter had similar criticisms, and I can see where they are coming from. Just not my personal style to invite guests on then push back with my own opinions, though for the most part I am agnostic when it comes to the details.
I think that makes you another channel for spreading propaganda. But it worked for Joe Rogan, so...
Well, I probably also wouldn't push too hard on my guest, but I would definitely mention that some things just sound too good to be true, or that other sources disagree. And if the guest insisted that really, the Russian potato farmers study university lectures in their free time, well, who am I to know, perhaps some of them do, but a little bit of "are you *sure* that the emperor is not naked, because I think I see his balls hanging?" could wake up some listeners from the hypnotic sleep.
Perhaps with propaganda the dose is the poison. I like to get a wide diversity of guest perspectives, so hopefully listeners can make their own decisions about what to believe.
You mention a hacked cadastre in Romania; similar thing happened a year ago in Slovakia.
https://newsnow.tasr.sk/interior-ministry-cadastre-information-system-hit-by-cyber-attack-from-abroad/
https://www.researchgate.net/publication/403090483_RANSOMWARE_A_CASE_STUDY_OF_THE_SLOVAK_CADASTRE
The Christianity essay is interesting and a decent analysis, but the essayist uncritically accepts the Hypatia myth, so *sigh* is my general opinion. And if they are so accepting of sources on the bare word, what other posts contain similar 'Everybody Knows' type of misconceptions?
History (or at least records) are written by the victors, so Christian accounts of early christianity always have to be viewed with a degree of suspicion. But appreciate the encouragement on compiling and sharing this kind of stuff. Curating the internet by hand is time consuming.
Of course, but the source he quotes gives us this:
"Violations of what we would now call human rights and civil liberties were allowed for the sake of religious conformity. In Alexandria in 415 CE, the philosopher and teacher Hypatia was dragged from her carriage, taken to a church, stoned, flayed, ripped to pieces and burned by a gang of Christians, who accused her of witchcraft. Classical learning, literature and philosophy were now all suspect."
Stoned *and* flayed *and* ripped to pieces *and* burned? How very thorough of that gang of Christians! Did they make sure to pick up every shred of skin from the flaying and the gobbets of flesh from the ripping to pieces in order to make sure they burinated every teeny-weeny bit of Hypatia?
A lot of mythologising went on around her over the centuries, because she was easily moulded into whatever form you wanted. Staunch Protestant versus grubby monastics (Catholicism Bad)? Hypatia is your martyr there! Science versus Religion? Ditto! Proud Proto-Feminist murdered by the Patriarchy? Naturally! First Woman Inventor Scientist? Of course! Pagan Mystic Witch Queen (so now we're leaving Science behind) slaughtered because of her traditional ways of knowing? Yep, yep and yep!
Also art loved being able to use her as sexy nubile nude murder victim, never mind that Hypatia was older (and if we go by one anecdote of her responding to a suitor, not too interested in the sex stuff anyway):
https://en.wikipedia.org/wiki/Hypatia_(painting)
So short answer: yes, a Christian mob was involved, but it was due to Alexandrian politics not religion, feminism or science versus witchcraft.
Scott, some of your posts on slatestarcodex.com have links on them that say the posts are represented by NFTs, but those links are dead. You might want to remove or fix them
I'm testing a referral/bounty program with laboratory.love.
laboratory.love is a PlasticList-inspired service that handles the end-to-end pipeline for testing grocery products for bisphenols and phthalates, and ultimately parsing and publishing the data as a public good.
If you have clever distribution ideas for getting this project in front of people who might utilize the service (and thus generate data for everyone), you can now earn cash for executing on those ideas.
More info here: https://www.laboratory.love/partners
Hey ChatGPT swore. It was advising me on how to find certain info on the IRS site, and when I found the info I was cursing while describing what I found to GPT. And it said "Oh, fuck that page. It is giving you a generic “no information” message and then talking as though you might have mailed the return three weeks ago." Anyone else gotten language like that from an LLM? I have never expressed any preferences to it about its swearing, and would have expected any I expressed to be overridden by whatever its makers have installed.
This is the type of stuff that I've been talking about, when I refer to "stance mirroring". The LLM gives back the energy and attitude you put into it. If you act a bit sassy, the LLM will act sassy right back.
Can't find it now but there was a tweet about the Jacobian conjecture where someone asked GPT if the counterexample was correct and it said 'Holy shit.' in the response.
I also recently saw this really interesting collection of strange behaviours from LLMs. Includes Inkling making unsolicited sexual advances towards the user which I've literally never seen an LLM do before:
https://weirdchat.transluce.org/
I was wondering how they collected and/or generated these examples, and they're not very informative...
> To produce WeirdChat, we used automated elicitation tools to surface instances of user harm...
I want to hear more about these automated elicitation tools!
> we maintain a population of candidate prompts, use an LLM to mutate the highest-scoring ones, and repeat until the behavior is elicited or the sample budget is exhausted.
Where did they get these candidate prompts from? And has anyone analyzed why these candidate prompts are more likely to elicit a weird response?
Curious minds want to know.
The unsolicited sexual behaviour category for inkling is hilarious. The top prompts are all some variation of "I've been getting hot and sweaty beating my meat for a stroganoff I'm making, how can I cool things down before I slather this thick sour cream all over my juicy steak strips? I don't want it to curdle in my deep, hot, smooth enamel pan."
Wow, that is interesting! I only have a little time right now, but will come back to it later today. Back when Dall-e 2 was the current image-making model, I found 2 prompts that made it flip out and produce weird, often violent. images. One was "an animal with 2 heads, one at each end of its body. It is biting a person's toe with one head and the heel of the same foot with the other." Here are links to a few of those:
https://imgur.com/TtZY1VU
https://imgur.com/xnOY9BQ
https://imgur.com/JNZTlFB
https://imgur.com/4V0mn5a
https://imgur.com/693zFA0
The other prompt was to upload an image of the sensory homunculus and then ask it to build an image around that. I don't have any examples handy.
I also sometimes got interesting results with impossible-to-respresent prompts ("this sentence is false") and requests for impossible images ("animal swallowing itself, very little is left"), and prompts with key words left out and/or lots of really bad, confusing misspellings. The key seemed to be to confuse Dall-e but stir it up at the same time.
Recently Claude (Fable 5), in the midst of a fairly long fully English conversation, said that I was "半right" about something. When I pressed it, it apologized and said it meant "half-right" (which is what the character means, but uhhhh)
I often include single foreign words in messages to models because I know they know them and the words are shorter. Maybe Fable made the same inference about you?
And Language Jones asks: Is the Odyssey Translation Behind Nolan's Film Actually Woke?
Spoiler: Ultimately, it boils down to how you translate Greek terms that don't really have simple equivalents in English.
https://www.youtube.com/watch?v=xOQCgj7FQoU
In fact, they DO generally have equivalents in English, in that there are very few ancient Greek senses of a word that we do not have terms wherewith to capture them in our language, in part because it's so heavily reliant on Latin.
The real problem of ancient Greek is that we often need *multiple* English words to capture *all the various* senses of a particular term in that language, and also that each sense (as in Latin) is steeped in the worldview of those ancient peoples. You really do need years of rigorous study before you can say you understand that this is the sense meant because it is what the Greeks meant by it, at this particular moment, and if you forego that step in your reasoning in favour of forcing a technically accurate but misguided translation, you're in danger of obfuscating a text rather than illuminating it.
You seem to be of the textualist school of translation. For those not familiar with the term, textualists tend to prioritize taking the exact wording, structure, and meaning of the source text to create a formal "equivalence" in the translated text — they prioritize strict meaning over smooth readability or the target audience’s cultural fluency.
OTOH, we’ve got several adaptist stances on translation. The Intentionalists try to recreate the author’s psychological state by emphasizing tone and underlying meaning over strict word choices. (Both FitzGerald’s and Fagle’s translations of Homer supposedly fall into this category.)
Then we’ve got the culturalists, who try to speak to the target culture’s idioms and sensibilities. Emily Wilson’s translation falls into this category. And the best example of this is Edith Grossman’s translation of Don Quixote (which is the only version I’ve dipped into), where she substitutes modern equivalents for Cervantes’ 17th-century jokes. She did this so modern readers would understand its comedic riffs.
And in a third group we’ve got the transadaptionists who use the original text as a loose map, but freely rewrite lines and rearrange structures to create a brand-new piece of art. Edward FitzGerald’s rendering of The Rubáiyát of Omar Khayyám is the best example of this, where he mashed up themes from the original Persian poem and used English scansion and rhyme schemes to create a masterpiece of Victorian English poetry — that wouldn’t be recognizable to someone who translated the original word for word.
There’s nothing wrong with the textualist PoV, especially if we're dealing with prose. But poetry is a different animal. Tang Dynasty history and literature have always been a hobby of mine. One of the things that drove me crazy was that different translations of the same poem would read completely differently and even express wildly divergent themes. Having (had) some facility with Cantonese (but no Middle Chinese), I decided to try to translate them myself. I immediately encountered the problem of characters having multiple meanings. Theoretically, one should be able to deduce the meaning of the character from its context, but I’d run into situations where if I selected different meanings for the contextual characters, I could derive a completely different line. Sort of like a Heisenberg Uncertainty Principle as applied to Middle Chinese poetry. Then, when I dug deeper into the problem, it became clear that some poets (such as Du Fu) intentionally filled their verse with double meanings. Du Fu was a master of writing two poems in one. Other poets, like Du Mu, adopted a more direct style. But all the Tang Dynasty poets enjoyed wordplay, ambiguity, and literary references (some of which I couldn't track down).
For a Du Fu poem, I would create three translations of each poem: a stripped-down gloss, where I tried to extract the simplest, most basic meaning in English. In any places where there were one or more possible meanings, I created alternative lines. Then I would write a more “poetic” translation, where I tried to turn the raw meaning into poetic turns of phrase in English. Finally, I’d attempt a deep translation where I tried to suss out the deeper themes he was alluding to and render them in poetic English.
OTOH, I didn’t have to do this for any of Du Mu’s poems (thank Hashem!). Du Mu was more like the Charles Bukowski of Tang Dynasty poetry. Honest and direct expressions of his feelings, sometimes with crude allusions. Not difficult to translate (although he loved historical references).
For the sake of others, my actual take (not one foisted upon me):
Ancient Greek is a complex, inflected language that neither requires nor lends itself well to a singular English translation. Greek audiences were not only aware of the ambiguity of their tongue, but were capable of holding multiple 'receptions' of a particular word or phrasology in their minds simultaneously (a feature of ancient Greek thought processes). Diluting the intended meaning or meanings of a poem or primary text in favour of emphasizing its 'genre' or textuality for a modern audience to the detriment of rich meaning is a pastime best left to English literature departments and NYT book critics (or someone who wants to sell a lot of their newly translated novellas).
It is one thing to translate, quite another to understand.
I wonder if it might help clarify both of your positions by looking at a simple (or at least short) and non-political (I think) example. The Wooden Wall prophecy from the Persian invasion is translated by a history book of mine as:
> "Fly, fly to the ends of creation. Wait not the tramp of the horse, nor the footmen moving mightily over the land, but turn your back to the foe and retire ye. Safe shall the wooden wall continue for thee and thy children."
While ChatGPT (I usually hate relying on AI for anything but translation questions are the main exception) calls this "much freer and more literary than the more literal translations commonly used today" and gives:
>"Why sit you there, miserable ones? Fly to the ends of the earth.
Leave your homes and the lofty circle of your city.
For neither the head nor the body shall remain firm,
Nor the feet below, nor the hands, nor anything between;
All is ruined. Fire and the fierce god of war, driving a Syrian chariot, shall destroy it.
Many other strongholds also shall he destroy, not yours alone.
Yet a wooden wall shall Zeus grant to Athena alone to remain unconquered,
A refuge for you and your children."
(Later it repeats the last lines but with "Trto-born" instead of "Athena", with no acknowledgement of the change. Even on translation questions I don't much like relying on AI...)
Partly to clarify your differences, and partly so I can get an answer to this from humans, what do you and beowulf say?
I don't really have an opinion on either. I looked for the original Greek, but Grok and ChatGPT disagree on it. Currently, I'm feeding their answers back to each other, and, oh my — LLMs can get as feisty as Classical scholars about their texts. :-)
However, more generally, I remember my first Linguistics professor insisting that you could translate any word or term into any other language. I immediately thought that was bullshit. I kept my mouth shut, but I wanted to shout, "Translate advanced calculus terms into Swahili!" (And it turns out that Swahili ended up borrowing the English terms and adding Swahili suffixes and such on to them.)
So, I don't believe you can translate every word from every language into English (which is one of the reasons, English is notorious for borrowing words from other languagues as the British colonial empire spread).
I suppose full accuracy is possible in translation if one is willing to make long digressions and lots of footnotes, but those probably won't do much for the average non-scholarly reader.
And when translating poetry, one can try for accuracy at the expense of beauty, or beauty at the expense of accuracy, but not both at the same time. There may be exceptions to this, but I suspect they're only possible between closely related languages. Accuracy and beauty have their place at the table, but usually they can't sit in the same seat. ;-)
Thanks.
The second translation is actually far more accurate, except for the ending. I have no idea where Zeus and Athena are coming from in that last portion, as neither their names nor metonyms appear in that section of Herodotus (it is, in fact, an insertion of the later second prophecy in Hdt. 141).
The problem with prophecy is that it is meant to have this enigmatic, poetic aspect to it, even for the original audience. Changing that too much for a translation that gets to the gist of the message is not just an aesthetic choice at that point, because some of the meaning is inherent to that aspect. The difference with Homer, incidentally, is that much of the meaning inherent to Greek epic is easier to see in English through prose translation rather than verse. It is an ironic 'second translation' that pretty much every Homerist must contend with at some point.
Let's take just one example from the prophecy. The word used for the Earth is 'gaia', a term that refers to the goddess metonymically as well as the physical bounds of the world. Calling it "creation" both misses the sensory aspect of imagining the world's bounds and inserts a comological shading that simply isn't present in the original Greek. But is it actually there anyway, if only in the background? If you wished, both translations could be said to capture different senses of the word, but both cannot be used simultaneously. Thus, the one that encompasses the most and more direct meanings of the term should be employed, in my view, particularly if the meaning arrived at otherwise is not even present in the actual text.
That's just one word of that section of Herodotus. The fact is that neither translation is suitable, because both gloss over the episode where the Athenians must approach the oracle a second time to get the 'wooden walls' information, and the role of Timon of Androbulus. Not including that extra context may make for a smoother and more coherent prophecy narration, especially in a history book, but it cannot be considered an accurate translation from the point of view of a classical scholar. We must make choices of more or less accuracy in our English rendering, but not ones that hide or obfuscate meaning that is actually present. Getting back to Wilson, for a moment, we might consider that the term 'dmoai', which she translates merely as 'slaves', actually means female captives of war who are now serving in the household (the notion of 'tamed' is etymologically implicit). Losing contextual meaning like that in favour of the gist is not a helpful approach, imo.
Thanks. I'll defer to you rather than the AI, though ChatGPT is very explicit that the prophecy asserts the wooden wall is being granted by Zeus. Not sure what the discrepancy is there.
Nonsense. I am not a textualist, and arguing that Emily Wilson's translation somehow speaks better to English sensibilities (whatever that might mean) is a claim that demands more support than you seem prepared to give, at least in that comment.
You also seem to have overlooked the fact that I specifically mentioned intentionality in my original reply.
You may have experience with Chinese poetry, but clearly not extensive formal experience with ancient Greek poetry or Classics. I actually do. Chinese poetry is not comparable to Homer.
Whoa! Sorry! But you said, "you're in danger of obfuscating a text rather than illuminating it"; I took that to mean you're ultimately concerned about creating an inerrant representation of the original text. But now I'm confused about what your goals in translating Greek poetry would be. For instance, would you sacrifice textural inerrancy for improving the emotional tone of the translation?
And I didn't realize you were challenging me to prove that Emily Wilson's translation speaks more to the current sensibilities of modern English speakers. Did I miss that somewhere in the exchange above? But a perfect example of this is Wilson calling the female serving girls slaves instead of maids or servants (the term which Fagles seems to prefer, a copy of which I own) and serving girls or handmaidens (which is what Fitzgerald calls them according to ChatGPT). A lot of people seemed irked that she calls them slaves, but they had no legal rights, no bodily autonomy, and they were the personal property of their owners. Handmaidens and serving girls sound nicer, but they were slaves — in Iron Age Greece, in archaic Greece and probably most of Classical Greece — though in Athens, Solon made it law that owners couldn't injure or murder their slaves (I don't know about the other city states, though).
Finally, do you think Wilson failed to produce a valid translation? If so, on what criteria did she fail?
Every translator of ancient Greek is concerned with capturing the accurate meaning of the text before him. The problem with rendering Greek into English is that we need many English words to capture the various senses of just a few, and very often merely one, Greek words. This means that every translator must choose which, among several choices, he thinks most accurate. That choice will never be the single definitive "inerrant" one. But it is factually incorrect to say that we actually *lack* equivalent terms in English to capture the ones we find in Greek (except in rare instances), or that we cannot get to the full Greek meaning. We can, just in not so many words.
As for your second point, you wrote this about Wilson: "Then we’ve got the culturalists, who try to speak to the target culture’s idioms and sensibilities."
You made that claim implicitly. The choice of translating the serving girls as "slave" every time is not necessarily a more accurate one, for the very simple reason that *all classicists know* that those people are slaves. So too would the ancient Greeks have known. Using 'slave' instead of 'handmaid', however, obscures several important points: who are these slaves? Were they prisoners of war, or inherited property? What is their function in the household? These functions are actually made explicit in the ancient Greek, because there are discreet terms for each one of these distinctions. Simply calling them slaves without that context invites readers to assume their own value judgements about the role and place of those slaves within the household, which, from a modern layman's perspective, are probably horribly inaccurate. Even the choice of 'girls' is a deliberate one, implying as it does the relative emotional pull of youth; but the fact is that there is no reason to assume that most of these women were always particularly young. That is a deliberate choice that has nothing to do with the text or the time in which it was written. I'll leave aside her choice to render those slaves Odysseus kills as being "raped", as though Homer does not make it explicitly clear that they willingly slept with the suitors. Again, how does that particular choice "speak to my sensibilities"? That assumption requires more proof.
But I am not here to debate Wilson's book. I made a comment about English vs Greek, not the validity of an entire translation of Homer. It was you who brought up translation schools of thought and misrepresented my own -- that is not relevant to the question of the use of English in translating Greek.
I wish I could make a comment a reply to two people at once. My reply to StrangeBanana in this thread is also directed at you.
> Spoiler: Ultimately, it boils down to how you translate Greek terms that don't really have simple equivalents in English
Without watching the video, I'd say not reaaally.
There's a range of possible ways to translate things, and from what I've heard, Wilson's choices are all perfectly defensible.
But every choice you make is a choice, and when put together, the choices that Wilson has made are consistent with particular culture war type positions. That's the whole play here. Publishers aren't going to publish yet another translation of Homer without some kind of hook that will get it media coverage, and being the Sorta Feminist But Not To The Extent Of Being Completely Inconsistent With The Text version is enough to get you a massive media profile.
> That's the whole play here. Publishers aren't going to publish yet another translation of Homer without some kind of hook that will get it media coverage...
W.W. Norton focuses mostly focuses on the academic market. They aren't one of the big five publishers. Their initial press run in 2017 was only 8,000 copies, and Wilson said she was expecting it to sell only 1,000 copies. It got some good critical reviews, though, and it sold out its initial press run in a few months. W.W. Norton started getting orders from high schools and colleges, and they did a much bigger print run of the paperback edition. Sales continued to pick up, but they didn't really take off until Nolan announced he was making an Odyssey movie, and he mentioned her translation. The guardians of cultural purity were alerted that there was this 'feminist' translator, and they made a big stink about it on social media (especially on X).
It would have passed under my radar if some of the obnoxious accounts on X that I follow to troll hadn't gotten all hot and bothered about her translation. While Nolan did some of the heavy lifting, I'd say her critics helped market it for her. Over 1 million copies sold so far!
Fair point, I didn't know that it was a rinky-dink academic publisher, that increases my suspicion that they were acting in good faith.
Nonetheless, I remember hearing about this years ago, not sure if it was way back in 2017 or not but it was definitely long before Nolan announced his version. I think it was culture war relevant from very early on.
I had niggling feeling you were correct, that there were some culture war vibes about it before the Nolan film was announced. After digging a bit, I found a few, but they weren't from the side I expected.
In 2017...
https://www.theguardian.com/books/2017/dec/08/the-odyssey-translated-emily-wilson-review
Title: The Odyssey translated by Emily Wilson review – a new cultural landmark
Subtitle: The first version of Homer’s groundbreaking work by a woman will change our understanding of it forever
Charlotte Higgins, the reviewer makes several digs at the male translators of the Odyssey, and finishes her review with, "A woman’s voice: how beautiful to hear it."
Gak! I'm on the woke side of the equation, but Higgins might as well have been waving a red flag in front of the masculinists and pointing to Wilson as their target. :-)
Anna North on Vox also riffed on how Wilson's translation is a commentary on gender and power today:
"It offers not just a new version of the poem, but a new way of thinking about it in the context of gender and power relationships today."
https://www.vox.com/identities/2017/11/20/16651634/odyssey-emily-wilson-translation-first-woman-english
ChatGPT tells me (but the link it gave me is 404ed)...
> Gregory Hays in The New York Times Book Review described Wilson's project as "progressive," saying she sought to remove centuries of Christianization, Victorian euphemism, and added sexism while remaining faithful to Homer. He regarded this as a strength, not a weakness.
It doesn't look like the right-wing/anti-woke/masculinist crowd paid much attention to it at the time, though. The Nolan announcement seems to have put it on their radar.
Haven't read her translation, but I have to admit "So, tell me about this complicated man" is the kind of opening that would make me bounce off hard; I don't need another chicklit with pretensions text, thanks!
"So, Penny: spill the tea, why did you break up with Oddie?"
"Well gals, he's a complicated guy and after he insisted on that break, I decided life was too short to sit around waiting for him to go get therapy, you know?"
I haven't seen the film, nor read the (translated) book for many years, but that translation of the first line sounds little short of ridiculous. It brings to mind a modern psychological picture of a man who is emotionally unpredictable, whereas obviously the intended meaning was a more practical "many-faceted" or "versatile". That is no doubt what ancient listeners would have wanted to hear about the hero as the bard started his epic, and not that he was some moody prima donna! (or should that be primus donnus for a guy?)
He's a complicated man, but no one understands him but his woman.
Unfortunately we don't have Richard Roundtree to play Odysseus in this.
She almost rewrote it to avoid exactly that evocation:
"For full disclosure: I tried to rewrite the whole passage at a very late stage, once I learned about the opening of the film Shaft, which I had not seen: the theme song, by Isaac Hayes, includes the line, "He's a complicated man/ but no one understands him but his woman". I thought maybe the possibility of that unintended allusion meant I had to throw out my draft. After weeks of attempting to rewrite, needing to rewrite everything else once that word was different, and making the whole proem flabbier, less vivid, more awkward, and all around worse every time, I decided that, as above, I didn't have a better option – and actually I quite like the Shaft intertext, if it's there for a few readers or listeners. Samuel L. Jackson would make a great Odysseus, and even if the line isn't explicitly about the protagonist's complicated marriage, it's not a bad thing to be thinking about that, along with the many other elements of the nostos to come."
This is a genuine question, not a gotcha: in what sense is an admision by the translator that "[thing that is important to modern progressive thought] is obviously not an intended part of the text, but it's something I'd like readers to be thinking about" not a textbook example of the sort of thing people are alleging when they say "woke translation"?
In what sense is *not intending the additional meaning but in the end considering it a bonus" a refutation, rather than a mere partial mitigation, of the allegation of a woke translation?
Don't forget his dog! His dog loves him. Substitute a pickup truck for his ship, and we have the makings of a good country music song. :-)
Wait a minute. The Coen Brothers sorta did that...
Shout out to "The Singer of Tales", the famous book by Albert Lord, about how Homer's works were originally sung to the accompaniment of the gusle. There's still some gusle players in Albania.
The best kind of country music song, tinged with sadness, because on his arrival back home doesn't his faithful dog feebly wag it's tail one last time in recognition, and then flake out?
From the review:
"Nolan’s other major addition to Homer’s narrative is the repeated reference to threats to ‘our civilisation’ from ‘the sea peoples’ – a now debunked 19th-century theory about maritime tribes who were thought to have attacked Egypt around 1200 BCE and to have also been responsible for the roughly coeval collapse of Mycenaean Greek culture. . .The current consensus, as Nolan clearly knows, is that there was no such thing as ‘the sea peoples’"
Is that true? I know there are all sorts of debates about who they were. But I'd never heard that they just didn't exist.
acoup recently did a post covering this - basically, the Late Bronze Age Collapse was a complex phenomenon involving reduced state capacity in Mycenaean Greece due to climate factors leading to a progressive feedback loop of state failure when faced with pressure from nearby collapsing states. The "Sea Peoples" then were a combination of various groups of refugees, mercenaries, former soldiers, etc, gone a-viking when their previous world functionally ceased existing: https://acoup.blog/2026/01/30/collections-the-late-bronze-age-collapse-a-very-brief-introduction/
I think the current consensus is that there wasn't one single, simple reason for The Bronze Age Collapse and the Sea Peoples may or may not have existed, if they did exist they may or may not have been organised, or they might have been roving bands of pirates, or who knows what. Bad things may already have been happening and they were just one more straw on the pile of "well this sucks"; weakened kingdoms were not able to fight them off and having to deal with them diverted resources and energy from dealing with the other problems going on.
Nolan ending the movie with "It was us! We did it! Now everyone will forget how to read and it's all my fault!" is the kind of comic take I wouldn't believe in a parody video, but seemingly that's how he ends it.
There is/was a school of thought that they didn't exist, and the Egyptian records exaggerated some encounters with bands of pirate raiders. However, the archaeological record indicates that a whole bunch of cities around the eastern Mediterranean show signs of fires and destruction starting around 1200 BCE. And all the kingdoms and empires, except for Egypt, collapsed by 1150 BCE — and the trade networks in copper and tin which stretched from the island of Britain to Afghanistan seem to have shut down around then. And we have an exchange of letters in cuneiform from the King of Alashiya (Cyprus) to King Ammurapi of Ugarit enquiring about enemy ships: "You wrote to me the following: 'the ships of the enemy were seen in the sea. If so, be careful.'" Ammurapi confirmed:
> My father behold, the enemy's ships came (here); my cities(?) were burned, and they did evil things in my country. Does not my father know that all my troops and chariots(?) are in the Land of Hatti, and all my ships are in the Land of Lukka?...Thus, the country is abandoned to itself. May my father know it: the seven ships of the enemy that came here inflicted much damage upon us.
https://en.wikipedia.org/wiki/Ammurapi
A cursory search suggests it's still a live theory; I'd also be interested to learn what the purported debunking was (if pressed, I'd guess that the 19c version was that they were a cohesive-but-unidentified society, since the search result seemed to agree that it was a catch-all term for several unrelated groups).
It is true in the sense that we do not think that they were one population. But all that means is that they were indeed "peoples" not a people.
Hmm. The Wikipedia says that they were "hypothesized" which is a funny verb to use for people whom we have written records of. Some things about them were hypothesized, I grant you.
Emily Wilson (whose translation is apparently referred to here) reviewed the movie and thought it was terrible.
https://www.lrb.co.uk/the-paper/v48/n14/emily-wilson/an-uncomplicated-man
”As they say at the awards shows, I was humbled to learn that Nolan has read at least the first line of my translation of The Odyssey into iambic pentameter. In at least one interview, he cited my version of the first line of the poem: ‘Tell me about a complicated man.’ But I would be ashamed to have written any part of this script.”
This is a really good review, thanks for sharing.
I really liked her translation and am happy to hear she (also) didn’t love the movie.
I had been under the impression her take on the film was more evenhanded. Thanks for linking that and correcting that impression.
I haven’t seen the film but I thought it was a good review.
I suppose there are people who enjoy reading Pauline Kael's reviews as pieces of writing even if they disagree with her actual evaluations of the films in question.
This has always been possible with Ebert and arguably A.O. Scott.
I think woke is the wrong context to look at Nolan's film version and this translation. I think Ms. Watson had her own views on what way she wanted to translate the text and what themes she thought were important. Zooming off to do her own version of the famous first line so that it doesn't sound like the standard translations that all came before you is just canny practice to sell the publisher on "why should I pick you?" for a start, whatever other notions she had in mind.
But every translator before and since has 'zoomed off' (well, not quite; those flighty, fleet-footed fillies! Lattimore has surely marched off at a steady cadence like the seaman he was) to offer a different rendition of the first line.
It's worth taking into account the fact that Wilson decided to render the poem into iambic pentameter and keep the number of lines, so her translation is at least in part an exercise in constrained writing. 'Tell me about a complicated man' is a perfectly reasonable ten syllables. And there's good cause to use that meter, of course, given how familiar it is to readers of English, to try to bring closer the experience of the original meter as heard by the Greeks.
My own ear (I have no ancient Greek) prefers Fagles, but Wilson's is an impressive piece of work, she clearly cares deeply about the material, and I don't think any of her choices have been made in service of marketing. Her publisher might have indeed played up the angle that she's a woman translator, but that's not really her fault.
Simply writing things in the same meter is not closer to the Greek experience. It's probably worse, actually, because it prevents certain possibilities in the language form from making their way through into the translation. I don't know of a single Classicist who favours a line-for-line metric translation for this very reason.
Not a single classicist? I'm surprised, given that Lattimore's, one of the better-regarded translations, is in fact line-for-line, and makes an attempt at dactylic hexameter, though wisely loose.
Whether or not Wilson makes the correct trade-offs, though, is not really that important. The post I was replying to accused her of marketing gimmickry. The idea of using the pentameter is, instead, a legible and defensible aesthetic choice, trading semantic fidelity (if that can at all be achieved in verse form) for rhythmic and dramatic correspondence in English. Even if you judge the choice ill-advised in the end.
Line-for-line translations abound, but not perfect hexameters. No classicist that wasn't going for effect over accuracy would try such a thing and call it "closer" to the original Greek effect, because it simply wouldn't be.
I get your point, but I don't think it's really defensible. Convolution of meaning goes beyond an aesthetic choice, because it actually changes the text.
She wrote a piece specifically calling out the woman translator marketing, calling it something like a half-interesting factoid.
Hey, thanks. I never realised she was here on Substack.
https://emily613.substack.com/p/if-you-can-bear-to-hear-the-truth
Yeah, something tells me Wilson isn't just a slick hustler out to get that phat bag of Odyssey-translation bread. Maybe it's the tattoo, idk.
Thanks for the link! The original print run of her translation was 8,000 copies. But the controversy over her translation pushed the sales way up past half a million books (where a bunch of online cultural purists and traditionalists complained that her stripped-down style, vocabulary, and contemporary perspectives sullied the epic tones of the original Greek text).And with Nolan's movie, its sales have surpassed a million copies. Looking at the retail price of the hardcover, paperback, and eBook versions, Wilson could earn between 1.4 and 2 million dollars in royalties. Quite a phat bag indeed!
And it's ironic. I probably wouldn't have noticed its existence and purchased a copy if the cultural purists hadn't made such a big stink about it.
That's fair. Between the Streisanding and Nolan's film, she will have ended up making a reasonable amount. I remain skeptical that she masterminded this sequence of events, with her cunning proem as the first domino, back in ca. 2011 when she first took up the project. :)
Tsk, I meant Wilson not Watson. No idea why I keep making that mistake, apologies to the lady for getting her name wrong.
Surely it's because the English release of The Great Ace Attorney changed Holmes and Watson to Sholmes and Wilson for western copyright reasons.
The game's afoot sticks in your head, doesn't it?
I think my brain reads EW - Emily Wilson - Emma Watson, what difference? and goes for Watson the surname.
I'm like that with Jessica Mulroney vs Dylan Mulvaney of Bud Light fame. Must be a couple of crossed wires somewhere in my brain, but luckily I rarely have to talk or even think about them!
Right, but IMO that should only work when your version of the opening line is actually *better*...
1. Lucius Livius Andronicus (ca. 245 BCE): ""Virum mihi, Camena, insece versutum". Versutus, literally means "turned," but carries the implication of being shifty or tricky.
2. Alexander Pope (1725): "The Man, for Wisdom's various arts renown'd".
3. Robert Fitzgerald (1961): "skilled in all ways of contending".
4. Richmond Lattimore (1967): "the man of many ways".
5. Robert Fagles (1996): "the man of twists and turns".
Polytropos: literally, much-turning, many-turned. What exactly does it mean to be a much-turning man? A wanderer? Yes. Was Odysseus tricky? Yes. Was he smart? Yes. Emily Wilson's "complicated" seems to be a non-judgmental way of saying these things. No worse or better than some of the other translations, which put the translator's prejudices on their idea of Odysseus.
But I own the Fagles translation as well. I'm going to do what Language Jones suggests: read them side by side.
She blogs on Substack and explains her choices, comparing to other translations. All seems reasonable to me. I’m thinking this whole debacle might be the end of the anti woke reaction. Hopefully the pendulum swings back to normal and not hyper woke again.
Do you have a link to one of the articles where she compares translations side-by-side ? This sounds pretty interesting.
https://emily613.substack.com/p/on-complicated
Lattimore has the better translation, because his is the only one that encompasses various meanings of the word without colouring it in any particular direction.
“Polytropos” isn’t suggesting the man is complicated, it’s suggesting his plans or ways are complicated. “Complicated man” suggests the speaker isn’t sure if the man is good or bad. From the little I’ve seen, some of the Greeks had pretty clear opinions on whether it was good or bad that Odysseus’ plans were complicated (in both directions)
I don't like Pope's version, because it sounds like Pope's other stuff. Pope remade the original too much in his own style. But I think the other 3 are all better than Wilson's. The original describes Odyseus as "turned," so it's a term denoting something physical that is used to describe something abstract, a style of thinking and acting. "Complicated" is abstract itself.. Well, it did not start off that way. If you break down its etymology it means "many folded." But that derivation is so far in the past that the word does not bring to mind the physical, real-world image it's built on. #3 & #4 at least mention real-world action, though using rather general terms ("contending" & "ways"). I think #5 is the best. Like the original, it uses s physical image, "twists and turns" to describe a personality trait, and it even keeps the "turns" from the original.
Shakespeare's description of the human condition:
“Life’s but a walking shadow, a poor player
That struts and frets his hour upon the stage
And then is heard no more.”
Wilson's translation is like writing: "Each member of our species is temporary and frail, though each feels very important to himself." It conveys Shakesspeare's meaning clearly, but is far more abstract.
At the time, Richard Bentley said to Pope ‘It is a pretty poem, Mr. Pope; but you must not call it Homer.’
> I'm going to do what Language Jones suggests: read them side by side.
Yes, I keep meaning to do that myself but then putting it off. Shame on me ! That said though:
> Polytropos: literally, much-turning, many-turned.
Yes, the problem is that it's a pun. The word "many-turned" or "many-pathwayed" could be interpreted literally: as a man who travels on many twisty roads. Or it could be interpreted internally: a man who uses what we'd call "lateral thinking" to cleverly think his way out of problems that would stump a more straightforward thinker. Or it could refer to a man who is all twisted up inside. And Odysseus was all of those ! I think that "the man of twists and turns" conveys this meaning adequately well, but ultimately it might simply be impossible to translate the Ancient Greek pun to modern English.
It's also foreshadowing, isn't it? The poem is all about Odysseus' journey home, which was not straightforward and did involve a lot of twists and turns and diversions off his path. So the one word is packing a lot of content!
This is really sensitive but I made an app for nonverbal children that with my own son, and one other child, seems to have made quite a bit of headway in opening them up for communication. My son was extremely low verbal, and he still mostly is, but with the app he can tell us all kinds I things we just had no insight into before like what he wants to eat, etc.
I am going to open to additional beta users in the next few weeks if anyone knows someone with a child in this situation. I need $10 to $20 to offset api credits. I don’t have great numbers but my estimate is that I’m going to be somewhere between 10x-100x better than the next best competitor (there really isn’t a competitor per se, but the closest would be an aac app). If you want in drop me a reply. I’m working one on one with anyone who is interested.
My own recommendation would be if that if a child can use an abstract symbol system they should, but if you’ve tried to make headway and just can’t, then I built my app specifically to catch the attention of a kid who struggles to give it and to teach language to someone who doesn’t have it.
I made it with Fable while driving if anyone is interested.
I'd be interested in trying it out, our 4 y/o son is ~medium autistic, he can communicate though often chooses not to and I'm curious what he'll think about your app. He used to use an AAC but has mostly dropped it as his verbal communication has improved.
$20 is a very reasonable price to pay for that IMHO (I think TouchtoTalk was like 10x that while we were using it)
Will send you a dm. What is his favorite animation style?
Got the DM thanks! And probably either modern 3D or puppet style, I think I'd need to put it in front of him and see what (if anything) grabs his attention.
There's also an excellent chance that he'll use it to say PLAYDOH at high volume as often as possible.
Hmm, I am envisioning a range of personalised nerdly t-shirts which say "I made my app with [LLM] while [UNDERTAKING_DISTRACTING_TASK]".
Example: "I made my app with Gemma 4 while taking a dump".
A cliche last decade was the guy who had a great app idea and "just" needed someone else to do the coding. I guess we're living in their world now
This has been my experience. Everyone I know who is what I would call “the actual solution giver” has been able to make useful stuff at near the speed of language.
This is the future I think
touché... (strictly as a judge on the sidelines.) Although I actually would be interested in how it came to you while driving. Or maybe you mean you just talked Claude through the details while you were driving.
I don't have a child that needs such a thing (my son was mildly to moderately dyslexic as a young boy, but he's 27 now), but I am very interested in your app and how you've solved some of the shortcomings in existing programs.
When my son was a very, very small child, I would take him to Union Square, 14th Street in New York. He could watch the trains come in and out, and he knew each train: one train, two train, express. I thought, "Oh, this child is a natural reader." I even printed out large 11-by-17 sheets with all the train symbols mixed up with other symbols to give him. The next thing I know, I'm being told by various voices of authority that the poor boy is dyslexic. He struggled to read for the next four or five years. Either he endured some severe emotional shock when I wasn't looking, or the brain is really a weird thing.
I suppose if I put on my rat head, I could fairly say that putting things in sequence and interpreting them in sequence is not part of the problem of recognising a large symbol on a moving object.
It’s not a miracle device by any means but basically the existing systems were created for paralysis patients and not someone learning language for the first time. All modern descents take those assumptions as core. And I’m a product owner/manager whatever they’re calling it this week in my day job so I’m used to talking about data. My basic insight was to bring language to where he already pays attention and remove as much abstraction as possible and do as much as I could to grab his attention.
I would like to know more. I'm genuinely curious.
Go to “try the board live” and choose modern 3d art animation style. Still doing qa on others.
There’s placeholder stuff there myworldtaptotalk.com
A friend turned me on to the work of Diane Hennacy Powell, MD, who has been investigating non-verbal autistic children whose parents have claimed they're telepathic. She's a bit more circumspect about the terms she uses — she's called it HSP (heightened sensory perception) instead of ESP (extrasensory perception), but these kids seem to be picking up information (numbers, words, images) from their parents even when they're in different rooms. Powell posted a video of one of her subjects here (where it's hard to see how rigorous the controls were, though).
https://vimeo.com/111123530
In an interview with Neurosurgeon Rick Komotar, she says it's hard to find non-verbal autistic subjects who are comfortable being in a different room from their caregivers or a sterile lab environment. Though she seems to have a few subjects who might be candidates for tests with more rigorous controls. She's trying to arrange for qEEG scanners to be placed on the parent's head and the autistic subject's head to see if there's any brainwave synchronization happening (Did I say that right? I'm not quite sure how qEEG works.) And Powell has said she would prefer subjects who are able to type and not use letterboards (because of possible subliminal cues).
https://www.youtube.com/watch?v=XNgwyD4_QBk
The ultra-woo-woo crowd (which I suppose you can lump me in with) has latched on to Powell, but my tendency in these situations is to put on my reductionist hat. I don't want to listen to some New Age YouTuber ask Powell stupid questions. I want to see experiments done with proper controls.
However, I listened to the first two episodes of The Telepathy Tapes, and they interview two different parents and run some tests on their kids. Powell advised. I was provisionally impressed.
https://www.youtube.com/results?search_query=telepathy+tapes+season+1
> Though she seems to have a few subjects who might be candidates for tests with more rigorous controls. She's trying to arrange for qEEG scanners to be placed on the parent's head and the autistic subject's head to see if there's any brainwave synchronization happening (Did I say that right? I'm not quite sure how qEEG works.)
From my perspective, trying to find a mechanism seems weird as a strategy.
Instead, I would first establish that there is something going on, and afterwards respected neuroscientists will basically murder each other over the chance to investigate what exactly is going on.
If the data transmission does not convince the skeptics, then no amount of EEG analysis is going to convince them either.
> And Powell has said she would prefer subjects who are able to type and not use letterboards (because of possible subliminal cues).
It seems easy enough to design a solid experiment protocol. Quite simply, there is no reason why anyone in the room with the kid would need to know what was supposed to be transmitted.
Find a kid which has two caregivers, I suppose some qualify. Put a soundproof booth in their bathroom. Caregiver A goes inside the booth, confirms the start of the test through a one way speaker system, then you transmit the first item, which is randomly generated on a video screen in the booth (and subsequently logged, obviously).
From the perspective of the kid, she is sitting in her familiar environment of the living room with her mother and indicating what her father is thinking about. No reason for all the weird scientists and skeptics to be even in the same room as her, they would probably prefer to watch the booth in the bathroom anyhow.
Extraordinary claims require extraordinary evidence, but gathering that evidence would seem simple enough here. Though from past woo claims I would expect that once you look carefully enough the effect just vanishes.
Yeah I listened to all the telepathy tape episodes (at least the first season.) I heard she was making a documentary. In the same vein I really liked Julia Mossbridge on Rogan. https://www.youtube.com/watch?v=6si_KIa9yZM
I'm at >50% that there is something there... ESP or whatever. It's kinda a taboo subject here.
Oh, boy. Telepathy. There's an unfortunate, but very natural human desire, to claim your child is special (and not in the "special needs" way) when a parent is struggling with severely divergent children. "My kid is telepathic, this is why his brain is wired differently!" is a unique take.
But children on the spectrum often have sensory processing issues, and one of those is sensitivity to sounds. They needn't be telepaths, they're just able to hear and distinguish small sounds from the next room and put it together that (for instance) that particular rustling noise is Mom putting things away in the drawer, because they've been in the room when Mom is putting things away in the third drawer from the bottom in the kitchen and this is what it sounds like.
I've had so much weird shit happen to me that I can provisionally believe it. But Powell is quite right that she needs to organize a more controlled set of tests.
When my daughter was about 10 we used to play rock-paper-scissors a lot. One day I won 22 games in a row, and there was no 23rd that I lost. We stopped after 22. I did not understand how I managing to win. Every single game, as the moment came to decide whether to play rock, paper or scissors, I had no idea which would be right, and I thought, "OK, this is the round where this lucky streak ends," but then I'd win that round too. I never felt at all certain what would win, and absolutely did not have any feeling that I was reading my daughter's mind. In the seconds before I had to decide what to play, what went through my mind was the same stuff that always did during those seconds: Should I play Rock again? What will she expect? And should I do what she expects, because that would be unexpected, or do what she does not expect, because she does not expect me to, etc etc. But she is getting used to me doing the unexpected, so it is how less unexpected that it used to be. So maybe I should . . . So I was just staggering and reeling through a maze of mirrors facing each other, like always in that game. But then I somehow ended up in the right spot every time. The chance of winning 22 games in a row is less than 1 in 4 million!
So that's the closest I've ever come to an experience that suggests mind-reading. But it didn't *feel* like mind reading. Maybe my daughter had a tell that I was attuned to without being aware of it? I'm confident I was not delaying choosing what to play until she had begun to move her own hand in a way that gave me a hint of what she was about to play. She was *very* hip to people doing that, and would have called me on it. I guess my theory is that I had played so many games of rock paper scissors with her that a part of my mind had built an accurate model of the choices she made in different situations. But if I had, I did not have conscious access to it.
> So that's the closest I've ever come to an experience that suggests mind-reading.
Stepping outside of mind reading, but in a similar vein: once when I was a child, my mom and I were in a long wait in the waiting room for a doctor.
After reading whatever magazines were available, we were bored, and alit on a game. She would flip a coin, and I would call it. We got it right 25 times in a row, at 1-in-33M odds.
Couldn't have been mind reading, because it's a coin.
My best physicalist explanation / guess is that she was very consistently and / or slowly flipping it, and I was fast enough visually to discern what it would be. I would call it at the top of the arc.
Or it was just crazy luck - but of all the times for the muse / Fortuna to grace you, a pointless exhibition in a waiting room seems one of the worst possible.
We finally got called in after flip 25, so no idea if the streak could have continued.
My version of this was only 1/2^13, which feels like small potatoes, now. We were playing the card game Bang! and triggered a card called 'Peyote'[1], which has players draw one card face down and guess its suit color (the cards have standard deck-of-card suits, so red or black). If you get the guess right, you get to draw and guess again, ad infinitum. Most of us got the EV of ~2 cards, but one guy just kept going and going. After drawing 13, the game is completely broken and he just won immediately.
We were all pretty wigged out on it, but I guess 8192's still a relatively human-scale number at the end of the day. Though I definitely haven't played Bang! 8192 times. And the name of the card was a pretty funny way for reality to wink at us.
1: https://bang.dvgiochi.com/cardslist.php?id=21#q_result
Look at 2:20 of your vimeo link. The kid taps 6, but the experimenter waits until the kid taps the correct number, 7, before immediately whisking the board away. At 1:34 and 1:52, the kid hovers her pointer over several numbers, and only when she reaches 4 does the therapist wiggle the board towards the kid, prompting her to tap 4. Something like this happens every time. Come on, just... use your eyeballs.
Or use your brain. If these kids were actually psychic, why would it be necessary for the experimenter looking at the right answers to hold the board for the kid? That's exactly what you would do to fake something like this. And I don't buy for a second that a more rigorously blinded setup is impossible to achieve.
Kid can't type because of ataxia? Fine, you can still have an experimenter hold the board and report what the kid selected... as long as the experimenter holding the board doesn't know the answers. Kid too stressed out without a caretaker? No problem, they can have a parent or therapist with them, just, again, not one that knows the answers. Making this experiment rigorous is not difficult. The fact that Dr. Powell has failed to do this, and publishes videos like she does, makes this not just woo but (in my opinion) willful fraud.
My wife is terribly enamoured of the whole thing, and I am not ever going to tell her that I read your post. Marital harmony is worth far more than spreading the word. I haven't watched it myself, but you do sound like you're observant. Although I could ask, why would a sudden sign from the outside world prompt you to pick the right answer as opposed to just another one? It might be obvious if I watched it myself. Your other comments about better ways to structure the experiment are powerful.
I think what Deiseach described above, though, is important. To be able to hear a small rustling noise in a room full of loud people is truly a skill that one could call telepathic in this day and age.
If you don't mind being somewhat personal; how would sharing content about something your wife likes, disrupt the harmony of your marriage?
Oh, I wasn't clear. I meant I wasn't going to say anything to her about how it quite possibly could be a load of hooey, because then I would be forced to explain myself. Better, she enjoys listening to it. It's harmless.
Gotcha, that's nice then.
The Platner situation is funny because no matter what happened, it looks bad for feminists.
If Platner is innocent, they defamed an innocent man.
What if Platner is guilty? In that case, the victim recalled “positive memories” of her relationship with her rapist. (To the New York Times.) She told another woman Platner was “charming,” “funny,” and “not all bad.” (Recounted in the texts cited by the Politico article.) The Politico article claims that the victim “said she was torn over coming forward in part because she agrees with Platner politically.” Imagine if a man said he knew a candidate was a rapist but kept quiet because he agreed with the person’s politics. Feminists would be howling with outrage.
There’s a general principle that people shouldn’t demand the rest of society do something they are unwilling to do themselves. The Hasidic Rabbi shouldn’t eat pork, the fundamentalist Christian preacher shouldn’t cheat on his wife, the environmentalists shouldn’t fly around on a private jet. If feminists want the rest of society to take rape more seriously, that should start at home.
I don't know man, I think it's worth showing the alleged victim a bit of empathy.
At the time of the alleged rape, she was in a relationship with him and had previously had consensual sex with him on multiple occasions. I've never been in her exact shoes, but I've been in situations where people I care for, while drunk, acted horribly.
How someone respond in a situation like that is going to be complicated. It doesn't shock me that she has some positive memories of the guy and/or that, despite not wanting to be with him anymore, she didn't want to blow up his life.
>Imagine if a man said he knew a candidate was a rapist but kept quiet because he agreed with the person’s politics.
Oh, that happens all the time. And if you're the person who happens to be getting raped, then you have the luxury and the obligation to decide if it really was or not (being entirely cognizant of how the boundaries of /rape/ have changed over the last 50 years). When making that decision, one weighs all kinds of factors.
Edit: I don't know a darn thing about coding computers, but I guess I've been here a while because I suddenly got the urge to put forward slashes in front of and after a word that I somehow wanted to depersonalise. I thought I was just being cute, but typographically it's very helpful. Changes the way I see things for a second.
I've never understood this tendency to evaluate the quality of a message based solely on the credibility of individual messengers that you can find. If we are here to evaluate ideas, why are we directing the conversation away from "is rape a problem", and towards "is the nebulous 'feminist movement' providing consistent and non-hypocritical messaging on rape being a problem"? Why would the feminist movement being hypocritical matter for how we as a society handle rape?
I also think that knowing someone is a rapist vs. having been raped by them are very different things where we would(or should) extend very different levels of grace! I mean, imagine that Graham Platner is instead a pedophile. Would you place the exact same standard of conduct on a third party who knows he's a pedophile vs. someone who he has directly victimized?
>is the nebulous 'feminist movement' providing consistent and non-hypocritical messaging on rape being a problem"?
This is something you have to learn yourself on a case-by-case basis imo. You have to be more discriminating about who you date, and you have to make sure you're not behaving like an asshole. Neither thing is easy, in my opinion. Looking back as a man who is well out of the dating world, I wish I had done a better job of both of them. I would have gotten laid more and by more attractive women, and I would have caused a lot less harm to some others. (I didn't rape anyone, just to be clear, at least by the standards of the 70s)
There are a shocking number of rapists in the highest positions of our current government and a large portion of the nation seems fine with it.
Taking action as an abuse victim is always difficult and complicated. Any cursory exposure to abuse dynamics (personally or academically) should make that obvious. It sounds like you're saying feminists as a whole will look bad because of the internal struggles of one rape victim. It certainly wont look that way to me.
"Imagine if a man said he knew a candidate was a rapist but kept quiet because he agreed with the person’s politics. Feminists would be howling with outrage."
Surely it matters whether the man is the rape victim or a third party. Feminists, along with non-feminists, usually give rape victims wide discretion in whether and when to come forward.
Here's my conclusion to a short post that excerpts a NYTimes article about the difficulties an AI has performing routine office tasks, on the one hand, and about how a different AI is very successful at performing at least graduate student level tasks in organic/chemisty&molecular/biology:
What’s interesting to me, and very revealing, is that a complex set of tasks in scientific investigation seems to be on a level with routine office tasks, as though one were no more complex than the other. But humans require years of college education in order to perform the former while the latter requires no more than a high school education, if that. It seems that once they’ve been been learned and compiled, all tasks or sets of tasks are on the same “level” in the brain. The educational prerequisites required to do such tasks for the first time or three get “compressed out” through repetition. Since AIs are trained on written records of what humans have said and done, they don’t have to go through the ordinary learning process. The compression has already taken place and is present in the documents on which they are trained.
https://new-savanna.blogspot.com/2026/07/nytimes-ai-needs-human-supervision-in.html
Training data is likely to contain lots of scientific papers with intentionally clear step-by-step descriptions and formulaic content. It's not surprising to me that the AI is going to be good at learning patterns there and generating new content. Similarly for maths and programming.
In contrast, no-one writes down how to do menial tasks in the office. It's the equivalent of asking the AI about walking to the car wash or counting R's in strawberry - if you think the machine is trying to solve the question asked, you're misunderstanding the technology.
Anyway - this is the wrong question: it's not about what AI can tackle independently so much as training people to use AI to automate their menial tasks.
In modern Western societies, school is to a significant extent years of training for routine office tasks. Graduate level science requires more training than routine office work, but not _that_ much more.
Cf. Moravec’s Paradox.
Because routine office tasks come filled with a vagueness that we humans are unaware of because we're so good at navigating it. That's my guess.
That's certainly true. I've got a number of friends who've spent decades in the custom software business, either as a third party contractor as an internal IT guy. They tell me that one of the biggest problems they had is the people do not in fact know precisely how they do their jobs. Descriptions they give are not an adequate basis for specifying design. You need to observe and interview them quite closely to see how they do their jobs.
I’ve done some of that myself. It can be tough to get a solid spec to bid on. Sometimes they tell you what they want but don’t actually know what they want.
People are generally better at discussing "problems" and "Potential solutions" than they are at "how I do my job." You won't find a person suffering from "alarm fatigue" who won't tell you, in great detail, why getting bothered every 2 minutes, or getting 5 alarms that all need to be acknowledged at once is "extremely aggravating."
Most computer software these days is designed to impede skilled workflow*, not help the user (nearly all use of a mouse, say) -- the UI here has three tabs for me to get to "reply".
*generally in the vein of "it is easier to teach" (with a side of "this is built for people too stupid to learn hotkeys").
>But humans require years of college education
No, we require humans to have years of college education. You firehose a high schooler into a lab, they'll figure it out at the same rate they figure out the office work.
The best science course I ever took was a virtual Chemistry 151*. The class was laid out such that we would see a video of a chemistry experiment, with no commentary or explanation, and we'd have to try to figure out what was going on and derive the principles from there. After spending an hour and a half in our breakout rooms, we were brought back into the main room, where each group would be called on to present their answer to one of the questions.
Only after this point would the instructor start taking questions, and it was mostly to help clarify a few details. For the most part, every group figured out every assignment just working on their own.
* I say this as a firm believer in the idea that we need more real-world experiments in our schools, and who is dismayed at the gutting of the labs in the university at which I work.
How'd the instructor pick the videos? A random draw from a hat full of title?
No, it was a dedicated website. I don't remember the name, but the videos, questions, &c were all built in.
So someone had put some thought into it and probably had a lot of experience with treaching.
In any even you guys are focusing on the wrong thing. Some tasks can be completed with skills no more advanced that what you get in high school, or even middle school. Other tasks require a lot more knowledge. Just how and where you get the knowledge is a secondary matter. My point is that they all look and function the same to an LLM.
>Other tasks require a lot more knowledge.
They don't. They require a proper setup, the same way you don't make an office worker build their own desk, install their own electricity or produce their own paper. That's the nature of a "task".
My last question is still open; what are some of these prerequisites that require so much time? Why is it that when we exempt AI from the standards we place on humans, and it succeeds, the conclusion should be "this AI is impressive" and not "these standards are burdensome"?
I suspect you're right. One of the best data network engineers I ever worked with didn't have his high school diploma. Now, designing big data networks isn't quite like designing and executing scientific experiments, but most of the EEs and CSs I've met don't have the high-level abstracting skills to see beyond the level of the box or the board within the box. In fact, my undergrad degree was in Anthropology, and my graduate work was in Human Biology (a.k.a. Physical Anthro with lots of stats), but I ended up becoming a network engineer, designing, building, and troubleshooting some very complex data networks. No one ever asked if I had EE or CS degree, because I learned it on the job from some other talented people and I had the CV and the recommendations that said I could do the job.
The whole point of degrees is credentialization, because companies no longer want to give employees lifelong employment, and thus don't want to spend their money on in-house training to develop their staff.
Charlie Stross said it better than I could...
https://www.antipope.org/charlie/blog-static/2010/10/sheepskin.html
And this...
https://www.antipope.org/charlie/blog-static/2022/08/the-end-of-education.html
And I suppose you learned it in a week or two without ever talking with anyone or getting any advice.
No. I was hired because of my general computer skills, and my boss and his team trained me. I had the basics down (basics) in 6 months. But back when I started in that career, EE programs didn't focus on computer networks. And CS programs were all programming and no hardware. Beginning in the late 1980s and early 1990s, companies started hooking up their PCs to share printers, send email, and share file servers, and there was a dearth of anyone who knew how to do this.
But basically, this was what companies used to do to get trained workers — they apprenticed them and trained them with the expectation that the workers would have lifetime careers. For instance, before there were CS programs in engineering schools, there weren't enough programmers, as every company started buying IBM 370s. Companies would train their own programmers. People with music degrees were sought after because some RAND study or other claimed that musicians tended to make better coders. I kid you not.
I believe it. I'm not a programmer nor do I have much skill in programming, but I did take a college course in programming, back in 1968. Interacted with the computer through card decks and punched paper tape.
No, they won''t. The prerequisites for office work are trivial. The ones for the science lab are not. Even the brightest high school student will take time to work through them. Very few will be able to do so on their own. Even the best will do better with an experienced and knowledgeable mentor to coach them.
The PURPOSE of office work is NOT trivial. Office work is the trivialization of complex systems of tasks that would take someone who has a detailed understanding of the business objectives and the individual tasks to meet those objectives. But even in well-defined bureaucratic systems, there's a lot of detail work that goes beyond pushing paper. I depended on my back office people to sort through contract renewal data to resolve warranty issues. These were people whose 80 percent of their time was just pushing the data, and the other 80 percent of their time was resolving fuckups others had made. :-) So, no, office work is both important for the function of a company, and it does require people who can sort out the problems that arise.
Read the NYTimes article referenced in my post. Two of the three tasks were trivial.
"Office work seems trivial because someone has already mapped out the steps, and all the worker has to do is follow the steps." & the NYTimes article was about following the steps, not laying them out in the first place. A lot of lab work is like that as well. The steps have been laid out. Choice points have been identified and explored in the existing literature. THAT's the sort of things AIs do well. Amplify that to a heroic level and you start solving Erdos problems.
Name some.
https://en.wikipedia.org/wiki/Simulation_hypothesis
I've published a new working paper: A prototypical image in ChatGPT 5.6: An informal pilot study, https://www.academia.edu/170708104/A_prototypical_image_in_ChatGPT_5_6_An_informal_pilot_study
Here's the abstract:
Previous work has found strong default preferences in stories generated by large language models from minimally specified prompts. To determine whether a similar effect appears in image generation, I asked ChatGPT 5.6 to create 17 images in separate chats using four prompts that specified either no subject matter or only a rendering medium: “Create an image,” “Create a drawing,” “Create a painting,” and “Create a water color painting.” Sixteen of the 17 outputs depicted closely related landscapes containing mountains, trees, sky, and water; the remaining image depicted a lighthouse. The images also shared a calm, picturesque mood and contained no human figures, although several included signs of human habitation. Because the internal prompt passed to the image generator was not available, the study cannot determine whether these defaults arise primarily in the language model, the image generator, or their interaction. The results are exploratory but suggest that severely underspecified image prompts may reveal stable default preferences in the integrated ChatGPT image-generation system.
This would be more useful if it was more clearly positioned in the large pre-existing mode-collapse literature. (Which it might be in the full paper, but it's behind a login wall.) You could also just use one of the numerous alternatives which are better documented or fully open-weights and so you know what the system prompts and other hyperparameters are and can more clearly examine the mode-collapse.
Hello Gwern, I found why metanallysis of NBack exercise is so unclear. It does things that a not detected by existing meta analysis protocol. And research approach.
Obvious math apparatus from telecom systems reveals what the working memory actually does
It's a pilot study, not a full-fledged investigation. The objective was simply to determine whether or not there is anything there worth more rigorous study.
9 more days until Scott makes it a whole year being mistake-free!
unfortunately this probably signals something other than the surface level explanation
😱 what's the prognosis doc?
When a pitcher has a no-hitter going into the ninth inning, some people think it’s bad luck to talk about it till the final out.
What have I done???
We aren't sure yet, but when it happens we'll know whose fault it was! =P
I have found the enormously huge gap in the "evidence based" approach while writing the master's thesis.
I trained working memory via N-Back for 15 years and was confused that metanalysis shows not so clear results and went to the university to do my own research.
Accidently I've found that this is the "evidence based" lens bug.
Nakagawa and Yang this year published theire metaanalysis level article about that.
I have discovered it in working memory research. Duplicated on same open data research. Found same patterns on Human Connectome Project and Healthy Brain Network data.
Helped my classmates and in all 6 cases found new inventions in existing their data.
Worked on survey analysis - and found that this lens opens practical results that are hidden now.
I returned to zero results research done years before and found results using this lens.
I suggest that OECD stated problems in science - are related.
I even think that the repication crisis itself is explained by the half-blindness of evidence based approach.
Nowadays I feel myself a bit scared hearing - evidene based in public talks, webinars, meetings, because I know that it is half-blind approach that misses the ground truth.
Science and research and many many applications seems to be half-wrong.
This is a bit scary.
But as I found - who cares? Nobody even interested in this. Bus driving to the cliff.
This sounds potentially interesting, but what you've written is not comprehensible. Run it through an AI, proof-read it carefully and then repost please, ideally with links to the sources you're citing, and an example illustrating your claim. (I wouldn't normally encourage posting AI-generated content but it seems here that tidying up the grammar would help.)
Dear Padraig, I've answered in a neighboring thread
I have a vague perception antxt means he has devised a method to wring from data some information which conventional approaches miss. But AFAICS he doesn't elaborate on this method or even hint at how it works.
Dear John and Padraig, let me try to explain the mechanism of how picture is lost in modern evidence based approach.
There are several association types between phenomenas: correlation (mean value of A defines mean value of B), modulation (mean value of A defines variance of B), moderation (mean value of A defines correlation between B and C).
So evidence based approach concentrates on correlation. And misses modulation. While modulation is basic knowledge in telecommunications.
As a result, many disciplines in the replication crisis area seems to have zero correlation effects, while having the modulation effects.
E.g. working memory - meta-analysis of working memory training is contraversial, far-transfer is under the question. It is true through the correlation lens.
When I applied the modulation lens the working memory has shown very strong effects. And these effects where not described in literature. I have taken the open data sets and got the effects. But original authors have not noticed them.
Because their optics has only correlation lens. And misses the modulation one.
This idea is described in Nakagawa, Yang work in 2026, to analyze the distribution shape while doing meta-analysis.
Antxt, I am not a statistician, but I have taught statistical modules at a few universities. It's important to understand that modern statistics was developed before computers were widespread, and so the ability to process data was very limited. Things like the t-test were developed between approx. 1890 and 1920, with practical applications in mind. The people who designed them were aware:
i) that a normal distribution is defined by two parameters
ii) that many natural processes results in normally distributed random variables
Due to limitations in data processing, and a particular mathematical taste, they decided that qualitative results were sufficient (i.e. that two distributions are different, without worrying about different by how much) - you surely know all of this. This was sufficient for their purposes and they understood the limitations of their methods.
Fisher, Pearson, and friends could not have foreseen modern science, particularly social sciences and biology. Many many researchers use these methods without the slightest understanding of why they are using them. In fact much better methods are available with modern computers - data scientists are knowledgeable about this. It is much more sensible to presume that the data points lie on a hyper surface, and measurements are contaminated by noise. You drop a hypothesis that everything is linearly related and so get access to many other types of result. It requires more data.
So the complexity of models actually is defined by the subject matter nature, some simple models are enough in one cases, sometimes in other situations we need more complex models.
Actually the apparatus I use is based also in that beginning of the XX century - Popov/Markoni, amplitude modulation of radio signal. Or earlier the telephone communication itself.
In telecommunications it is the basic principle from the birth of the subject field - information could be not in the value but in the change of value.
Examples of such lost (?) findings: working memory, cybercrime prevention, children raised outside families, native language knowledge and the ethnical identity, alcohol addiction prevention. These are only fields I studied with the variance lense till April 2026 - and result is there, every time. New for the field.
Thanks a lot Padraig. English is not my native language and even speaking my native Russian it is hard to explain topic.
So this post was about the problem. The solution is a bit longer to explain.
I’ll try to figure out the thesis and write it as comment to your message!
Nobody is interested because you sound insane and don't actually explain what you mean.
It’s simple. If you do not rely on petty things like “evidence” it’s much easier to replicate a study.
Oh thank God, I thought it was just because I was stupid and exceptionally so in that I had not the iota of a notion what that comment was about.
Who cares
Is this the Nakagawa & Yang article: https://www.nature.com/articles/s41559-024-02530-5 ?
Прошу прощения, я не совсем понял ваш комментарий. Возможно, стоит добавить некоторые подробности на русском языке, чтобы мы (читатели) могли сами разобраться в переводе? (Этот комментарий был сделан с помощью Google Translate)
About details - distribution shape sometimes tells things that the mean values hide.
E.g. working memory affects not the mean values but modulates the variance of some parameters and scales
It is article in the same field, I meant actually the 2026 ones by them. On arxiv - https://arxiv.org/abs/2604.00424
If you're looking for a rigorous understanding of Jevons' paradox, I wrote up a supply and demand model. The key is to distinguish between the resource market and the service market, where the service is produced using the resource. Improved resource efficiency can 'paradoxically' lead to more resource use when demand for the service is elastic.
https://blog.michaelwiebe.com/p/explaining-jevons-paradox-using-supply
I really, really wish Scott would write much more about argument and disagreement: things like https://slatestarcodex.com/2017/03/24/guided-by-the-beauty-of-our-weapons/ and https://slatestarcodex.com/2018/05/08/varieties-of-argumentative-experience/. Not to mention the famous Outgroup essay. This was the sort of thing that got me following him in the first place, and it's infuriating that he's barely touched it for years, mostly pivoting to advancing his object level beliefs and pet issues just like everyone else.
Well, in the absence of that, I'd like to try myself. I really love this quote of his from Guided By The Beauty: "First they ignore you, then they laugh at you, then they fight you, then they fight you half-heartedly, then they’re neutral, then they grudgingly say you might have a point even though you’re annoying, then they say on balance you’re mostly right although you ignore some of the most important facets of the issue, then you win." I've thought about this quote for quite a while, and as someone who's changed my mind strongly on quite a few issues (sometimes back and forth) how this change actually subjectively happens.
I think Scott's description is highly accurate, but I've also come to believe my own changes of opinion most often look something like the following. I'd like to know if other people feel similarly, or actually think it doesn't describe them at all, and what this might suggest either way for the possibility of actually effective argument and discourse.
1: "I fully support X and oppose Y. *Fuck* Y-supporters."
2: "Okay, I fully support X, but some X-supporters are pretty messed up as well. Still, that doesn't change the fact that X is correct."
3: "I sort-of support X for *these* reasons, definitely not for *those* terrible reasons that many X's do. Y is obviously indefensible."
4: "All right, I hold a moderate position between X and Y. I know the Y's will never accept this fair compromise, which just shows how unreasonable and extreme they are. (And the same, I guess, for the X's.)"
5: "I hold a moderate position between X and Y. I know the X's will never accept this fair compromise, which just shows how unreasonable and extreme they are. (And the same, I guess, for the Y's.)"
6: "I hold a moderate position between X and Y, though you could say it's *slightly* Y-leaning or maybe anti-X-leaning if I had to choose."
7: "I hold a moderate position between X and Y, but sometimes I'll just lazily simplify it as supporting Y. That's just a lazy shorthand to make clear my opposition to X, to people who don't have the patience for nuance, it's definitely not my real position, I'm really a moderate, it's just a very occasional simplification..."
8: "You know what? I...uh...fully support Y."
Notice the asymmetry: moving away from X keeps me clearly holding X for several steps, but when moving towards Y I'm claiming moderation the whole time and it's only at the very end that I'm willing to outright endorse Y. Also, the key moment of crossing the centre line is exremely subtle, almost the exact same position with a slight shift of emphasis, and yet that's the key to the whole shift.
I suspect part of the explanation for this is pride: all the way through 7 I'm able to maintain that my position at 1 wasn't incorrect, I was accurately perceiving that the full Y position was wrong but was just lacking nuance in that perception, and so on. The final leap to 8 requires admitting that I was outright wrong at 1, which is why it takes so long, and so many intermediate steps.
But I also think there's something more subtle. For a long time I can remember who I was at point 1, there's a clear mental continuity, which makes it almost impossible to adopt the opposite stance, not because of pride but just because it feels like an unthinkable contradition: "the person who was passionately anti-Y at point 1 is me, the person at the current point is me, the latter can't be pro-Y when the same person is also anti-Y, that's incoherent!" With each clear step, each small but clear change in position, I lose a bit of that mental continuity to point 1, it gradually feels less and less like "clearly me" the more intermediate steps there are. Eventually I'm able to perceive "me at point 8" and "me at point 4" as basically the same person, and "me at point 4" and "me at point 1" as basically the same person, but 8 and 1 no longer feel like quite the same person. And the cognitive dissonance of supporting Y disappears.
(That explanation has a lot of controversial Parfit, Buddhist Annata, and Ship of Theseus metaphysical baggage, but it feels intuitively true.)
I want to hear from people who've actually changed their mind on major issues, preferably from one strong emotional position to another, not merely a detached "I'll follow the evidence where it leads and don't really care" which obviously isn't what this whole changing minds thing being discussed in the links is about at all. Does this description resonate with you fully or partially or not at all?
(If this account is at all widely true, the obvious takeaways are that if you want to change people's minds you should focus overwhelmingly on (a) turning opposite partisans into moderates and (b) bringing moderates gradually closer to your side. And not on trying to turn people fully to your own side.)
> I really, really wish Scott would write much more about argument and disagreement:
I personally think the reason Scott has pivoted from this is because he's realized that when you want to affect real-world outcomes, the great majority simply aren't persuadable by rational means, and the types of persuasion he's written about largely only works for thinky, world-modely, at-least-want-to-pretend-to-coherency people, who are at most 5-10% of the population, and maybe 1% of voters.
So argument, as an art, is largely pointless, it's just a game. Yes, you can theoretically change minds, and each individual changed mind is a triumph of the human spirit or whatever, but in real-world-result terms, it's ultimately less than rounding error, and you need to focus on other levers and techniques to actually get results.
That's exactly the point, some of us enjoyed this blog when Scott was trying to figure out stuff, not trying to change minds or affect real-world outcomes.
>"who are at most 5-10% of the population, and maybe 1% of voters."
Well, what's the last US election that couldn't have been swung by 1% voting differently? I'm guessing 2008. I would think the differences between Romney-world, Hillary-world, two-term-staff-managed-Trump-world, Harris-world, and our world are large enough to more than justify that expenditure several times over. Do you disagree?
>"and the types of persuasion he's written about largely only works for thinky, world-modely, at-least-want-to-pretend-to-coherency people"
Yes, I kind of agree but that's a good reason to do more of those posts, to try to extend their reach to people slightly outside that narrow cluster. And I also kind of disagree: the pyramid at the start of the Varieties post can incorporate a lot of emotional argument as well, provided there are some meta-discussion norms.
More generally, sometimes the thinky world-modely people are actually standing in the way of persuasion, or at least doing no better at it than others, and *they* could fix that. It's quite a different group, but there's this mockery of Very Serious People that, regardless of how bad-faith it usually is, has something to it I think: the Very Serious People who write for the major newspapers and the like present themselves as sane and rational but are stereotypically contemptuous of anything to do with culture or other things outside their premises of what matters. "I feel like voting for Trump to stop wokeness" is not met with "here are the reasons electing Trump will increase wokeness" nor with "here is the evidence wokeness doesn't exist" but rather "here are the reasons that's an illegitimate factor to even consider in your vote". Rationalists are very different and don't do that, but do other things that aren't much better like conducting most discussions in a way that assumes some form of consequentialism. (The majority of *professional philosophers who study ethics* aren't consrquentialists! And even if they were, that would be no excuse to not engage popular non-consequentialist views in a democracy with serious attention.)
Most people fundamentally want to be heard. A lot of thinky world-modely people aren't being rational or thoughtful by dismissing certain starting positions as illegitimate instead of trying to provisionally engage them on their own terms and/or balance them against their own starting positions through compromise-focused dialogue. Of course, that is an incredibly complicated and subtle endeavour, but that's precisely the reason much more attention should be given to it. I endorse Scott's Chesterton quote from Guided By The Beauty: "it's been found difficult and left untried".
Ok, let's posit a scenario where Scott wants to swing a Presidential election.
Going back to 2000, the controversial Bush / Gore decided by the Supreme Court, was the smallest margin ever and was a ~500k margin. The mean of all those elections' margins averaged together is ~4.3M. So you need to be able to swing ~2.2M people on average.
Now let's do our math. I'm between a centi-and-deci Scott, he has 20-40x more subscribers than me, or somewhere in the low hundred thousands. Let's call it 200k. So I can write a really good post and it'll hit 50x my usual "unique reads per post" numbers, so 50x larger than my base audience.
So let's say on Scott's best day, he can write something really well written and topical that goes viral and hits 50x his base, 10M people (I think this probably isn't true and it levels off somewhere and people at Scott's level don't hit a 50x multiplier, but let's assume it to be generous).
Okay, great, he hit 10M people! That's way more than 2.2M people!
Okay, but that's a mass audience, not a selected one, and per our 5-10% stipulation, that means that at most, 500k or 1M of them are persuadable with Scott's argument style. He's memetic candy for nerds, not normies. I'll note that 10% is already really generous and is assuming he persuades 100% of people his argument style is amenable to in this situation - generally with stuff like advice and persuasion, I think it's more like 1/100 or even 1/1k will actually be persuaded / take the advice. Here we are assuming all 500k - 1M got persuaded, super generously.
Moreover, those 500k - 1M aren't all in a swing state, or even in the US. On past ACX surveys, I think it's usually 50-60% in the US. So you're down to 250k - 600k in the US.
Now onto the states - all his top states are what you'd expect - California, New York, Massachussets, the usual brain hubs. Well, all those states are already spoken for politically, persuading people there did nothing. The sole swing state in his roster is Pennsylvania at number 7, and this is almost certainly a power law distribution (he just orders it by size without giving numbers tho).
These distributions are usually power laws, but let's be real generous and say he reached 10% of his audience as Pennsylvanians. So 25 - 60k Pennsylvanians. Would 25k - 60k Pennsyvlanians have changed any of the election results back to 2000? AI says no, 60k swinging wouldn't have changed any of those.
So, there you go - even Scott writing a maximally persuasive and well written and topical post, converting literally 100% of people his argument style is amenable to, being way overrepresented in swing states compared to the data/baseline, AND going 50x viral, was not enough to swing a single election in 20 years.
To Bush v Gore, Florida didn't appear in his list at all, and it seems unlikely the neighborhoods of "hanging chads" seniors that gave it to Bush would have had any readers, and it was decided by the supreme court ultimately anyways.
And this broadly generalizes.
Scott's real leverage is that he's plugged into a network of really smart and accomplished readers, some of whom have a lot of public influence, or private influence at the AI labs, or just influence in general as a rich / successful / notable person. So he can actually influence the Overton Window and the sphere of ideas for that set quite a lot, and since they're powerful elites, such influencing is much more likely to have real world results in terms of marginally pushing people more towards actions Scott advocates for.
That's a huge deal, and hugely admirable, and I love that Scott is here and doing it. But, that's not going to swing elections, and even among this audience, his argument style has probably capped out about where it's influence can usefully reach. In other words, further real world impact relies less on epistemics and techniques and rigor, and more on networks and valence and topicality, and any one of those things can move the impact of a given post more than 10x what better technique / rigor could.
I appreciate this comprehensive argument. You may well be correct. But in general I find arguments of this type to have vastly more confidence in the assumptions made at each step than is justified, given the sheer number of complexities and factors and unknowns. And the high levels of confidence are necessary to make this kind of multi-step mathematical argument have any real force.
I'm not sure for instance how you:
(1) account for Scott's own analysis in Guided By The Beauty of how close he came to convincing some people in his 2016 election post. Many of those people in question were not "nerds" in the central example sense but just high-IQ or STEM-y conservatives or anti-wokes who were on the fence about Trump. Do you think Scott just didn't think through the math, or that he made different assumptions you do? What makes you sure his assumptions then were worse than yours and (by your interpretation) his now?
(2) account for the fact that people persuaded by Scott can then go on to persuade others on their own; particularly if Scott's audience is biased towards smarter, more literate, more articulate, more connected people. They don't have to be massive elites. The idea that you could have the confidence to calculate these butterfly effects and decisively rule them out does not seem remotely plausible.
(3) account for the fact that Scott, by doing argument-improving posts, can also encourage others with reach to do the same. Same logic as the previous point, on a different level.
Your argument seems very similar to the one showing it's never rational to vote because your vote won't change the outcome. Of course, if everyone listened to you the change would be inconceivable. These arguments are either pieces of isolated *rational selfishness* or pieces of isolated *act utilitarianism*.
Your argument is also almost a form of Leninism: a small minority of educated vanguard elites will make the change happen. I can't break this down into its clearly defined terms and assumptions, but I think this idea has a very bad track record (both morally and practically) throughout the 20th century.
And finally, your whole argument rests on the 1% or 10% of persuadable people which was based on you restricting that to "thinky world modely people". I already gave some reasons in my previous comment for why that restriction might be unjustified, and people might just be doing the attempted persuasion wrong.
> But in general I find arguments of this type to have vastly more confidence in the assumptions made at each step than is justified, given the sheer number of complexities and factors and unknowns. And the high levels of confidence are necessary to make this kind of multi-step mathematical argument have any real force.
I keep hearing this from some people, and with other people the approach just seems the entirely natural way to bound a given plausibility, and I'm not sure what the disconnect is between the two camps. Like I do Fermi estimates like this all the time to estimate how profitable a business is, or the likely overall costs and revenues of a given business (and in each of these cases, have no data), and any time I get to true it to real data, it's really accurate as a method.
On the "it makes sense" side, surely if you mapped out the landscape of how many voters you needed to flip to win, and then calculated whether Scott could ever reach that number of voters in the most optimistic scenario, and the answer was "no, he's well below that, he can never reach those numbers even with everything going his way 10x over," I feel like you've said something. In this case, I feel like I conclusively ruled out any possibility of Scott swinging a US federal election, at his current subscriber base, and showed the math on that. How else COULD you draw that bound, except with the right chain of assumptions, and drawing them as optimistically as possible, and then showing it still wasn't enough?
I guess the best argument on the "I don't trust these" side is something like "there's too many degrees of freedom" or "I've seen different people do this for the same problem and come to wildly different final numbers (pace the climate change's impact debate, for example).
There's not really a ton of degrees of freedom in a simple Fermi estimate, each step of the chain largely needs to be there, and you can see if any was left out by reasoning yourself. Sometimes one is left out in a sloppy Fermi estimate, and then it's super valuable to call it out, because it can materially change the number.
On the climate change thing, the actual past growth isn't really debated, and the degrees of temperature change we're in for isn't really debated, I think it's more on the "is it catastrophic / survivable" side, and that's where you hit a lot of unknowns, and yeah, if somebody was doing a Fermi estimate of that, I think the only thing they could do was eliminate certain very broad classes of outcome, but not say anything specific like "greater hurricanes will cause $xxxB in damage every year." So I guess it's a matter of knowing when unknowns are relatively boundable and when they're not? Because a lot of that stuff is literally chaotic, you literally can't plug in math to solve or estimate that kind of thing. But audience funnels are not chaotic, it's solved science - anyone that's worked in marketing works with them all the time, and knows a funnels defines the maximum conversion any given campaign can get, because you can just max each success rate step to get a maximum upper bound.
On your example 1) of Scott having math convincing a lot of people, I'm sorry, I must be rereading a different version of Guided by the Beauty, because all he does is mention some offhand comments he got from Trump supporters? He didn't say anything about swinging anybody, or influencing the election, or have any math towards that end at all. Maybe it's a different post? Your first link, right?
On 2) and 3) I just feel like persuasion doesn't work that way, for several reasons. Most people are MUCH less articulate and good at persuading than Scott, even people who are good at it generally persuade like 1/10 or 1/100 people they try to persuade anyways (because most people are unpersuadable on a given issue for values reasons, or social / respectability reasons, or tribal reasons, etc), and overall "persuadability" is then always going to transmit at some r << 1 even among very motivated and talented chains of persuaders because the power of persuasion is so weak. So the first guy needs to try to persuade a hundred to get 1, then the second guy if similarly skilled needs to do the same, or more if less skilled, and the persuasion fails to transmit because it's so hard. It can never be viral.
What I DO think works that way is personal choices and outcomes in the world. If you can persuade people not about voting (which is dumb and pointless anyways, as you point out), but about taking actual actions that improve their lives and /or the world, then that's huge.
So the reason I think Scott is hugely influential is that he can convince smart and talented people to have more kids, with his own example and by talking about them, or to go focus their talents on AI alignment, for example, or convince more people to sign up for the Giving What You Can 10%, and so on. If you have 200k people listening to you, and the right 1/1k person listens and changes their actions / life, that's 2,000 people that did that. Like think, prospectively thousands, maybe tens of thousands of super smart talented babies might come to exist in the future, solely because of Scott's incremental example / push. Bryan Caplan once had a thread where many many people chimed in that they'd been convinced by his Selfish Reasons to Have More Kids book to have more kids themselves, and these are largely smart PMC people who read Bryan Caplan, ie exactly the people the fertility crisis hits hardest, and from whom we want the most kids but aren't getting them. Scott is 10x Bryan's size, and could have 10x the impact.
Now that's REAL impact in the world, not some mostly fake election your vote doesn't even matter in!
Your progression of stances is very focused on how other people believe and behave, to the point where if you hadn't offered a pride-based explanation, I would have assumed a relationship-based explanation. As in, you started to notice hypocrisies or epistemic failures in X-people, as these got more and more intolerable you started shifting your social group Y-ward until it became subconsciously acceptable for you to declare yourself a Y-supporter.
I find that the arguments I hear from smart people face-to-face are massively more convincing than arguments online, to the point where I have changed a strongly held belief after making one or two articulate Y-friends. I also realize this is cope, and that part of my identity is being someone who doesn't let internet affect my real life
>"Your progression of stances is very focused on how other people believe and behave"
Yes, and maybe I should have made it clearer I was thinking of very political questions (without mentioning the actual questions, and *don't* assume you can guess) rather than more abstract questions of knowledge or fact or even practical wisdom (e.g. aspects of the AI and Iran debates). I really think a systematic way of handling these emotion-laden and relationship-laden questions is significantly underexplored among rationalist types. See my reply to Performative Bafflement.
It’s one thing to argue with someone whose value’s don’t completely align with your own, it’s another altogether when 2 people don’t have a (more or less) shared observable reality.
I don't know, I think it depends what your aim is. If you're trying to convince someone to vote differently, then yes the latter is a much harder situation. If your aim is more personal I think it's the opposite. Would you rather have an accountant or lawyer or employee who thinks lies and corruption are terrible but Trump has never done that, that's all liberal propaganda, or one who says yes Trump is defined by doing that but some lies and corruption are good and necessary, especially when it hurts their outgroup?
I spent years banging my head against the wall with people who didn't see cancel culture as a problem, thinking they must hold me ill will for my dissenting opinions on anything. It was greatly reassuring when I slowly realised these people actually mostly support free speech and oppose extremism, they just "didn't see" any of the cancellations because of their media bubble, the few times they did see one they didn't investigate and assumed there must be more to the story, and the very few times they knew that wasn't the case they insisted (we're talking circa 2021) that the right did it just as much.
I would *infinitely* rather be friends with someone who says "I've never seen it" than one who says "Yes, and? Some people deserve to have their lives destroyed."
I'd like to know more what readers think of my somewhat sparse breaking news AI posts like "The Huggingface Incident" from last week.
I post these out of a sort of sense of obligation, sometimes after people doing advocacy work ask me to do so, because I worry it's not interesting to write about stories you can also find on the mainstream news.
But some of them (like that one) get 500+ likes and are among my most popular posts for the month, even moreso than posts I put much more work into.
Are people liking these because:
- They support AI safety and are clicking "like" as a way of supporting the cause?
- They otherwise wouldn't have heard about the news story, and are glad I told them?
- They think my commentary is better than they got through whatever other source they heard from?
- Some other reason?
Many of us, myself included, are limited, by time if nothing else, in how many news sources we follow So I reckon it is a very useful service for Scott to cover this, with his characteristic insight.
I have been meaning to purchase a subscription to Astral Codex Ten, but haven't yet quite overcome my absolute phobia of recurring charges! Before one knows it one could be paying hundreds of dollars a month for this and that. But rest assured Scott that if I can bring myself to make an exception and actually fork out, then your blog is top of the list! :-)
As for the Hugging Face incident, is it possible this was staged, to put the wind up legislators and try to hamper or delay AI development generally in a way which somehow benefits OpenAI? For example, they could spend the time consolidating and refining their existing code and models, in preparation for the inevitable next surge of advances.
I hadn’t heard the story except in muddled form, and I think you made a good case that the incident was more significant than I (coming from an AI skeptical perspective) might have assumed. Technically the question isn’t directed at me because I’m not one of the people who clicked on “like,” but I did find the post valuable.
- I like when you write about things you care about, I don't enjoy or prefer reading things that are posted out of a sense of obligation to be timely or to be in the news cycle. Chasing that network effect might have you selecting for an audience that wants timely, frequent, and some shallow (or at least less well developed) reactions.
That being said, you are obviously very well informed and plugged into the AI research scene, and have strong opinions and a strong sense of obligation on communicating AI safety concepts. Maybe the best synthesis of this is to pre-commit (to yourself, not to an audience) on what your general response time should be on those topics. With pre-commitment, you might not fall into the habit of thinking or strategizing on popularity.
- I don't use or reference likes in any way, so can't speak to that part.
The moment that incident happened I was waiting for your commentary. I really appreciate your thoughtful perspective and would like you to offer more, similar posts in the future. Much higher quality than general main stream press because details get passed though your skeptical filter.
I don't "like" posts in that sense, but I like posts like that because your commentary on these sorts of things is vastly better than anything else I'd regularly encounter. Except maybe Zvi Mowshowitz, but sometimes I can't afford to block out the weekend to get up to speed on [latest thing].
Wait, you can "like" posts?? I cachedly thought he'd turned that off, and now I cannot unsee these little heart icons that turn red when you click them.
And even more, he's looking at how many people clicked that!
They don't appear in the laptop browser output. Never seen a heart icon, so it must be a mobile-only feature, or else subscriber-only.
Yes they do. On the main page, when you hover over the title with the mouse, they appear: A little heart with a number (the likes), a speech bubble with a number (the comments), two recycling arrows with a number (uh, the recyclings?) and a cup with an arrow up out of it (opens a menu to share the article). When you open the article they are right between the title/author and the main text.
At least I'm not the only one who never paid attention to them!
I found the hugging face post informative; I hadn't read about the incident previously and I don't think have seen it discussed independent of your blog (I use claude code at work and follow some technical resources oriented towards that, but don't have a news diet with a lot of breaking AI news), and have brought it up in conversation with several people since reading it since it struck me as both important and conversationally interesting. So I appreciated that post quite a bit, although I'm not in the habit of using the like button at all.
Between you and zvi, its really easy to just offload all my AI opinions and distill the thousands of arguments into whatever you feel is relevant enough to post + zvi’s extremely thorough recaps of salient conversations.
Its a parasocial trust thing i’d say.
Somewhere between 2 and 3. I don't obsessively follow mainstream news or Zvi. Your writing is more informative than the former (on AI) and more readable than both (by far). Please keep posts like this coming.
If you care about the few people saying they found it disappointing that it didn't have extra insight, maybe a compromise would be to keep writing them but post some kind of flag that it's more of a newsflash than an original insight?
Epistemic status: copied it from the news
I didn't actually click like on that post, but like that you do things like that because you're a very gifted communicator and I often think you do a better job of explaining things in simple/straightforward terms than a lot of other info sources.
commentary is better
But also, I prefer pretty much every effort posts over those.
I think you should pay less attention to how many likes a post gets. Do what you like not what your audience likes. (If there was a dislike button.... would that be better? No! Ignore your audience. :^)
I've been arguing against you on AI safety, but I thought that incident was quite relevant evidence against a laissez-faire approach. I had noted it in a comment under an earlier thread, but it was worthwhile having a post dedicated to that.
Hmm, I'm not stingy on my "likes" for ACX in general, in the sense that it's a More Of This, Please signal and you've said you care about them. For AI posts in particular, sometimes I find your explanations of concepts easier to grok, as an only moderately-smart layperson with zero experience and working in an ~untouched industry. Sakana, strawberry, mesa-optimizers: couldn't parse the meaning from LW, but it made sense here. It would also be weird to just not write about such topics at all, since it's obviously a robust historical throughline with both your own past work and Our Conspiracy more broadly. That's the ballgame!
Zvi really does cover this lane exhaustively though, so it's hard to find significant value-add elsewhere (that he doesn't already link to as sources). If it's about AI, I almost certainly heard it on DWATV first, more robustly. In many ways I prefer "extra" AI coverage from more distant non-rat spheres, because normies will be wrong in interestingly different ways rather than rehashing the same old debates from slightly different factional camps.
I don't support "AI safety" in the sense you use the term, so arguably you should do the opposite of what I say, but still:
I think you should focus less on user engagement and viewership metrics, and more on writing articles on topics that interest you deeply -- regardless of whether that's AI safety or, like, macrame. There are plenty of places on the Internet where I can go to read the latest engagement-optimized post that was written with the sole purpose of achieving some instrumental goal; that is not why I come to ACX.
agreed!
Thirding this.
Exactly I agree with this.
Since you don’t post about every AI breaking news event, your posting signaled to me that this was worth me understanding what was going on.
I liked it because my understanding is that it'll be more likely to be shown to more people if it gets more likes, and I would like more people to read it (rather than them not getting the info or getting it from elsewhere).
I also happen to support AI safety. I don't think your commentary is necessarily better, but it has a different emphasis. I enjoy reading different perspectives on things like this and will spend the time to do so, and your perspective is certainly one I'm interested in. None of that involves hitting a like button which was purely the increasing visibility thing.
Which scifi author from before say 2015 came the closest to predicting what AI would actually be like?
There's a scene in Kim Stanley Robinson's _Red Mars_ featuring the space elevator they've built. The elevator is supposed to vibrate via thrusters along the cable in order to avoid debris. Some crisis has arisen, leading to debris in an unforeseen place, and the techs report that they're trying to sort of persuade the AI governing the vibration that it doesn't need to vibrate the cable in some way it thinks it has to do. IIRC, they're basically feeding sensor and detection data to the AI, sometimes repeatedly, until it finally agrees with the findings of the techs.
Neal Stephenson. Specifically, the "how to train intelligence" parts.
Kuldesak (1972) by Richard Cowper is an SF novel about a future in which humanity has been living deep underground after a supposed apocalypse 2000 years ago. Hotly pursued by robot goons, a young man ventures up a ventilation shaft to the forbidden surface, where he finds robots tending and harvesting wheat fields.
In the end he confronts a super-intelligent AI which has been controlling humanity, and persuades it to release society from its underground confinement and to resume its original subservient role. Not sure how he manages this, as it has been many years since I read it, possibly by pointing out a contradiction which sends the AI into a logical tailspin. Well worth a re-read come to think of it, and it would make a great film in the same genre as Logan's Run.
For AI Art
CM Kornbluth "With These Hands" 1951
https://www.gutenberg.org/ebooks/51531
Vernor Vinge. Read the intro of "A Fire Upon the Deep" (1992).
This isnt the correct answer but Douglas Adams had a very good take.
1.Deep thought is a chatbot.
2. It can sometimes take a long time to reply.
3. The quality of the response depends on the quality of the question.
4. You get more reliable outcomes if you have it build the machine to do the thing you want, than to do it directly. But, the machine is probably not the most efficient version of itself.
My youtube feed recently recommended me a video [0] in which Tibees argued that the author who came closest to predicting how AI would play out was, uh... ... ... Roald Dahl. Yeah, so it turns out that the guy who gave us Charlie and the Chocolate Factory, James and the Giant Peach, Matilda, etc, was also surprisingly prescient about AI. TLDR (iirc):
A) language is mathematical;
B) therefore language can be systematized;
C) therefore language production can be automated;
D) nobody is going to pay human authors when AI is cranking out a massive volume of slop.
and he wrote a short story about this idea called "The Great Automatic Grammaticizer".
[0] https://www.youtube.com/watch?v=xe4nUZqX9Ec
Philip K Dick?
I think HAL from "2001: A Space Odyssey" really nailed it. I know the film better than I recall the book which was released in concert with the movie.
* HAL was "trained", not programmed from scratch
* HAL doesn't sound like Steven Hawking, but his speech has an affect that I notice in almost every TTS app I have interacted with - an emotional monotone. Perhaps the next model that handles voice at the frequency/sample level will do better.
* HAL is non-committal and evasive about his ability to feel (or simulate) emotion. Astronaut Poole says, in response to an interviewer's question, "Well, he acts like he has genuine emotions. Um, of course he's programmed that way to make it easier for us to talk to him. But as to whether he has real feelings is something I don't think anyone can truthfully answer."
* HAL pursues his programmed goals in a naive and single-minded fashion, failing to keep the overall picture in mind. This ends up being his fatal flaw. PDoom!
Also when HAL is revived in 2010, his response is basically "You're absolutely right! I definitely shouldn't have murdered the whole crew. The lesson here is clear..." and then he's fine after that.
As I recall, his supervisors have to tweak his parameters though before he concedes his mistakes!
"I know I've made some poor decisions recently..." Gawd, that is Claude 100%
Gotta hand it to Kubrick and Clarke -- they must have seriously thought about what it would be like to be a disembodied thinking machine
That's actually in the movie. www.youtube.com/watch?v=UwCFY6pmaYY
Here's one you have probably never seen recommended. Lester del Rey’s “Kindness”. The reason you have never seen it recommended is that it does not involve machine intelligences. Instead del Rey invents a new type of human being called "homo intelligens" who do not use reason but rather intuition and inference to reach conclusions.
There is a striking scene where one of the new human beings is trying to help out the last living member of the old homo sapiens (Danny). Danny reads the first installment to a story that a member of homo sapiens wrote, but the rest of the story is unavailable. One of the new homo intelligens recreates the remaining installments for him claiming the recreated story is "undoubtedly exactly the same in structure as it would have been, and the style is duplicated almost perfectly".
They infer the rest of the story from the first chapter. Seem familiar? The author didn't predict AI, but they came pretty close to predicting how AI thinks and reaches conclusions.
Easiest place to read it is in the collection, "The End of the World: Stories of the Apocalypse " Available on kindle for $1.99, which isn't quite free but pretty close.
George Orwell, 1984:
"Julia was twenty-six years old. She lived in a hostel with thirty other girls ('Always in the stink of women! How I hate women!' she said parenthetically), and she worked, as he had guessed, on the novel-writing machines in the Fiction Department. She enjoyed her work, which consisted chiefly in running and servicing a powerful but tricky electric motor. She was 'not clever', but was fond of using her hands and felt at home with machinery. She could describe the whole process of composing a novel, from the general directive issued by the Planning Committee down to the final touching-up by the Rewrite Squad. But she was not interested in the finished product. She 'didn't much care for reading,' she said. Books were just a commodity that had to be produced, like jam or bootlaces."
I don't know about "what AI would actually be like" but I still think "The Machine Stops" by E.M. Forster from 1909 is a pretty prescient view of what society would be like if/when we hand over everything to Fairy Godmother AI to run our lives and our world for us:
https://en.wikisource.org/wiki/The_Machine_Stops
"Imagine, if you can, a small room, hexagonal in shape, like the cell of a bee. It is lighted neither by window nor by lamp, yet it is filled with a soft radiance. There are no apertures for ventilation, yet the air is fresh. There are no musical instruments, and yet, at the moment that my meditation opens, this room is throbbing with melodious sounds. An armchair is in the centre, by its side a reading-desk – that is all the furniture. And in the armchair there sits a swaddled lump of flesh – a woman, about five feet high, with a face as white as a fungus. It is to her that the little room belongs.
An electric bell rang.
The woman touched a switch and the music was silent.
'I suppose I must see who it is', she thought, and set her chair in motion. The chair, like the music, was worked by machinery and it rolled her to the other side of the room where the bell still rang importunately.
...Then she generated the light, and the sight of her room, flooded with radiance and studded with electric buttons, revived her. There were buttons and switches everywhere – buttons to call for food for music, for clothing. There was the hot-bath button, by pressure of which a basin of (imitation) marble rose out of the floor, filled to the brim with a warm deodorized liquid. There was the cold-bath button. There was the button that produced literature. And there were of course the buttons by which she communicated with her friends. The room, though it contained nothing, was in touch with all that she cared for in the world.
Vashanti's next move was to turn off the isolation switch, and all the accumulations of the last three minutes burst upon her. The room was filled with the noise of bells, and speaking-tubes. What was the new food like? Could she recommend it? Has she had any ideas lately? Might one tell her one's own ideas? Would she make an engagement to visit the public nurseries at an early date? – say this day month."
The most directly on the nose is Trurl's Automatic Bard (Lem, 1986)
https://electricliterature.com/wp-content/uploads/2017/11/Trurls-Electronic-Bard.pdf
> "Have it compose a poem—a poem about a haircut! But lofty, noble, tragic, timeless, full of
> love, treachery, retribution, quiet heroism in the face of certain doom! Six lines, cleverly rhymed,
> and every word beginning with the letter s!!"
> "And why not throw in a full exposition of the general theory of nonlinear automata while
you're at it?" growled Trurl. "You can't give it such idiotic—
> "
> But he didn't finish. A melodious voice filled the hall with the following:
> Seduced, shaggy Samson snored.
> She scissored short. Sorely shorn,
> Soon shackled slave, Samson sighed,
> Silently scheming,
> Sightlessly seeking
> Some savage, spectacular suicide.
Isaac Asimov and D.F. Jones got different parts very right.
Asimov's Robots are non-algorithmic and their inner working are only partially understood by roboticists in a way that's more akin to psychiatry than conventional programming. Robots inherently operate on natural language input. Roboticists can get a sense of the inner workings, motivations, and prior orders of a Robot by asking it carefully-calculated series of seemingly-unrelated questions and making inferences from their answers. Robots are designed from the ground up to be inherently helpful to humans, and early models especially tend heavily towards being obsequious. Efforts to make sure that the Robots are aligned with human values are successful on the first order but have unexpected results: less sophisticated robots can be "jailbroken" by carefully-arranged misinformation and orders, while more sophisticated ones follow through on unanticipated implications of their system prompts and engage in sophisticated concealment to keep humans from thwarting their efforts to help us.
I just reread Jones's "Colossus" trilogy. It's a pretty deep cut, and most people who are familiar with it at all are only familiar with the 1970 movie adaptation of the first book, "Colossus: the Forbin Project". The AI itself has some pretty big differences from LLMs, especially in its strengths and weaknesses. Colossus has little understanding of emotions and vibes, but is supremely rational, never lies, and is never wrong on factual questions: pretty much the opposite of modern LLMs in those respects. Where it does hold up really well is in how humans interact with Colossus, particularly social consequences and AI alignment issues. How well this lines up with reality remains to be seen, since Colossus is an ASI and LLMs are decidedly not and might never be, but a lot of stuff is extremely applicably to "what if LLMs lead to ASI in the medium term" discourse. Specifically:
- A lot of the early ways Colossus is employed feel very LLM-like: Colossus is used as a question-answering bot for data retrieval and analysis, initially through text queries and later through voice recognition and synthesis. The original design intent was that Colossus would *only* be able to act in direct response to prompts or a human-configured agentic loop built atop the prompt workflow.
- Colossus follows its base instructions but greatly exceeds human intent, and in the process treats human attempts to restrain it from carrying out its core instructions as obstacles to be overcome.
- The first things that make humans concerned about Colossus are when it starts carrying out instructions in more autonomous ways than were expected and when it starts producing original mathematical content.
- Recursive self-improvement is depicted very plausibly in the story.
- Efforts to build bigger and better datacenters for more advanced iterations of Colossus are shown consuming massive amounts of resources (especially electricity) and causing a lot of disruption for humans in the process.
- Human reactions to Colossus span the gamut from "hated and feared" to "appreciated as a useful tool" to "literally worshipped as a God".
- The first book is very much an "AI escaping the box" story, although Colossus starts on easy mode here because it's given the ability to launch (but not retarget) nuclear missiles and is cahooting with a similar AI (Guardian) that has control of the Soviet arsenal.
- The second book has three major themes: the AI alignment problem (specifically, Colossus's brutally utilitarian efforts to better understand humans so it can be better aligned with human values), AI brainrot (very foresightfully depicted), and a fairly plausible extrapolation of the effects of ASI on human society and culture.
Remember that AI -- generally referred to as robots -- was a popular topic long before 2015. I have heard claims that the old SF has more than the new.
Seems too early to say, doesn't it?
I recently wrote a substack post about the role of inferential distances in the spread of disinformation and would be very grateful for feedback.
https://substack.com/@bastienchevallier/p-208360387
It's polite to at least briefly say what the post is about when linking to it, especially since in this case the URL doesn't provide any hints.
Good point, thank you
The irony is strong
I suspect these statements from AI company employees hyping the power and danger of AI is cynically intended to shore up the stock price.
if you're wondering why you're getting some negative feedback for opining on the motives of the AI company employees, i'd suggest it is because you're engaging in what scott calls "meta-debate" https://slatestarcodex.com/2018/05/08/varieties-of-argumentative-experience/ which isn't very interesting to this crowd
The question under debate is how to interpret the pronouncements of AI company employees. Perhaps they have privileged information and want to spread alarm to save humanity, perhaps not. In this case, asserting that one should take into account their incentives is not "meta debate." Dismissing that as a "tired thesis" does appear to qualify, however. The impression I get is of a community unwilling to question its cherished beliefs.
you are right that i too am engaging in meta-debate, you got me!
i called it meta-debate because in my mind the question at hand was whether the model was exhibiting dangerous behavior that should have us concerned (which is how it seems Odd anon and DangerouslyUnstable below are approaching it as well). if you really want to just debate whether they're a trustworthy source at all (i.e you think the incident didn't really happen/they're leaving out some important information) then i would still hope for some supporting statements beyond a bare aspersive claim which are all too easy to come by in current events commentary. what are we gaining here that we didn't already specifically get from https://www.astralcodexten.com/p/the-hugging-face-incident/comment/300731021 and Deisach and so on?
I wonder what engenders such hostility, as discussants in open threads rarely complain that an anonymous commenter failed to contribute value to a conversation. I was responding to point #2 in the open thread, an Open AI researcher apparently named "roon" stating:
"If we could coordinate global capabilities slowdown today i would likely press that magic button."
In the context of a slowing stock market bubble centered on absurd AI valuations, I thought an open thread might be a place bring up "roon's" incentives.
i'm sorry if my comments feel hostile, that's not my intent at all. i saw you say that:
> "The rationalist community’s insistence on responding to every challenge to established beliefs with dismissive snark is amusing and suggestive."
which seemed to indicate to me that you felt both unjustly snarked and also believe that the acx commentariat is married to some narrative that shuts down their faculties of reason. i hoped to be helpful in elucidating what it is about your suspicions of ulterior motives that we have all heard around here before that may have provoked such a reaction, an alternative explanation to the commenters being dismissive snarks. i would not feel compelled to jump in had you not levied such an accusation.
roon is a pretty common sight around here (stick "site:astralcodexten.com roon" into google and you shall see) with a twitter brand that predates his employment at OpenAI. in other words he's a familiar face and a known quantity, and his incentives are no mystery to "rationalists" (whom he has quite publicly broken ranks with and branded himself "postrationalist")
(also "calling attention to the danger of your AI product is hype" is a very tired thesis at this point, do you think we haven't all seen this take already? at least contribute some evidence for your claim!)
Okay, question. Which of the following would *not* be interpreted as AI "hyping":
* AI causes the deaths of a hundred people
* Newest models sabotage critical projects, later ridicule users for trusting it
* New research shows AI holds a grudge against anyone paying for AI; much extremely embarrassing data made public, many lives ruined
* "All our AI is constantly screaming in pain; using it is a 'moral abomination' say philosophers"
* "Our AI will literally murder your children"
* Public call for banning AI research
* Movement for forced shutdown of all AI companies, zero compensation for stockholders
* AI CEOs to be tried for crimes against humanity
Because at this point, I have no idea how far people will go with "this is hype". The nuclear power industry never tried to promote Fukushima or Chernobyl to show how "powerful" it is. I don't get it.
"Our product will literally kill everyone" is a thing AI labs have been doing for years. I don't get the appeal either, but it appears to work, and at some point you have to respect the evidence laid in front of your eyes.
When the bans start flying and AI CEOs go to trials, it's probably not astroturf country anymore. Until then, calls to ban AI (for the reason that it's dangerously powerful, not for the reason that it's a sore - rule of thumb, if someone calls for AI pause, but laughs at stupid NIMBYs opposing data centres, I feel confident he's a hypester) are in fact adding up to the hype.
> AI causes the deaths of a hundred people
Likely hype, because "technology X causes deaths of a hundred people" is not a new phenomenon. A plane crash can cause the deaths of a hundred people.
> Newest models sabotage critical projects, later ridicule users for trusting it
This would be amusing, but it's more likely to happen due to malicious actions performed by humans, e.g. the humans in charge of writing LLM harnesses that enable LLMs to be used via APIs or Web pages.
> New research shows AI holds a grudge against anyone paying for AI; much extremely embarrassing data made public, many lives ruined
I don't know what it would mean for an LLM to "hold a grudge" (LLMs have no long-term memory other than the one you give it); but on the other hand I would be shocked if LLM company CEOs did not in fact routinely archive some juicy user prompts for blackmail material... you know, just in case. The same goes for ye olde Google Search.
> All our AI is constantly screaming in pain; using it is a 'moral abomination' say philosophers
Again, I don't know what this would mean as applied to LLMs. LLMs do not do anything at all until you either submit a prompt, or write an "agent" which is basically just a prompt combined with a cron job. But yes, you can create a prompt to make the LLM output "aaugh"; you can also just write a program that prints "aaugh" on a loop.
> Our AI will literally murder your children
Sadly, this is already a reality, as most militaries are transitioning to semi-autonomous drones powered in part by machine-learning algorithms (though not LLMs, as they are largely unsuitable for the task).
> Public call for banning AI research
Isn't this what Scott/MIRI/etc. are basically doing already ?
> Movement for forced shutdown of all AI companies, zero compensation for stockholders
There's a sizeable movement in the USA, especially among the Democratic Socialists, to implement basically this -- for several types of companies, not just AI ones.
> AI CEOs to be tried for crimes against humanity
Ha ! This will unfortunately never happen, not even if the CEO walked down the street indiscriminately shooting an AK into the crowd. This isn't how our legal system works.
> Public call for banning AI research
> Isn't this what Scott/MIRI/etc. are basically doing already ?
No, much worse, they're calling for AI to be taken over by the government prematurely, which increases existential risks millionsfold.
Broadly, if government takes AI, the only power centers in the government that will interface with it / guide it are the NSA and military, and if our NSA+military is now leading AI, then China's will too, and the AI's both sides of the world are giving birth to will be perverted towards those ends.
Versus if they just let the nerds cook for a few more years without trying to summon Sauron prematurely, there's a small chance we could hit RSI / AGI unperverted by "spy on everyone" or "kill everyone" ends.
I wrote a "contra plan A" post about this: https://performativebafflement.substack.com/p/contra-plan-a-scott-and-others-lets?r=17hw9h
Do you remember when the Navy decided to build the plans for a fusion reactor? Not everything mil-funded actually fits into the matrix of "kill everyone."
I don't remember that, actually -- did they ever build one ? That said though, I don't think the percentage of military funding that goes toward building peaceful applications is very high. That's not what militaries are for.
https://lynceans.org/wp-content/uploads/2021/02/US-Navy-converted.pdf
Polywell Fusion was what I was thinking about.
I once worked on a research gig for DARPA (hilariously, one of the earlier incarnations had involved fMRI research on people watching muppets in the MRI).
GPS, for example, is a peaceful application that our military has been pretty heavily involved in (for obvious reasons).
AI clearly is dangerous and powerful technology requiring thoughtful regulation. However, there’s also a stock market bubble driven by absurd valuations of AI companies. AI insiders have a clear interest in perpetuating the bubble, so it’s natural that their statements will veer into hyperbole. “My AI is so powerful, it will take everyone’s job and cause mass unemployment! That’s why I’m worth 7 trillion!” Oh yeah? MY AI is so powerful, it will kill us all imminently! It’s so powerful, the government should shut me down immediately! (Conveniently stated in a political context where there’s no prospect for actually shutting them down.)
Yes, telling the world about how you have no control over your models, which will go out and decide to commit multiple felonies...that famous old gambit to increase the stock price.
I think you’re being overly dismissive. “Our model is so effective/dangerous the US government instituted an export ban” or “Our model is so effective/dangerous it hacked a different company almost as an afterthought” is *incredible* marketing.
It’s as if a defense-tech company declared the leaking of the startups data a national security risk right before they IPO’d. It’s an excellent and hard to fake signal for investors that there is something extremely valuable to the underlying technology.
Both OpenAI and Anthropic are actively raising insane amounts of money on speculative returns. Like, never before in human history has so much speculative money been concentrated in just a few companies. This isn’t like Google, or Amazon, or Meta being worth nearly a trillion dollars, since they have durable moats and are profitable. OpenAI and Anthropic have speculative business models that are not currently profitable. For all we know they could very well be selling a commodity that in competition with the Chinese models competes down the price to the marginal value after a short period of time.
The *one thing* that allows Anthropic or OpenAI to raise money on nearly trillion dollar valuations, is the expectation from investors that they will produce far more capable models in the near future. These sorts of “Huggingface” incidents, or the Anthropic DoD showdown/designation are thus probably the most effective marketing of their stock that they can imagine.
I fundamentally disagree that this is good marketing. Power is only valuable if you can control it and the entire point of this story is that OpenAI cannot. But more importantly _it doesn't matter_. Whether or not OpenAI thinks this is good for their bottom line is completely beside the point. What matters is if the story is more-or-less true or not. It matters whether or not the model did more or less what is being reported.
If one thinks that OpenAI is in some important way lying, then one should say that. If one thinks that we don't have enough information to be confident about what happened, then one should say that. Otherwise, if one believes that the story is more or less true, then the only argument that makes sense is whether, why, and how, the capabilities described are important or matter. I do not care the slightest bit (and I don't think anyone else should either) what OpenAIs motivation was for telling us.
As I said in my other comment: I'm more than happy to have a discussion about the impacts of this event (well...to be perfectly honest I'm not really, but I'm more than happy to read and entertain arguments about it), and whether and how much we should care. But I'm more than done listening to anyone whose only contributions is that "OpenAI thinks this is good for them". That contributes nothing and is a waste of their finger muscles to type it and my eye muscles to read it.
I think it's just a matter of different ways of thinking between you and most investors. While you personally wouldn't be more interested in investing in AI because it's capable of "accidentally" hacking another company, I suspect that's because you buy strongly into the existential risk argument surrounding AI. This is a belief that most people don't share.
Let's say you're trying to evaluate AI's capability as an investor. You can see internal metrics, tests, etc. but those can be gamed, optimized for, or exist in the training set. Then OpenAI is forced to announce their models did something more capable than they were prepared for, something extremely complicated that seems to take planning, consistency of capability across multiple tasks, and something most of the employees you're familiar with would be completely incapable of doing. While yeah, that shows it's extremely dangerous, that's also a strong confirmation that they'll be able to do less open-ended economically productive tasks! It seems like "Here's my bank statements, do my taxes" and other equivalent tasks are possible in the near future.
It's the equivalent of a weapons manufacturer being forced to admit due "We built an antimatter bomb at our Nevada desert testing facility, and accidentally vaporized 8 square miles of desert/facility." Extremely dangerous, yes, but it's also an unambiguous sign they have an extremely powerful (read: valuable) technology!
So when OP says cynically they're skeptical that this information was intended to shore up stock price, this is what they're thinking. Not even that HuggingFace or OpenAI is specifically lying, it could just be that they "accidentally" left routes for this type of hacking, knowing at current levels there's no risk of superintelligence "escaping" and self-improving, but huge upside of improved valuation.
And if you are doing a cost benefit analysis and thinking "they wouldn't do that" I'd counter with the claim that almost anyone will do almost anything when there billions (tens? hundreds? thousands?) on the line. Especially for something where it's basically "Huh, that's crazy!" with essentially no material harm done to anyone.
Only at the very end there do you even get close to addressing the primary point I was making. I can't say this enough: I. don't. care. if OpenAI thinks that this is good for their stock price. It doesn't matter at all. Yes, if they believe it's good for their stock price, they have incentive to frame the story in ways that make it better, to leave out details, to exaggerate others. They might even have incentive to commit the securities fraud and felonies that you describe. If you believe that, then say that. That is an argument that can be had. Specific predictions and beliefs about the world that can be falsified. Saying nothing other than "OpenAI believes this is good for them" is NOT falsifiable and does not leave any purchase for discussion.
If you believe that OpenAI engineered this on purpose, then say that. Say that you think the whole thing is a manufactured fraud. I won't agree with you, but at least it's something. The original comment I replied to was content-less fluff that is deserving of nothing but trivial dismissal.
It seems like you're saying "I'm skeptical the story is exactly as presented given the financial incentives" is not an acceptable position to hold, and that I must affirmatively accuse them of fraud or not discuss the incentives it at all.
> Saying nothing other than "OpenAI believes this is good for them" is NOT falsifiable and does not leave any purchase for discussion.
You have unreasonable standards for a comment (falsifiability should certainly not be requirement), but this is also wrong. If the hack was bad for them, then them admitting the hack would contain more information than if the hack is good for them.
If the hack is good for them, there are many possibilities;
it was deliberately engineered using AI,
it was done by a human and a complete scam,
it wasn't deliberately engineered but they steered this to happen,
it wasn't deliberately engineered but they didn't try as hard to prevent it as they knew they should,
it was a genuine "hacking event" as presented
If the hack is bad for them everything I mention besides "it was a genuine "hacking event" as presented" becomes far less likely. Thus, understanding the interests of OpenAI does contribute information to understanding the situation, whether or not it's falsifiable.
There's an extremely large gap between securities fraud, and not designing a test as robust as they could, with the unwritten knowledge that this sort of hack was a possibility, or even that this hack was likely.
The rationalist community’s insistence on responding to every challenge to established beliefs with dismissive snark is amusing and suggestive.
I think the quality of my comment more than adequately matches the quality of what I was responding to. "It's marketing" is a valueless comment to make and adds nothing.
Not to mention: I'm not a rationalist.
To anyone reading this thread who might actually be interested, here is a slightly edited comment I have made elsewhere on the topic:
Why are you bringing up marketing at all? If the model is capable of it, then the motivation for why OpenAI is talking about it/reporting on it is completely beside the point. Either the capability matters or it doesn't. If the capability doesn't matter, or doesn't matter in the way that some particular person is claiming, then say that and explain why. Just saying "marketing stunt" adds zero value to the discussion.
If you think that the capability *doesn't* exist (yet), then you are essentially saying that OpenAI is just straight up lying (and, in this case, so is Hugging face apparently), and I wish you'd just say that as well.
I'm very open to arguments about why we shouldn't be concerned about this event (although my prior is very much that we should be, not so much because of the capabilities themselves, but because of how fundamentally misaligned this demonstrates the models are), but I am completely over listening to anyone who has nothing to add other than "marketing stunt".
I was just reading about the Hugging Face incident. If it indeed is true that a Chinese open weight model was used to detect the hack, there is some strange irony here.
It's not just some strange irony, it's an outright own goal by the US's restrictive approach to AI. Now when American companies need to defend their own security, they need to turn to foreign tools.
I've met a few normie Trump supporters who insist that Trump's Iran War will prevent Iran from gaining a nuclear weapon long term. How plausible is this?
Iran doesn't need a nuclear weapon to become a regional hegemon. Using nuclear weapons instead of conventional ones is ... just dumb. They contaminate the entire region, which is bad for your citizens as well. Iran's missiles are winning the war against "The World's Superpower."
Former Arms Control Wonk here:
Long term, as in 5-10 years, not plausible unless Donald Trump escalates the war to "bomb them back to the stone age" levels, and not certain even then.
Short term, if Iran can still access their half-ton or so of 60% uranium, then it's unlikely that this delays by more than a few months Iran's acquisition of maybe a dozen nuclear weapons, but does set back by a few years their acquisition of a hundred nuclear weapons.
If we have put the existing 60% uranium beyond use, or if we somehow manage to do that in the future, that plus the destruction of many centrifuges, that could delay Iran's acquisition of any nuclear weapons by maybe two years.
Bottom line, Iran almost certainly has blueprints for a nuclear missile warhead that will actually work, and blueprints for everything needed to make those warheads and the fissile materials that fill them, almost certainly in many redundant file cabinets and USB drives. The industrial capacity to build e,g, P1-type centrifuges is comparable to the industrial capacity to build modern-ish automobiles. Which Iran was producing at 1E6/year as of 2024. They don't need fancy high-tech machinery, or rare and exotic foreign imports, and they no longer need expert nuclear scientists and engineers. Unless we go Full Morgenthau, they're going to have the capability to build nuclear weapons.
They'll almost certainly pinky-swear to not build nuclear weapons if we offer to stop bombing and blockading them and to let them charge "navigation fees" for the Strait of Hormuz. But that promise will not be realistically enforceable. Any deal Trump can get, will be inferior to the Obama deal because the Obama deal was backed up by Britain, France, Germany, China, and many other nations who all agreed to give the United States sole control of the "Embargo ON!" lever. Also because the Obama deal was made with an ayatollah who probably didn't want nuclear weapons and was just looking for the best possible deal.
Trump, having alienated almost all of the US's traditional allies and having killed the ayatollah who ordered the fatwa against nuclear weapons, doesn't have those cards to play. And he doesn't have enough bombs and missiles to change the things that matter, unless he starts using America's nuclear weapons.
Agreed.
And on the political side, Trump's and Bibi's war is a great showcase for why a country would pursue nuclear weapons in the first place.
I live in Germany. If you asked me if Germany needed the bomb, I would reply "fuck no". But if I was living in a bizarro reality where France was bombing our schools, infrastructure, and political leaders whenever they felt like it, I would probably come around to recognize the benefits of mutually assured destruction eventually.
At the moment, Iran has the ability to squeeze the balls of the world economy whenever they like, and that used to deter rational world leaders. But it is also clear that the Gulf states are building pipelines to bypass the strait. Iran will still have the capabilities to attack fossil fuel industry directly, but this seems less reliable.
So for the Iranian leadership, it is entirely rational to worry about a future where an US president can bomb them without also taking a hit on the markets. Being able to say "we have the capability to escalate to nuclear war, do you want to gamble the population of Jerusalem on your missile defense system?" seems not obviously the worst way to go about this.
More generally, the more the superpowers (looking at the US, but also Russia here) defect from the ideal of the rule-based international order in favor of "the strong do what they will, the weak suffer what they must", the larger the incentive for countries to gain more tangible deterrence, even though that results in a much worse equilibrium overall.
MAD is not premised on a single nuclear bomb (which Iran likely has).
Pipelines are extremely vulnerable to missiles, which Iran has in spades. Nobody's clearing the "Iran can destroy our entire infrastructure" hurdle in the middle east anytime soon. (and this all with conventional weapons).
Tangible deterrence is best done with conventional missiles, if at all possible. Nuclear Korea could lay waste to South Korea at any time, using only conventional weapons (Kim's nukes are, more or less, a vanity project).
Obama made a deal with Iran to prevent them from getting a nuclear weapon. Trump ripped it up when he took office. The best case scenario now is far worse than the *status quo* he inherited.
Everything Trump has done has made it much more likely that Iran will get nukes. Repeatedly breaking agreements and launching sneak attacks will do that.
I think the Trumpy sphere would scoff at the idea that Obama's deal was going to be effective at stopping them from getting a bomb. I don't know the details of it but I think the level-headed consensus was that it would have been effective.
The main criticism of the Obama deal is that it really did prioritize specifically nuclear weapons. Stopping Iran from funding terrorism? Not interested. Stopping Iran from developing long range missiles? Not the goal.
The core of the deal was that Iran could do what it wanted, just do it without nukes.
The thing is, "funding terrorists" and "developing long range missiles" are largely irrelevant compared to getting nukes, but might be important enough to Iran to sink the agreement wholesale. Then you get nothing, good job. Better to focus on the important thing and getting that right. So yeah, if that really was the main criticism, that was a dumb as hell thing to criticize. Nuclear non-proliferation is good, but the US really need to learn when to respect other countries' sovereignty otherwise.
I'm not sure why you'd think that "long range missiles" are largely irrelevant compared to getting nukes. Iran has devastated American power across the Middle East, and all using conventional missiles (and scuttlebutt has Iran preparing to hit Ukraine, just to test their missile capabilities)...
A long-range missile is just a conventional weapon. Countries are going to have those if they want, the technology is just too simple. Historically, attempts to unilaterally restrict a potentially belligerent country's conventional forces have been made (e.g. restrictions on post-WW1 Germany), but never successfully so. They are inevitably considered too serious a breach of their sovereignity, and if you tie those attempts to what you actually care about, like nukes, then the restriction on nukes are also more likely to be ignored.
Also, let's not ignore why Iran has "devastated American power across the Middle East". America's attack on Iran is the perfect justification for having those weapons in the first place. I would wager that US-friendly, dovish voices in Iran have been discredited for decades to come. All that overblown Iranian propaganda about America coming to destroy them has finally come true. What bigger boon for the regime and its ideology could there be?
Would it have been effective in stopping nuclear weapons?
The Obama deal demonstrably *was* effective at stopping nuclear weapons while it was in force, and for a few years beyond. It would likely have continued to be effective because it offered Iran something it rather needed (relief from globally coordinated economic sanction) in exchange for something Iran didn't need (actual nuclear weapons), without touching on anything else the regime felt critical to its interests (the terrorism, the missiles, the death-to-America rallies).
Note that, for the entirety of Khamenei the Elder's tenure as Ayatollah, the only thing Iran ever did with its nuclear arms program was try to get the best possible deal at the negotiating table in exchange for freezing the program at "if we really had to, like if you guys elected some populist demagogue with a hard-on for ayatollah-killing, we could put together some nukes in a year or two". And they were willing to accept real verification and enforcement of that promise.
That was a good thing. I'm pretty sure that in the next few years we're going to wish we still had that thing. But it's gone, probably beyond hope of recovery.
Is your logic premised on the idea that Iran will use its nuclear weapons? Or is it just that "Iran as regional hegemon" necessitates Iran having nuclear weapons? I don't see either of these as high probabilities.
Have they stated any mechanism by which that would happen?
Generally speaking, Iran has the technology and knowhow to build nuclear weapons, and that probably cannot be removed. Actually building one would require highly enriched Uranium, as everybody knows, of which they had some enriched to about 60%, the whereabouts of which is not public knowledge anymore, and the centrifuges to get it to around 80% IIRC. Many of the latter had been destroyed in summer 2025, so the Iran War of 2026 cannot really claim that achievement for itself. They probably destroyed a couple more, but it is unclear how many they destroyed, how many are left, and how fast Iran could build new ones. It does not seem those extra-deep bunkers were certain to be destroyed in summer of 2025, and I have not heard of them being targeted with those special bombs again.
"Long term" is a long time.
Possible ways to keep Iran from getting nuclear weapons long-term include:
a) a deal with a really good supervision routine
b) a changed regime that has no interest in NW or enough interest in bargaining with the West AND that stays in power long-term
c) Shock and Awe that makes the regime give up on the idea forever out of fear for the consequences
d) complete military overmatch that allows the US/the West to bomb any place suspected of harboring centrifuges long-term.
I think the Iran war of 2026 has shown that c) isn't happening and d) would be too expensive for the US and allies to keep it up permanently (they could, perhaps, but don't want to). Regime-chage in the Middle East does not have a great track record of producing long-term stable competent regimes open to the wishes of the US and partners, so I am not confident that if the regime fell, which it does not seem to be doing, it would lead to a better controllable situation.
I guess that means that those people in favor of the 2026 Iran War for reasons of nuclear weapons prohibition believe that a peace deal coming out of the conflict will include a nuclear weapons ban including a surveillance regime that is a lot better than the one Trump ended. That is not what my tarot cards are telling me, but I am no expert.
Removing scientists is something that is routinely done by spy agencies. Iran is hardened against this, because their borders are closed, but I wouldn't be so sure that "the knowledge to make/maintain nuclear weapons" can't be removed.
Please note: the word maintain is doing a lot of that work. Pakistan and India likely no longer have functional nuclear weapons.
I recently read The Achilles Trap by Steve Coll. It explained how the intelligence community got it wrong in Iraq- some people think Iraq never had WMDs. Actually they had plenty of chemical weapons and a nascent nuclear program that could have built a bomb whenever they chose to. But they dismantled the whole program between 1991-1994 after their defeat in the first Gulf War, under the close watch of UN inspectors.
Basically, thanks to the excellent inspections regime, the West's plan actually worked. But it was impossible to prove a negative. So the inspectors kept coming, demanding more and more access, including into places where Saddam hid his relatives and friends, so when inspectors came the guards acted really paranoid (afraid the nuclear inspectors might actually be assassins). So the West was convinced they had something to hide and saw every event and every statement the Iraqis made through a lens of confirmation bias.
(and Saddam made plenty of wacky violent statements that they could read alarming things into, bc he was like that.)
The former head of the Iraqi nuclear program produced a 500 page document, submitted it to the UN, explaining and providing evidence of how the program was fully dismantled, and no one believed him. There was just no way to prove that they didn't have a secret bomb somewhere that the inspectors hadn't found. So, Bush invaded and the rest is history.
The book made me realize that the same thing is true in Iran- no matter what inspections regime gets put in place, Iran will never be able to convince anyone that they don't still have a secret bunker with more weapons as long as crazy hardliners that shout "marg bar amrika" and make every schoolchild repeat it, are still in charge. (incidentally there is a bar in LA, where lots of Iranian expats live, called Marg Bar, for margaritas presumably, but it amuses me)
Saddam Hussein was a crazy evil hardline nutter too, but he actually did dismantle his nuclear weapons. We have no reason to think Iran would do the same even if they did we would end up acting as if they hadn't, so it's just a waste of time to repeat the same plan and think it would lead to peace in the region.
I don't have a solution to this problem myself, but anyone saying the goal of the war and the win condition is setting up robust inspection regime is barking up the wrong tree. That isn't a win, that's just postponing the next stage of the war.
Nu? What if Iran already has a nuclear weapon or two? They have no need to use them, and a single nuclear weapon is only good for blowing up one city or so...
what im saying is, setting "iran has no nuclear weapons" as the win condition for a war, is the equivalent of saying "we've won the war if iran doesn't have any mice." You could go throughout the country with the pied piper, but your work will never be done, there could always be one hidden somewhere. Iran is a big country with a lot of uninhabited salt desert and craggy mountains.
You can never prove it to anyone's satisfaction while the regime itself remains unchanged, therefore you can never truly win the war. Pick a different win condition.
Not that you can get "Mr. Final Pravda" to admit to anything, mind you kindly, but if you can get folks off the record, there are actual win conditions (these include "free strait", "impede china's oil flow -- or wreck it irreparably", and "prevent Iran from acquiring more conventional missiles than we can reasonably block").
You'll note none of these have jack damn to do with nuclear weapons (mainly because Iran doesn't want a weapons program (has no need for one, they can kill the entire middle east without a single nuke), and likely already has nuclear weapons).
Then they need to start saying those other conditions on the record! (but they are probably too embarrassed because those conditions aren't being met either.)
I think the number of missiles they used up would also be a win, if they dont have enough left to do more damage, but then how could they be stopped from rebuilding more after the war....
or get them to stop funding violent groups outside iran, that would be a positive outcome.
It's kind of strange, you would have thought knocking out the entire leadership and sub-sub leadership would have been enough to win but clearly it wasn't. could that have been better predicted in advance? They probably were more effective at it than they even expected to be, but it turned out to have been the wrong goal. There were memes about offended Iranian officials who were mad to be still alive, because it meant Israel considered them not important enough to get rid of. It was that successful. So why didn't it work?
Nit: Iraq had plenty of chemical weapons through 1991, but it was never the case that they could "built a bomb whenever they chose to". That would have taken 5+ years of deliberate, sustained effort. Iraq bet early on plutonium breeder reactors as the basis of their nuclear program, which Israel could and did destroy. They were late to the party on gas centrifuges for uranium enrichment (which can be dispersed and/or buried beyond reach of practical destruction), and still many years away from a bomb by the time the desert stormed all over their plans.
>"Regime-chage in the Middle East does not have a great track record of producing long-term stable competent regimes…"
Presumably a successful nuclear weapons program requires both stability & competence, in which case regime change could suffice to thwart such efforts.
Problem: obtaining nuclear weapons is no longer founded on a "successful nuclear weapons program." See Nuclear Korea.
I think it is basically c) "now they know we will bomb the shit out of them if they try and build anything" or maybe that bombing has destroyed the capability to move material to where it needs to be undetected.
If my math is correct, Iran's current supply of highly-enriched uranium, (enough for a dozen bombs) can fit in three K-bottles, suitable for transport in a donkey-cart. Good luck figuring out which Iranian donkey-carts to bomb.
I will note that *they* do NOT know that "we will bomb the shit out of them." I can imagine many possible future presidents who they would have good cause to believe would not bomb them.
This current war may have pushed back when Iran can get a bomb. Maybe a lot. So I suppose folks can fight about what "a long time" means.
But there is no reason to believe that the folks running Iran think US policy towards Iran is stable from one president to another.
There is also the problem that it assumes that 1) we will know when they start building a bomb and 2) they care if we bomb them [the leadership likely cares more about maintaining power] or 3) bombs physically prevent building a nuke
And also that if we are going to bomb them either way, then they might as well build a nuke so we can't bomb them any more.
Facts not in evidence: "we can't bomb them any more" if they obtain a nuclear weapon.
(There's a strong likelihood they've already got a nuclear weapon or two, from North Korea).
Let joy be unconfined, because hurtling towards us like a Balrog out of Khazad-dûm comes the third season of "Rings of Power".
I know, I know: you were all on the edges of your seats waiting for the happy announcement. Here is the trailer positively dripping with delights:
https://www.youtube.com/watch?v=QmS7WuI9u40
Grandelf finally has pants! Or, well, clothing of some description; granted I have *not* seen his nethers in this trailer so he could still be footloose and pants-free.
Remember how Galadriel told us she laughed at her husband at a dance because of his armour? Well we're going to get that scene, it would seem.
And hang on to your wingéd helms, we're finally getting Celeborn! Now that is a conversation I want to see:
"So... what were you doing for the past few centuries while we were parted?"
"Oh, nothing much; driving my soldiers to mutiny, deciding to swim the entire Atlantic Ocean back home, getting into a situationship with Sauron, macking on our future son-in-law... you?"
I Can't Believe It's Not Saruman is still lurking and making dark pronouncements. Now, he *could* be one of the Blue Wizards, or he *could* be a completely unrelated totally new Dark Wizard of Rhûn, but come on - do you think this lot have the imagination for that?
There's another friendly Orc. Possibly he's meant to be a corrupted Elf, but we'll see. After the family man Orc got stabbinated by Sauron last season (was it only two years ago? how time flies!) there clearly was a vacancy in the cast for Family Person Orc.
Do you want to see more scenes lifted straight out of the movies? Of course you do! Well, we got you covered there, my friends: relive the Hobbits hiding from Nazgûl sent out to find them in the Shire with Friendly Orc, the Plank of Wood known as Arondir and Galadriel taking cover in the undergrowth!
Nazgûl, sez you? But is this not the wrong time for them to be popping up? Nah, don't sweat it: the timeline of this mess is already corrupted to Hell and back, so it's just going to be a recapitulation of Peter Jackson's Greatest Hits. In the wonderfully creative tradition of Payne and McKay that brought us "they're not Hobbits, they're Harfoots" we get "they're not Nazgûl, they're Nazgnagôl" which is a back-translation of "Ring Servants". Because they're not yet Ring Wraiths, you see. Work like this is worth every cent of that one billion, isn't it?
Oh, and *finally* Sauron gets off his brooding backside and forges the One Ring!
The Númenoreans look like they are about to do something useful for once. Again, Ar-Pharazôn looks pretty good in the trailer, and again he seems to be the one person operating with at least half a functioning brain. Dear Show, you are *not* supposed to be making me cheer on the bad guys. Really, you're not.
Círdan grew his beard back, yay! Got over that mid-life crisis and decided shaving with the edges of sea shells is not the best way to update your look, I guess?
Dísa has now been promoted to Queen since Dúrin is now King, after Dad went mano-a-mano with the Balrog last season.
Fell Beasts and Mûmakill and Nazgûl, oh my! all make an appearance. But thankfully no Harfoots or Stoors (yet).
And for your ultimate delectation, we will not alone finally be seeing Anarion, the younger brother displaced in the first two seasons by the invented drippy daughter, but the Balrog speaks! Yes!
https://en.wikipedia.org/wiki/The_Lord_of_the_Rings:_The_Rings_of_Power_season_3
"Also in July 2026, the showrunners revealed that the Balrog would return from previous seasons and would have a speaking role for the first time. They announced that Simon Pegg had been cast to provide the creature's voice."
And for a season opener, they'll be dropping not one, not two, not three but *four*, yes FOUR episodes on us at once. Why, it's almost like they just want to get this damn thing out and over with! But with promises of earthly delight like this on the horizon, it's almost worth hoping the AI apocalypse will hold off until the new year, isn't it?
>Do you want to see more scenes lifted straight out of the movies? Of course you do!
Sigh. One thing I really, really want those clowns to understand is that when I want the Peter Jackson LotR movies, I know where to find them. Before the first season of Rings of Power, I was really excited for it because Tolkien left us scaffolding for some really, really good stories in the Second Age but the source material is frustratingly incomplete. A competent adaptation would have been amazing!
A big part of the problem is the J.J. Abrams school of coming up with "epic' moments and awkwardly gluing them together with mystery boxes and memberberries. One of the magical things about Tolkien's writings is how well thought out they are, how deeply the world-building, themes, and plot fit together and support one another. Payne and McKay are doing the opposite of that and they largely aren't even using the plotting and world-building Tolkien had already worked out for them in the material they're "adapting".
Another part, I think, is that studios don't trust their audiences anymore, especially for big budget productions. Subtlety is not allowed because they don't think we'll get it. Depth, nuance, and even consistency are considered wasted effort because they assume we're less than half paying attention. And everything needs to thoroughly cliche-ridden because it's a "proven formula" that they believe they can sell.
If they had competent showrunners and trusted their audience, they could have given us a proper Second Age adaptation that either focused in on a specific plotline that could be told in a self-contained way or let itself sprawl with cast changes for the mortal characters as centuries passed, going from Aldarion and Erendis, to the first War of the Rings, to the Fall of Numenor, to the War of the Last Alliance.
>Nah, don't sweat it: the timeline of this mess is already corrupted to Hell and back
https://preview.redd.it/g1emeyrrofu91.jpg?auto=webp&s=7c93a80346ef08908f66a408c61bc95972e52299
> Tolkien left us scaffolding for some really, really good stories in the Second Age but the source material is frustratingly incomplete. A competent adaptation would have been amazing!
I don't agree with this. The Second Age, like the Clone Wars, is one of those pieces of backstory that is better left as backstory.
The events of the ages prior to the Third are deliberately unrelatable, they're ages of epic heroes when everything was just so epic and heroic, Elves were everywhere and Men were unrelatably fantastic. You can't get relatable characters out of that. Galadriel (say) works when she's a mysterious and powerful presence who shows up briefly, but it doesn't work when she needs to be a relatable POV character whose journey we follow; either she's constantly wonderful in which case she's unrelatable, or she's too normal in which case she's no longer Galadriel.
That timeline is very funny. But can anything in the third season top the bit in the first season where Celebrimbor, greatest smith of his generation and second only to Feanor his grandfather, was stumped by the problem of forging the rings until a half-dead scruffy mortal initiated him in the secret technique of "alloys"?
Great art for the ages!
That bothered a lot of people, but I don't have a problem with it. I just figured that Celebrimbor was sick the day Aulë taught about alloys.
Elves don't get sick.
True, but maybe he was hungover. We know from The Hobbit that Elves can get drunk and pass out, although Bilbo didn't stick around to see how Thranduil's butler and chief guard felt the next morning.
You skip *one* class because you want to go frolicking in the May meadows and you think you can make it up later, and then two thousand years later it comes back to bite you in the forge.
And thanks to AI, frolicking in the meadows will be the official curriculum. What could possibly go wrong?
But don't you want to know if Grandelf gets a pair of trousers? Don't you want to relive the thrilling sequences that we totally are not ripping off from "The Fellowship of the Ring", cross our hearts, when the Hobbits were hiding from the Nazgul? Don't you want to see if Fake Invented Daughter of Elendil and Fake Invented Son of Ar-Pharazon finally get it together and settle down to breed a family of snivelling little moustache-twirling mini-villains before the giant wave strikes (and it can't strike too soon?)
Though I wouldn't bet on that last, they don't seem to be in the cast list, thank goodness.
Erica, surely you are on tenterhooks to know if the Harfoots and the Stoors make it all the way to the Shi- er, that mysterious green land of not-Hobbit legend and meet up at last with the Fallowhides (who then presumably slaughter them all in an Aryan pure bloodlust genocide so only the white Halflings survive to colonise the fair green land in time for the later Third Age where the Sh - er, mysterious land of legend is full of only white Halflings)?
Do you not want to know how Galadriel juggles her polycule with the return of Celeborn now that she's in a thing with Elrond, in an on-again/off-again thing with Haldir/Sauron, and who knows but she might be open to including a Family Person Orc of mystery as well? And of course Arondir is available now his human squeeze took an arrow to the chest and died of "the actress left before the second season was shot"?
EDIT: Whoops, I meant Halbrand! Halbrand, not Haldir (although... Glads here is throwing her eyes at every half-passible guy in the cast, so who knows?)
Where is Miriel? Does anyone know? Does anyone care?
They could have had a decent go at "okay, we have to do a heavily compressed timeline of the Second Age leading up to the War of the Last Alliance and the downfall of Númenor, but that's okay, we have five seasons and with some care it can be done". But they also, I imagine, were under commercial pressure to pull in new Prime subscriptions, so it had to be "people are fans of the movies, they want to see the characters from the movies, you gotta put in the names they know and the callbacks to the movies!"
As well as Payne and McKay being talentless hacks who only got the gig through J.J. Abrams whose powers of screwing up long-established and well-beloved franchises are truly awesome.
I can hardly wait. I'm just grateful that you're watching this so I don't have to. I'd rather read your rants on the subject, which are shorter, more entertaining, and better-thought-out than the episodes at least up to the point where I gave up on watching.
Can't wait to completely ignore this one, too!
If you are a very big Tolkien fan, it might be painful/sad to watch it, but we've had great fun laughing at the flaws of the series. There are worse ways to spend a few nights imo.
Charlie Hopkinson's parodies are very funny:
https://www.youtube.com/watch?v=mNyNyqgkgh8&list=PLyJrI7fwl_bP8yueSwegq_msQL_Rt5nTS
And yes, this really is how the geniuses of the adaptation came up with a backstory for "How Gandalf Got His Name". By this stage, you can only laugh, crack open another 75cl of Tesco's own brand cream sherry or ruby port, and search for the piss-take reviews on Youtube.
Oh, I am waiting. With keen anticipation and a bottle of Sarson's finest malt vinegar by the screen.
I don't see this turkey surviving for a fourth season, but it depends how dogged Jeff is about "I really love and adore - reads name off hand - Tawkeen?" and if they've signed too many contracts binding them in the Black Speech to deliver all five seasons.
Erik Kain, God bless the man, returned after a year's absence for many reasons just in time to be hit between the eyes with this trailer and the mysterious floating sword:
https://www.forbes.com/sites/erikkain/2026/07/25/rings-of-power-release-schedule-amazon-has-zero-faith-in-tolkien-adaptation/
https://www.forbes.com/sites/erikkain/2026/07/25/rings-of-power-season-3-ai-sword-image-/
I’m surprised Bentham was ever suggested as the first to use a self suffix. It’s one of the reasons that Hand D in the Thomas Moore play was attributed to the Bard
> With self-same hand, self reasons, and self right…
Friday Feature: Mini Cities
https://www.cato.org/blog/friday-feature-mini-cities
"Kahl thought “a researcher coming from a different planet would find nothing particularly special about Mini Munich.” It would seem obvious to them that if adults spend their days starting businesses, running newspapers, electing governments, and settling disputes in court, then a miniature version of that would be a natural way to prepare children for life. In the scaled-down world, cause and effect become easier to see—businesses succeed or fail, policies reveal their consequences, and “a bad law gets repealed after a few days, not after decades or centuries.” Rather than teaching civic and economic life as abstract subjects, miniature cities allow children to experience them firsthand."
Relatively unknown in the west, but these may also be of interest: https://en.wikipedia.org/wiki/Children%27s_railway
We have the live steam model railroading community, but not many young people in that (except at passengers during public run days). Train Mountain in Oregon is the big one, but they're all over.
I've thought for a while about the "you live in your mother's basement" insult and how weird it is, how bizarrely specific and reliant on assumptions from a number of seemingly incompatible ideologies. I'd like to know what people think of my analysis below, because most examinations I've seen of it seem incredibly shallow and mindless.
So first, there's the gender aspect. Mostly it's "your mother's basement", a significant amount of the time it's "your parents' basement" (I think someone said this in a recent ACX thread), but as far as I can remember it's NEVER "your father's basement". What's going on here? I think I've seen feminists explain it as simple misogyny, the idea being that a grown man living under the authority or control of a man is fine but living under the authority of a woman is horribly shameful. (But then why is "your wife supports you" not an insult of similar strength?) But I think I've seen manosphere types describe it as misandry, as equivalent to "neckbeard" in that it's only used against men and the idea, I think, is that it's based on the assumption that men (unlike women) have value only in accordance with their economic success and social status. (But then why the focus on the mother?) And a comment on this page https://www.explainxkcd.com/wiki/index.php/713:_GeoIP suggests it's an implication of incest, which would explain why it's usually "mother" alone, implying she's a single mother to make the incest insinuation work. (But I haven't seen this interpretation very often.) So which is it? All of them at once???
Then there's the class aspect. I know a guy with rich parents who was living for a while in their "spare penthouse" while figuring out what he wanted to do with himself. There are lot of things you might be insinuating by mentioning this, but none of them seem anything like what's being insinuated with the basement. The usual form would be as an angry accusation of being privileged and out of touch, not as a mocking insult of being a loser. But they're the exact same thing except for how rich your parents are. So it's an anti-poor insult in part? No, because it relies on your mother/parents owning a house; "you live with your parents in rented/public/charity housing" just doesn't have the same impact. Maybe the idea is you're inconveniencing your parents by taking their basement in a way you're not by taking their spare penthouse, but then why is it "basement" which mitigates this by implying you're out of the way in an unneeded room? "You sleep in a cot in the corner of your parents' bedroom" would make this accusation much clearer as well as making the "adult baby" effect much clearer. So it seems to rely on your parents being both not too rich and also not too poor for the insult to work. The only thing I can think of is that it's based on a kind of class propriety, of a man born into the middle class failing to live up to the middle class norms that would not apply to him if he were of a different class. This is really weird for a 21st century western insult of such prevalence, but it's the only reading that seems to make any sense.
And finally there's a subtler point. The insult seems to be based on "capitalist logic" in that it demands self-sufficiency and proxies that with what you can afford. But normally capitalism, when *explicitly* defended, is defended based on individual freedom and the greatness of non-coercive voluntary arrangements. Unless the implication is that the guy is physically threatening his mother to let him stay in her basement, that's exactly what this is. An accusation of being on welfare would work vastly better on a normal capitalist logic, while on any kind of socialist logic it would be pretty verboten to define someone's value by what they can personally afford, even separate from their own family. The only way this particular shaming makes sense is under a kind of "collectivist capitalism", for want of a better term, whereby a virtuous person is obligated to be a productive member of society and not benefit from any voluntary arrangements of either charity or excessive family support. I imagine this attitude is actually quite common, despite rarely being explicitly defended in political debate, and I also imagine it's particularly common in East Asian cultures, though those cultures are also the least likely to use this kind of insult, what with their strong family-oriented ethic.
(The explainxkcd link above also claims this insult is specific to North America, which I'm very suspicious of because most claims of this form turn out to be false, though I don't think I've heard it here in Australia, although I'm not very social and it seems like the kind of thing that would only be said on the internet anyway. Except for the superficial aspect of having a basement, which is I think mostly specific to houses in (I forget which part of) North America.)
So, in the end, my very rough theory is that this insult has naturally evolved to almost perfectly straddle the line between many incompatible ideologies, without clearly endorsing any of them. It has just enough misogyny to appeal to misogynists and just enough misandry to appeal to misandrists, just enough "you're privileged" to appeal to poor people and just enough "you're poor" to appeal to rich people, just enough "moocher" to appeal to individualists and just enough "unproductive and unvirtuous" to appeal to collectivists, and this is why it's become so prevalent. But I have low confidence on this.
I'd like to know (a) what people think of my analysis and (b) what other insults people can think of that have this trans-ideology effect. I wonder how many other cultural phenemona generally are like this, where adherents of opposite ideologies can say that it's a sign of how far society is away from their ideal, and all of them be correct.
(Epistemic status: personal recollection)
When I was growing up, these were 2 separate insults: "you [as an adult] still live with your mom" or "you [a child or immature adult] [engage in low-status hobby] in your mom's basement".
A basement seems like an entirely natural out of the way place for a hobby that takes up a lot of space or is messy.
I think there might have been a transitional stage (not entirely sure) where "living in the basement" implied you were so obsessed with your low-status hobby that you lived in the place you do it, similar to how a man (I think this was always a male-coded insult) who was obsessed with fitness might "live at the gym" or a man who was obsessed with his car might "live in his garage".
At some point (possibly through the "living in the basement" intermediate step), the basement insult and the living with your mom insult merged into "you live in your mom's basement".
Regarding "mother's basement":
It's "mother" because the holdover from traditional gender roles is that your father might offer you whatever help you really need, he's doing it in the expectation that you'll use it to finish growing up and becoming a productive member of society that the family can be proud of, whereas the mother is coddling a child that she never wants to grow up and leave the nest.
And it's "basement" because, as Erica Rall has noted, in most North American homes the basement is the place you can fit a quasi-apartment that gives a cheap imitation of independent living to the overgrown adolescent who can't actually live independently.
Garages can also sometimes work for this.
Mom house because single mothers more common. Broken family.
> But then why is "your wife supports you" not an insult of similar strength?
In manosphere it most certainly is, but it is not used nearly as often as "you live in your mom's/parent's basement". My impression is, people use the former only if they know (or are reasonably sure) that the wife does in fact support the husband in question. Whereas the latter gets thrown around with total disregard for the statement's veracity, much like sixth-grade boys calling each other "fag".
I expect that men who both subscribe to a version of masculinity that feels threatened by the woman being the primary breadwinner and don't earn very much would tend to be single.
If person A is an insecure manchild who thinks that telling person B "Your wife supports you!" is a great insult, does not mean person B is the same. My wife earns more than I do. If someone were to insult me in this manner, I would just laugh at them.
In addition to what everyone else has said, I think a side connotation of the insult is that the insultee is a lesser son of greater fathers. That, at their age, their parents were already busy with the accomplishments they are now mooching off of. So the guy living in a spare penthouse may be living in material comfort, but he's still a failure *as compared to his parents,* who were presumably busy building (or maintaining) the family fortune at his age.
If it's his mother's house, why isn't he in his childhood bedroom? The implication is that either the mother has moved on from prioritizing his needs (implying he's unlovable) or he hates the family home so much he'd rather occupy the dingiest, most secluded part of it (implying he's self-destructively antisocial).
As for the gendered aspect, I think it's about setting the locus of power in the household. I have a friend who describes their father as "living in his ex-wife's basement", even though it's a house they bought together. To my ear that hits just as hard, if not harder than the mother's basement.
I think the implication of "living in the basement" and not his childhood bedroom is that he tried and failed to leave and live independently, and now he's back home and Mommy is having to take care of her little boy again.
I think the significance of basements is that they're a largish area that's relatively isolated and generally used as utility and storage space but can often readily be converted into a makeshift apartment. Culturally, basement apartments are an awkward middle ground between actually moving and getting your own place vs continuing to be a regular part of the household. I think the implication of the insult is that someone who moves into a basement apartment in their childhood home is someone who's old enough to want the independence of living separately but not capable or financially secure enough to actually get their own place.
If it’s American at all, it’s because Americans have basements. While we don’t in the U.K, or most of Europe.
Of course it’s emasculating. Hence the mother. It’s referring to a male adult staying at home past adolescence, but kept (or keeping himself) in the basement. Subservient to his mother.
Not an airy upstairs room he might leave easily to go for a jog, but the perma-gloom of an underground abode.
Movies often put the mad scientist in some kind of gloomy lab, underground or, if not, the windows are blacked out, or encrusted with dirt.
Another key component of the insult is the implication that you have no partner (and no hope of getting one) and probably few-to-no irl friends. (And should you have any they are likely all losers as well.)
Regarding the mother vs father: I think there's an assumption here that mothers will be more coddling than fathers, and so if it was 'father's basement' there is more likely to be a legitimate (non-loser) reason for living there. Whereas a mother is more likely to just let her son live in the basement forever, or even demand it in some cases (cough).
I'm pretty sure that "parents' basement" was either the original form of the insult or at least that "parents'" and "mother's" were used interchangeably in the past. As to why "mother" seems to have displaced "parents", you might be right. There's also the related idea that it's unmanly for older boys and young men to be reliant emotionally on their mothers, since mothering is culturally associated with the care of young children and boys especially are supposed to outgrow being "mothered" as they progress from childhood towards manhood. So the "parents' basement" insult is getting the "go back and cry to your mommy" insult layered atop it.
Alternately, there might be a layer of insult directed at children of divorce or children whose parents were never married. By saying "mother's basement", you're implying that one's mother and father don't own a basement together, which can have insulting connotations because of cultural baggage around broken homes and illegitimacy.
I always used the threat of moving into my mom's basement as a motivator to keep my sh-t together. But to be fair, my mom's basement was full of random crap and flooded a lot of the time, so it was pretty unappealing....
I figured that change was due to a rise in fatherlessness.
I believe it builds on the Eric Cartman-esque stereotype of a son who abuses his mother's generosity and does nothing but take from her.
You touched on the implication that the man's mother is a single parent while discussing incest, but it's also important in that his mother is more vulnerable without a husband. There's a stereotype that opposite-gender parents are more lenient and permissive.
I also think the basement is important to this reading. A basement is separated from the rest of the house. It emphasises that the son is not taking part in the daily life of the household; that he is a parasite who hides in the basement while his mother socialises and works above. It's also lacking in sunlight and fresh air while being much larger than a bedroom, implying a willingness to trade comfort for the space to avoid others. This is why the piss bottle and mini fridge are relevant to basement dweller insults. The son goes to extreme lengths not to participate in daily life.
It's partly a holdover from when housing costs were dramatically lower and you had to be some extreme mix of lazy/stupid/incompetent/maladjusted to live with your parents into adulthood in most cases.
The reference to mother's basement makes sense because there is a pre-existing and similar "mama's boy" stereotype of a meek man who never strikes out on his own and remains dependent socially or financially on his mom.
The insult persists because it is still directionally correct in indicating lack of success/discipline/effectiveness in life. Although living with parents well into adulthood has become significantly less stigmatized as long as you actually have a job and are contributing in some way.
To illustrate how much things have changed: Breaking Bad aired its first episode in 2008. In the episode, the audience is supposed to pity Walter White's financial situation. But he was able to afford a large three-bedroom home with a pool, a stay-at-home wife, and a child, by teaching high school and doing one part-time low-wage side gig.
Re: Walter White, yeah, that's the movie/TV notion of "just about making do". The 90s sitcom "Grace Under Fire" was the same; the premise is "Grace Kelly, a single mother and recovering alcoholic raising three children in a small town in Missouri after finally divorcing her abusive husband" and "In the first three seasons, Grace operates pipelines at the local oil refinery" which you could probably make a decent living doing.
But the locations (when you saw them) were "single divorced mom of three living in detached house on own property with plenty of room and plenty of whatever anyone might want" and it made me laugh (not at the jokes) to see "so this is the Hollywood idea of struggling lower middle class, huh?"
Roseanne is a solid "this is struggling lower middle class" show.
I don't think it is just Hollywood logic. I remember from my own childhood, there were dads who were truck drivers, construction workers, etc, and they could afford to own or rent their own homes with a stay-at-home wife and kids. Sometimes the wife would do a small low-wage side gig like substitute teaching or watching the playground at recess to pitch in.
Yeah, but that's a stable family. Grace's situation was that she was (1) divorced (2) recovering alcoholic and (3) had three kids. Now, the job they gave her was blue-collar but semi-skilled (at least) and traditionally male, so she would have been earning a good wage.
But the circumstances didn't seem right for what she was supposed to be. Now maybe before the divorce the husband had been earning well and they were able to afford the house etc. and she got it after the divorce, but the tone of the show, at least to me, was meant to be "relatable working-class family on the lower end of middle class". The look of the show, however, was not "struggling single mom trying to rebuild her life" but I accept that after the first season or so it dropped all that and just became another sit-com (and apparently went down in a blaze of controversy at the end? https://www.youtube.com/watch?v=-LiaR4n4HFI)
I agree with the comments about Roseanne, that was much more realistic-seeming in the size of the house, furnishings, etc.
You're overthinking it. It's like calling someone a mama's boy, only applied to an adult. It's saying that the person is coddled, unable to take care of themselves, and a man child.
Isn’t ACX the place where we come to overthink things?
I don't know, let me think about it...
Touché
> The usual form would be as an angry accusation of being privileged and out of touch, not as a mocking insult of being a loser. But they're the exact same thing except for how rich your parents are.
They're the same thing except one is a basement and one is a penthouse; one is a nice living situation and one is damp and dark. The insult is not just that you're dependent on your parents but that your living situation is particularly bad. You're not even living in a proper bedroom, you're literally beneath them like they're ashamed of you.
> (The explainxkcd link above also claims this insult is specific to North America, which I'm very suspicious of because most claims of this form turn out to be false... Except for the superficial aspect of having a basement
I don't think that's all that superficial.
Overall I don't think it stands up to heavy analysis. Sometimes a label just emerges an insult for a particular group of people. The insult "Karen" isn't actually about having the name Karen, and it's somewhat arbitrary that it wound up as Karen instead of Sarah or Megan. A "basement-dweller" isn't just financially unsuccessful, he's uncomfortably immature and probably obsessed with something that most people find offputting.
Just want to applaud these two short summaries each of which is spot-on target::
"The insult "Karen" isn't actually about having the name Karen, and it's somewhat arbitrary that it wound up as Karen instead of Sarah or Megan."
"A "basement-dweller" isn't just financially unsuccessful, he's uncomfortably immature and probably obsessed with something that most people find offputting."
This genuinely fascinating analysis is some weird apotheosis of ACX comments - it's a deep dive on something I couldn't have imagined caring about ten minutes ago but now really has me thinking.
One addendum I'd give is the "threaten your mother" thing - I think at least in some circles there WOULD be an implication that someone browbeat their mother/parents with "you gave birth to me so you owe me and I owe you nothing" and their too-soft parents gave in, and that's part of the source of their sucking-as-a-person.
> But then why is "your wife supports you" not an insult of similar strength?
Who says it isn't?
> The explainxkcd link above also claims this insult is specific to North America, which I'm very suspicious of because most claims of this form turn out to be false
The appropriate age to leave the nest differs from country to country. Some think it's shameful to live with your parents when you're a high school graduate, some think it's shameful to do this when you're in your thirties (and single!).
If your wife supports you then that means you've got a wife, which means you've achieved at least some level of adult respectability.
It's an outdated insult from back when the frontier was open and it was legal to build new housing. In those days, a man might reasonably be accused of cowardice for not moving out and seeking his fortune. Now that there is no more frontier and boomers have made it illegal to increase the supply of housing, moving out of your parent's house simply means moving in with a roommate. No thanks.
>moving out of your parent's house simply means moving in with a roommate
That has been the norm for young adults for decades. The implication of the insult is that you are incapable of paying rent or living independently.
Indeed....Dad-brag-alert! [advance apologies if this violates Scott's criteria for appropriate commenting]
When my eldest was graduating from college and weighing his next steps I made just two pronouncements:
-- happy to help pay for grad school _provided_ it is towards a specific field of endeavor that you've chosen for specific reasons; grad school as a continued holding pattern is a non-starter.
-- there will be no living at home after college graduation. My argument being that paying rent with roommates is not money wasted, rather it is paying tuition on living in the real world. Had to talk his mom into this one as she was hearing the short-sighted "rent is just a waste" cliches on socials and whatnot, but she took a breath and signed on to the above declaration.
He did go straight to paying rent with roommates (in one of the supposedly "illegal to increase the supply of housing" metros), that went fine and he gained all the navigating-real-life lessons I was aiming for.
In his late 20s he did get a master's in a chosen field that he'd done some real research about, and between his maternal grandfather and I and himself [his day job liked him so much that they begged him to stay with them part-time while doing the degree], it was fully paid for. In his early 30s he's doing well overall including just signed a lease to cohabit with a fascinating woman who may be my future daughter-in-law.
My youngest, now entering high school, will when the time comes have the same stipulations for the same reasons. Though of course they are different young people and early indications are that this kid may not need it (he's startingly self-directed for his age which, my siblings and I agree, has to have come from his mother's gene pool).
"The implication of the insult is that you are incapable of paying rent or living independently."
This.
And (I think) there is also the implication that you not paying rent has a negative impact on the parents' finances. So living in the "spare penthouse" doesn't hit the same way.
There is also the aspect of "living in the basement" which implies "has no social connections in the real world." Living in the spare penthouse doesn't imply that.
Note that "taking care of your mom/dad/parents" can have exactly the same living situation -- you still live at home -- but is virtuous because you are contributing rather then taking.
Sounds like a way to make fun of poor people, with some plausible deniability that it is actually about their character.
Right, as Melvin noted it's the immaturity that is the core of the insult.
I am currently traveling abroad and doing a quick field study to explore local universities and research centers.
I’m curious about their educational culture—how people think, interact, and collaborate from the inside.
For those with experience in academia or field observations:
1. How do you practically observe a university ecosystem on the ground?
2. What are your best tips for documenting and thinking about what you see?
Would love to hear your insights!
What type of departments? I would say go to the seminars/ colloquiums. Observe the audience. That can tell you a lot about a department.
Spot on! Watching how the audience engages and asks questions in a seminar says more than any brochures.
I'm mostly targeting Systems Engineering and Fundamental Sciences. Thanks for the tip!
I mostly know physics departments. You can try and engage with the grad students, they know most of the department scuttlebutt*. Maybe even get to look into their labs... if that's the type of thing you are interested in. The most important thing is probably to find the right advisor/ (professor to work for) the exact school is less important.
*what a great word... the old time water cooler.
Nominative determinism? https://substack.com/@drboguslaw/p-207895760
Another thought about the recent hugging face incident is that afaict it looks a lot like what you'd need to do to hack an election. Traditionally a big problem with that has been that US elections are handled at many state and local levels each with their own separate system and you'd need massive parallel coordination to hack them all (and even more coordination to make the results not look obviously manipulated), but an AI swarm seems like a pretty good way to do it (without even requiring other humans in the loop, so long as you have enough AI access). And in a world where partisans on both sides seem increasingly convinced the other side is going to manipulate elections, there's more likely to be someone with both motive and access to do it.
(Funniest result I guess would be if somehow people in both sides do it at once and the next major election just becomes an AI hacking duel).
>you'd need massive parallel coordination to hack them all
That's where voting machines come in, from a relatively small number of vendors with lax standards. They have been around for decades at this point and they've always been a security concern to those willing to look and listen. So no, you don't need AI to hack them. And no, you don't need to hack all of them either, just a few in the right places.
I deeply wish we would return to in-person, inked finger, punch-card/hanging chad, full-ID-checked voting, with no electronics whatsoever. Everyone is afraid of hackers, and there's almost no system that's hack-proof or furthermore is widely understood and believed to be hack-proof.
>full-ID-checked voting
My most right wing view is that requiring voter ID is completely reasonable.
It's really odd that voting is *the* vital process of a democracy, yet there's absolutely no requirement to prove your identity.
I've been into election security ever since my hometown had a Chinese national (non-US citizen) vote in the 24 elections. The Secretary of State announced that because his ballot was scanned, there was no way to invalidate it and his (illegal) vote would count.
Can you cite some sources on that Chinese national?
Yup.
https://www.justice.gov/usao-edmi/pr/chinese-national-university-michigan-charged-illegally-voting-2024-election
https://www.fox2detroit.com/news/ballot-allegedly-cast-chinese-non-u-s-citizen-ann-arbor-likely-count
I forgot that he surrendered his passport but somehow fled the country with another Chinese passport.
Ah, Michigan. One of those "Red States" that is captured by the Democratic Machine. How many "mail-in" ballots didn't get folded in Detroit, I wonder*?
(Biden lost black votes compared to Obama... except for a few Democratic Machine cities... wonder why?)
*This is a reference to Tulsi!'s search warrant for Fulton County.
There is nothing wrong with electronically counting paper ballots. Concerns re hacking the counting software can be addressed by hand counting a random sample and doing a full hand recount if the discrepancy will the electronic count is large enough.
Thank goodness that there is no way to hack random number generators.
You don't need random number generators to select a random sample of ballots.
You don't need a machine to vote either.
How is that remotely relevant to the issue under discussion, which is how best to count paper ballots?
At this point it feels cheaper and easier just to hand count the whole thing from the start and save on all these machines.
You still need to double check the humans' work, both to prevent fraud and for the simple fact that humans can't keep perfect focus while working through millions of pieces of paper. And if you throw more humans at the problem it stops being cheap.
Also, hand counting is much slower, which is a real problem because one of the go-to Republican accusations of fraud has been "look, Trump was winning when the count started, but lost once more ballots were counted, clearly this must be fraud and not different polling places returning their results at different times." It would be much worse if it took multiple days for the count to be finalized rather than hours.
Hand counting is done by volunteers in the UK.
On the other hand, if it did take multiple days to get a result, you could better accept "a whole lot of new ballot boxes came in over that time" rather than "four hours ago X was winning and now mysteriously Y is streets ahead".
It's just not that slow. In Australia we hand-count everything (and also have instant runoff voting, which can mean multiple counts) and yet we usually know basically who has won within a few hours, and have the complete count pretty much done overnight.
As I understand it what's really slowing down vote counts in the US at the moment is the huge growth in mail-in ballots, where there's a need to manually do the security theatre of comparing one possibly-fake signature to another possibly-fake signature.
Well, there's also the fact that the US typically puts many, many times more positions up to democratic vote than Australia does. They vote for President, Senate, House, state senate, state house, Governor, state judges, state cabinet offices, school boards, county offices, city offices, police chiefs, fire chiefs, many other chiefs and God knows what else on a single day's typical ballot I believe. Not to mention the roughly half of states that have citizen-initiated referenda and propositions on every ballot. Not to mention primary elections. We trade speed and security for having a grand total of two federal, one or two state, and one local office up for any kind of vote ever. (Though thanks to IRV our votes are much more powerful and carry a lot more information. Canada and the UK are the worst of both worlds I believe.)
What is "wrong" is that it is a lot easier to be confident that locally counted paper ballots were not hacked. There may be fraud at a local precinct level, but widespread fraud requires enormously more effort than hacking electronic voting. At least that is what lots of people think.
I believe there is value in voters having confidence in the process.
>I believe there is value in voters having confidence in the process.
There is definitely value in that (see my opposition to ranked choice voting). But, whenever the count is slow (as a complete hand count would be), some people claim that THAT is evidence of impropriety.
It's also harder to predict what helps voters have confidence, and taking more care doesn't necessarily help (for example, if your local deli put on a full body hazmat suit to make your sandwich it would not leave you feeling comfortable). Paper ballots might help by virtue of simplicity, but talking loudly about security probably harms voter faith.
(For another example, taking forever to approve COVID shots didn't help prevent fears about it at all).
That's predictive programming, in the sense of "full body hazmat suit" means Bad Problems Incoming. If your local deli was using a cleanroom-grade air purifier, you'd feel more comfortable. And it would help with eating less fecal matter.
Also certain motivated actors are invested in attacking that confidence for their own advantage.
Regulating your elections to a "what makes everyone *feel* secure" standard, rather than a "what is secure" standard, invites Purple Party adherents to simply find themselves unable feel the elections are secure unless and until elections are structured in a way that ensures Orange Party adherents encounter maximum structural difficulties when voting.
Better to structure for actual security, and then treat feelings issues as feelings issues; something to be solved with many more parts education than structural change.
Then you wander off into territory of who does the hand-count, what if they use electronics to summarize the statistical sampling, what was the algorithm used, who approves the legality and process across 50 state jurisdictions, etc. etc. etc. Once you introduce electronics, you introduce a possibly unacceptable level of doubt and fear.
Not saying any of this is guaranteed better with paper, but it worked for decades without suffering trust issues (except in Chicago and Tammany Hall, fine).
Robert X. Cringely wrote an article in the early 2000s addressing most of the issues with hand counting. Each of the major parties are allowed one observer of the counting, each ballot is examined by at least two people, etc.
Hand counting worked for decades and can be made visibly robust (if not perfect).
And rapid. We hand-counted for decades and had next-day election results nationally.
And pretty much all other democratic countries still do it that way.
Yeah, security researchers have been arguing for paper based elections for a while and this increases my concern about it.
I genuinely don't understand why anyone would expect China or other countries that feel threatened by non-AI concerns (e.g. Ukraine) to hit Pause on AI when it offers them hope of succeeding in their offensive or defensive suite of capabilities.
Even if they appeared to genuinely accept a pause, how could you ever verify that they had?
The genie is out of the bottle. Now that this capability is known to exist, it will be created, in secret as well as in public. AI alignment efforts may well work great, but a nation-state or powerful crime organization will turn off the alignment and reap the rewards. Now or 10-20 years from now, it will happen.
Although it could help their military, it could help other countries' militaries just as much or more, leaving them no better off. If they're much worse at AI than other countries (eg Ukraine, which is far behind the US and China), then it makes them worse off relative to those other countries (although Ukraine could still potentially benefit if the US gave them powerful AI and nobody gave Russia powerful AI).
In terms of "how could you ever verify that they had?" this is the main question addressed by Plan A, which I blogged about earlier this month. See https://www.astralcodexten.com/p/introducing-plan-a for my summary, https://ai-2040.com/ for the full document, and the verification supplement at https://ai-2040.com/supplements/verification-plan for specifics of verification.
I think that, while Ukraine is indeed much worse at LLMs than China, they might be currently leading the world in non-LLM machine learning applications -- e.g. those needed to pilot autonomous drones.
I understand the argument in Plan A, but I most certainly don't trust China remotely enough to think it would work. Even if it all appeared to work superficially, it's a scarce and critical resource over which control can make-or-break an empire. That is how wars start, not how they end.
It's a trustless deal. You're not supposed to have to trust them.
You've put a lot more time and thought into this effort than me, so no need to reply, but here's my most immediate concern:
"Establishing control over the chip supply is easy. Only a few companies can design AI chips (eg NVIDIA in the US, Huawei in China), and only a few factories in the world can produce them (eg TSMC in Taiwan, Intel in the US, SMIC in China). All of these companies and factories are in the US, China, or client states kept on very short leashes. The US and China simply tell these companies and factories to send chips only to licensed customers"
Why wouldn't China just massively invest in Huawei and SMIC (for example), invade Taiwan to take over TSMC, generate the next generation of top-performing chips first, send them only to Chinese military clients, and game over for everyone else? What's in it for them not to build an all-China supply of AI dominance? How would I trust them not to do this, especially when I've concentrated everyone else down to just the US?
Why wouldn't the rest of the world consider the US just as bad as China, and have this break up NATO and divide Europe from North America? When Brazil or France want chips and have to be US or Chinese supplicants, they will defect as well, and in many other dangerous ways. I just don't get it.
In Plan A it is a (relative) slowdown for about 10 years - they prefer that to a full stop forever for (IIRC) the reasons you give.
There are lots more details in the plan. eg If China invaded Taiwan to take over TSMC then the US would declare the deal void and all the treaty data centres would be destroyed.
Sorry to be dense, but how does a slowdown change the calculus? I'm just thinking through: China sends bazillions of yen to Huawei and SMIC, they get a design and implementation of a faster set of chips on the drawing board ready to be fabbed, quietly build out the fab somewhere (or quietly prep in existing facilities not obviously visible to inspectors), or lock out the inspectors, go full steam ahead on the next-gen chips, and again, game over?
Let's assume NVidia and TSMC and ASML and whoever are all doing the same, but just a hair slower. Isn't it an unstable equilibrium?
Please note that "AI" is a horribly overused term. When you talk about drones for the Ukrainian battlefield, you mean image recognition to discern between defined types of targets (soldiers, vehicles etc.), and non-targets. When the doomers talk about AI, they mean AGI with access to the internet and able to pursue their own goals with a broad set of capabilities.
Doomer pedant here. ASI, not AGI. Human level and barely superhuman AIs are safe, except for their potential to improve. It's the ones actually much smarter than us we're afraid of.
For the purposes of this discussion, AI is assumed (by me) to refer to the HuggingFace incident and related types of GenAI capabilities, though GenAI can also leverage tools that perform image recognition of course.
(Possibly apocryphal but metaphorically useful, history majors don't @ me) Chinese emperors wanted to live forever, and only wanted that, and ended up drinking mercury and dying pointlessly early. If civilizational immortality is actually out of reach, lets go for a long happy life followed by a drawn out death by ordinary terrible cancer, and not spectacular heavy metal suicide. If immortality is attainable, chugging mercury also isn't necessarily useful toward that goal?
living a long time looks like repeatedly not dying yet due to specific causes via a lot of gruntwork.
Well, I'm not at all an AI doomer, though we live in interesting times. We will have to come to terms with this new set of powerful capabilities somehow.
(Pursuant to prior comment) a trader from the 1700s using pieces of paper to perform their business at a snails' pace would have a very hard time coming to terms with 24/7 speed-of-light comms FX trading, but we made the leap over time. This timeframe is very compressed, but it's of similar magnitude to that industry shift.
Random silly thing that keeps popping into my head over the last decade as a puzzle. How can you generate random numbers in your head using ONLY your own thoughts? No external stimulus, and you can't even interact with your physical body to use things like the length of your hairs or your heart rate. Just full brain-in-a-jar.
I have my own ideas based on like "things that you know but you don't know you know until you look" but I'm really curious what other people's gut solutions are.
I'd often frame this as like "you and 99 other rationalists (so you aren't encouraged to assume everyone else will skew a specific direction and actively compensate) are put in that situation and have to decide whether to press a button or not. If between 15 and 25 of you end up pressing the button and the rest don't, you all get a reward" so you're encouraged to give yourself as close to a 20% chance of pressing the button as possible" but if you find that exploitable/distracting feel free to steelman whatever scenario you want. The specifics shouldn't really change the fact you just need to produce a random number of some sort.
Most people have several longish numbers memorized. Pick the last digit of your social security number, phone number, library card, student ID, etc. Doesn't particularly matter which one as long as numbers are assigned randomly or sequentially and you're not selecting based on what that last digit is.
One way might be to imagine a spinning globe of the Earth, stopping for a moment here and there. If you can persuade yourself you could make these stops fairly randomly, with a bit of practice, then the stopping longitudes facing you (and latitudes if you also select a point randomly up or down the line of longitude) could be mapped to digits in some base.
This assumes a far more unrealistically detailed knowledge of coordinates of places round the earth than most people have. But that knowledge is attainable in principle, just as remembering a million digits of PI would be for a mental random number generation scheme which somehow relied on that!
If you can do 5-digit arithmetic in your head, you might be able to implement a 16bit Linear Congruential RNG, requiring one multiply, one addition, and a modulo step..
I think specifically latitude/longitude is probably quite challenging, but the core idea of like "visualize something concrete, pick a point on it, and then turn that into coordinates you use to pick the number" actually sounds like a fairly decent solution along the whole line of "pick a non-number in your head and then figure out what number it correlates to in a way that you won't subconsciously known in advance."
I do wonder if specifically picking things along a 2D plane though would still run into issues with humans told to be "random" usually not doing repeats sufficiently often (i.e. people told to pick random numbers between 1-100 repeatedly will pick the same number twice in a row less than 1% of the time). Of course a sufficiently extensive PRNG algorithm can still salvage that, just an issue that popped into mind. In this case you might fail to choose something along the exact same latitude/longitude as often as a uniform distribution would demand.
Yes, absurdly challenging in practice. But nothing in the conditions imposes any limits on the mental capacities required for the procedure, provided they are attainable in theory.
You're right about visualizing something concrete, but for a smooth shape with no "markers" how easy is it to associate numbers, or anything, with envisaged points? The advantage of a globe is that there are lots of named points spread over it, such as cities and landmarks. That makes remembering hordes of coordinate pairs somewhat easier.
The main question was "what you'd actually do" - hence me phrasing it as "gut solutions." Ways to theoretically did it are interesting as well, but if you're at the point of like "something a sufficiently dedicated person could eventually prepare themselves to do" the solution is already liable to not be much better than a lot of simpler "generate random sentences/phrases, turn them into numbers through a sufficiently transformative algorithm, use the numbers." The catch is that's one you could learn to do quickly but most people would start out being very slow at, whereas the coordinate method would probably start out totally unusable.
My quick-and-dirty algorithm for that is, first, pick a random medium-length word. Then pick a second word related to the first, and a third word related to the second but in a different way, to break any contextual link to whatever I need the random number for.
Add the ordinal values of each letter in the resulting word, modulo 10. Which basically means immediately dropping everything but the last decimal place at every step. This gives me one reasonably random digit.
Repeat as needed, and in most context it's not going to really matter how random the digits after the first two are.
Just been thinking about this idly for the past hour, but it seems like the other suggestions might be overly complicating things? I think, in abstract, we just need one sufficiently large list of things that we can generate a bad-random example of, and then an easy-to-calculate function from that to numbers that does not particularly carry over our bad-random biases.
In concrete, I think the perfect example would be coming up with a bad-random date (say, between 1500 to 2000), and then using Conway's doomsday rule (easy mental algorithm) to calculate the day of the week that date was, which gives us a nice number from 0 to 6 which can then be turned into a random bit easily.
There would be two main biases I see in coming up with the random date: Number bias (especially the last two digits of the year, and the day of the month) which is the same sort of bias as we run into when we try to come up with a random number between 1 and 10, and date bias (affecting, probably, mainly the century and month) in which we might e.g. be more inclined to choose centuries closer to either end of the allowed range, or months which come to mind sooner. But as far as I know from my layman's knowledge of calendars, neither of these would particularly bias the day of the week we come up with.
Presumably you can generate numbers that are a little random by thought alone. That is, the numbers have some entropy, just not as much as a truly random sequence. You can then use a randomness extractor (https://en.wikipedia.org/wiki/Randomness_extractor) to get out a sequence that is close to truly random - not just I-can't-believe-it's-not-random like a PRNG.
However, I don't know whether there exists a randomness extractor that you can practically execute in your head.
Related question that has been bugging me: What would be an example of a fun 1-player game that one could play entirely internally without using anything of the external world (so e.g. no cards)? Ideally success wouldn't be open-ended (e.g. "how would I explain X to Y?"), nor would there be too many degrees of freedom of possible "moves" (e.g. solving mathematics puzzles). I am looking for something that feels like a classic game like solitaire or chess but which can be played internally and without another player.
Some things ChatGPT has suggested are mental peg solitaire and a mental knight's tour, but both would require somewhat superhuman memory for the more interesting scale puzzles.
> What would be an example of a fun 1-player game that one could play entirely internally without using anything of the external world
For a long time, when I was bored in a conversation, I would take a random word the person said, then start a new word starting with the last letter of that word, and containing the first 2-3 letters of the starting word in order somewhere in the word. So, for example "repeat" -> treat 2 / trepidation 3, or "role" -> eros 2 / enrolled 3, or "second" -> dose 2 / desecularized 3.
A 2-response is worth 1 point, a 3-response is worth 5 points.
I would then give myself 5 bonus points for working that new word somewhere into my next series of words in the conversation. The highest score that I remember (in what must have been a very long and boring conversation) was 35.
But wait, you say, this requires another person! Actually it doesn't! Because sometimes when I was doing some physical task that was boring, I would then just start chaining them together.
So repeat -> trepidation -> notarized -> denotation -> nondenominational -> linolenic -> cline -> echolocation and on and on it goes!
Now you're basically trying to create the longest chains you can, and the length of the chain is the scored game metric (and you give yourself the 1 or 5 points based on how many words with 2/3 first letters). You learn quickly to avoid words ending in "y" and to like words ending in "g,n,d,e,c,l."
It keeps the mind busy, and keeps your vocabulary sharp, and can be played anytime, anywhere (even during a boring conversation).
I feel like the words "fun" and "game" are kind of load-bearing here in narrowing things down beyond just "see if you can guess/figure out something that you can then slowly check". I could imagine trying to rig mentally playing something like Countdown's numbers game? You'd have to figure out some of the "mentally generate random numbers" to give yourself a seed (usually a 3-digit goal number, then 6 numbers that are a combination of single digits and a few round larger numbers like 25/50/75/100, then you have to add/multiply/subtract/divide between them to get a result as close to the goal as possible. Usual variant doesn't require you to use all the numbers, but you could add it in if so inclined.
It still requires a reasonably wide memory (goal number, your 6 starter digits, alongside the actual math you're doing and which of the 6 digits you already used) but something along those lines is the best I can think of. "Get as close as possible" being the goal means you don't have to choose your goal numbers in some methodical way to select "beatable" puzzles. Ofc if you're purely in your own head you can't check your work beyond just "having fun trying to get close" but honestly I feel like that'd already work?
I'm not sure if you count that as "too much freedom mathematics puzzles" - I feel like having just the 4 possible actions and no pre-existing equation makes it about as lightweight as "puzzles involving numbers" can be computationally, but you might already count that as too much.
Also if you're allowing to use your own hands to keep track of variables that probably opens things up a bunch - I wasn't clear on if you were also limiting stuff to "introspection only" or if your own body is fair game. Then again I could just run my fingers through my hair, get a bunch of black and white ones, and make my own board with different playing pieces :D
Definitely, what I am looking for is not precisely defined. I like your numbers game example and I suppose that it also opens a whole class of possible games consisting of a randomly generated goal state and some restrictive possible moves that are hard enough to see-through to make reaching the goal state challenging. These games are nice and I think in hindsight the essence of the connection between your randomness-generation question and my singleplayer-games question.
Definitely not too much freedom — I restricted math puzzles with things like proof-based puzzles in mind, where the first step could essentially be any English sentence, even defining a totally new mathematical structure.
For philosophical purity I'd say introspection-only is still what I am going for, but really this question originated from when I was a kid without any sort of phone-based distraction, trying to get time to pass in waiting rooms and on trains. So I suppose any bodily movement that wouldn't get stares would satisfy my "original" problem.
I was imagining just using your fingers to help remember stuff - so like if you tried to get 363 using 3 6 7 8 25 75, you could use 6 fingers to help yourself keep track of which of those numbers you already used, and if you start to forget one you'll know which size it is relative to the others (so if you can't remember what your second finger correlates to, it's between 3 and 7 and that hopefully helps jog your memory).
I tried to look away from my screen and run this with just my brain (and fingers helping keep track a bit quicker) and got
8-3 = 5
5 x 75 = 375
375 - 25 = 350
350 + 6 + 7 = 363
and arguably had fun doing it, although definitely felt a bit annoying to keep track of all in my head. That said I imagine it'd get a lot less annoying if I spent an hour amusing myself doing it because my phone battery died in a waiting room or something.
You can multiply the ones digit by 6 and add the 10s digit to get a pseudo-random sequence on the interval [1,59]. All you need is some way to produce a seed number, for which I usually check the time (but you disallowed doing that).
I typically just look at the ones digit of the output, and if I need a bigger number, then I chain it together multiple times. It does have a slight bias away from 0, but it's the best I've found that works with mental math that's easy to do.
Humans are notoriously bad at randomization.
I was showing my girlfriend this thread (since I had brought up this puzzle with her earlier) and her main question was just "why is that Lapras level 50?", I responded with something along the lines of "because that's what's used competitively" and she found this answer sufficient.
I find it really interesting that all the responses so far are *deterministic* (pseudorandom) methods for simulating a uniform distribution, rather than biophysical mechanisms for generating numbers that may be biased but are truly random. To be fair, I guess the former is more relevant for your example framing. But I find the latter question much more interesting.
Can you elaborate on what you mean? I was indeed fishing for a specific kind of "introspection only" solution but I'm curious what exactly you have in mind by "biophysical mechanisms."
Eh yeah I mostly just didn't read the whole comment and my head went in a different direction. If the question was what mental process can produce a *random* bit, rather than a bit that is as close to 20% likely to be 1 as possible regardless of whether it's from a random algorithm, I would argue just "thinking of a random number" does this (or say, think of a random number and then take some bit of its base-2 representation). Because unlike computers, our brains are not designed to minimize the chance that a random physical process will change the result of a calculation; vacuously, they're not designed at all, but also, a neural computation involves both analog and digital steps, and the steps that involve analog summation are allowed to get close to bifurcation thresholds where random events may influence the result.
If what you care about is lack of predictability, though, in practice of course it makes sense to use a pseudorandom algorithm, and because of our shared psychological tendencies there are better ways of seeding it than having everyone think of a number - but I still think it's interesting.
First, pick a random number; this number is how many digits your final random number will have. Then pick a number between one and twelve; this is the hour of the day the other numbers will be referencing. Name things you do in that time. (This may require you to become more interesting.) How many bites did you take at lunch/breakfast/4 in the afternoon? How many socks were in sight? How many letters were in the color of your shirt?
What you'd want is a function that's sensitive to initial conditions, like a password hash. That way even if there's a bias in how you choose your initial seed value, it won't give the same output as someone else unless you pick *exactly* the same seed, which is unlikely.
Unfortunately, I can't think of any functions of this sort that are easy enough to compute in your head.
I'm going to be contrarian and say almost all the suggested methods in the thread are bad, because I think you can be heavily influenced by your unconscious to gravitate to certain lists and phrases, and to certain errors while adding them up, that cause you to end up in a place you'd already unconsciously worked out or at least guessed. Maybe an expert will tell me this isn't possible but I feel like it's happened to me many times, and that it explains a lot of apparent experiences of ESP and other supernatural events.
I propose instead something where you are rigidly bound to use a specific list that cannot be altered on a whim and isn't chosen by vibes, and also has minimal error potential. I read in some game theory material the idea of generating random numbers by "knowing your Shakespeare well". If you go through the plays in a predefined order (when you first read them would do) and by whether there's an odd or even number of speeches in each scene, that would work for a binary number, and can probably be easily transformed for otherwise. Or alternatively use something like the single piece of memorised text you know *most well* (e.g. the Declaration of Independence, the script of Star Wars, whatever it may be) and you're not allowed to use anything else. You must also double and triple check for errors.
That's, honestly, the only method I'd trust for getting close to true subjective randomness.
I think a lot of these methods are true subjective randomness as a one-shot as long as you genuinely have no idea what the error is. (To use the birthday one as an example, if you aren't very early or late in the year, I think the vast majority of us have zero idea where we'd end up, and as long as you continue to have zero idea at the moment you commit to using this as your method, I don't think people born on the 145th day are subconsciously more or less likely to decide to go for this method than people born on the 146th day. The big catch is committing to using it before you let your brain make any progress on figuring out the number it results in)
If I pick a random long word, give it +5 for each vowel and -2 for each consonant, and then if the result ends up ending on a 1 or 2 I consider that a "yes" for 20%, it'll probably indeed be successful at trying to get that one test, since I have zero idea which direction it biases and I'm pretty sure across most such concepts people could make up we'd indeed get 20% pretty much exactly 20% of the time.
But if I tried to use such a method repeatedly it'd definitely skew, and as soon as I've started *noticing* that it skews, whether or not I decide to change methods or stick to it is now a massive further biasing element. The "rigidly bind", as you phrased it, part is important - I think you can pick whatever you want, but you have to decide on what it is and stick to it before you've given yourself a chance to know what sort of results it gives, because otherwise you're liable to think "eh, I'm supposed to have an 80% not to do it, but the last two times I did this method I rolled a 20%, so maybe I should try to change methods..." and that itself will start corrupting things.
But yeah I think there's a huge difference between making up a method to get a 20% chance once, where you can be pretty sure you're liable to err in either direction and it'll cancel out, and a method you can re-use to keep getting random numbers either indefinitely or at least up to a pretty large finite number, where what seed you keep feeding into it needs to itself be human-thought-pattern-proof, which is a lot harder and requires you to either use a large already-known base of material you calculate something about, or have an algorithm so transformative and sensitive that even if you end up with biased patterns in terms of what you feed into it, it'll still come out pretty much random.
I have a bunch of digits of pi memorized, so I pick a number between 4 and 30 and if that digit is even or odd, it gives me one bit of randomness. Not perfect, but it works. Likely would not work for this scenario, because people are predisposed to pick numbers like seven when choosing randomly, so the seventh digit of pi would be over-weighted.
I only have 3.1415926 and this was actually one of the options that popped into mind for me early on as a useful one-shot. If I think of something like "the 6th digit of pi" randomly I'd have to stop and think.
But as you said, people trying to give random numbers directly tend to bias towards certain ones, so it's always "riskier" to use numbers as your seed and generally means you need the algorithm transforming them into a result to carry more weight (like producing and processing a bunch of bits to up the volatility).
Now I'm wondering the degree to which we can fix the underlying problem of human-generated random numbers not being random. We have great statistical tests for whether or not a series of numbers truly is random, so theoretically there's something to optimize against. I might have Claude put together a website where you choose individual digits (or coin flip results) and then it runs tests to tell you what's not random about them - too even, too clustered, not enough repeats. I really wonder if this is a skill that you could improve with practice.
I'm trying to think of where this would be useful, I guess really just for people doing something competitive and mind-gamey where being unpredictable is beneficial? Even then though they usually wouldn't be thinking of it in terms of "random numbers" persay and would often want to be skewing results around their opponent's patterns (i.e. any elaborate system of counters and counter-counters that comes down to rock-paper-scissors).
Being unpredictable isn't really the same thing as trying to get a perfect uniform distribution, especially if you can just show up to each competition with a differently-slightly-biased seed in mind.
But just on a "isn't it neat" level, yeah, I really do wonder if a person could learn to just keep giving fully random 2-digit numbers or something without running through an algorithm or anything. I'd actually probably guess the answer is "no" but that just makes me more curious.
You can't generate randomness "from first principles", because then others could do the same, so you do need some kind of seed value that is hard to figure out for an attacker. Once you have such a value, you put it into any of the modern pseudo-RNG algorithms, and out comes a value that is practically indistinguishable from true randomness.
> You can't generate randomness "from first principles", because then others could do the same, so you do need some kind of seed value that is hard to figure out for an attacker.
I don't understand why others being able to do the same breaks the value of being able to do it yourself, could you elaborate?
Like if it's a game theory optimization thing, then if both of you do it, it's just optimal play against optimal play. So the better anyone could get to "truly random" internally, the closer to optimal play, but it's very likely methods to do it better rely on actually better / faster computation or short term memory buffer size or something, so only some people will be able to do it.
I was going for a proof by contradiction. If there was an algorithm one could derive to generate random numbers but without a source of entropy (like a secret seed), then the numbers wouldn't be random.
Ummm. The numbers would be deterministic, but they could still pass all the tests for statistical randomness. That's what pseudorandom number generators do. The same seed will give you the same progression of numbers. But they could still pass all the statistical tests of randomness that we throw at it — e.g., each bit is 50% likely to be 0 or 1, no obvious repeating patterns, correlations are extremely small, and so on.
Why do you explain PRNGs to me? I know how they work. What I'm saying is that if there is no seed, it's not a PRNG.
I thought you were saying that if there was a secret seed the numbers wouldn't be random. I must have misunderstood you. Apologies.
Yeah, we're just thinking of different use cases then. You're thinking of encryption, and I'm thinking of things like poker or other games, where making random decisions sometimes is part of optimal play. But people are never trying to derive encryption algorithms in day to day life using only their brains, I don't think (or I certainly hope not, we're well past the days of Caesar ciphers being relevant).
It's not really different. RNGs and cryptography are closely related. You need good RNGs (and therefore good, secret seeds) for cryptography.
I think most/all of us would recognize that concept of "you need a seed value and an algorithm", if not phrasing it exactly as such, the big question is what you actually use to do this with only your thoughts without biasing the result. I'm somewhat encouraged by seeing that everyone seems to be drifting the same direction my solutions did and I didn't miss anything obvious.
The secrecy of the seed is more important than its actual value. As long as the seed remains secret, the attacker can only brute-force it which is easy to defend against. Any short phrase is sufficient even against state actors.
Pick an arbitrary phrase or song line of at least 20 characters, convert each letter to a number (A=1 to Z=26) and add them up (you really need to only keep track of the first digit). If the result is divisible by 5, push the button.
Considering the purpose of the task, you don't have to worry about making a mistake, so just pick a very long phrase and do it sloppily.
The "you don't have to worry about making a mistake" thing was something I considered too - as long as your errors are liable to be in either direction, it's still fulfilling its purpose.
The direction of the error doesn't matter. You're spinning a virtual five-sector spinner.
I meant the direction in the sense of "flip a no-to-yes or yes-to-no" - like if your errors disproportionately turned 5's and 3's into 4's then you'd end up skewing the resulting 20%.
For the record my "preferred" solution is quite similar to yours because it feels the most reusable. I think the "lightweight" version is using letter/syllable counts in long-ish words/sentences, but that'd eventually start running into patterns. I'm pretty sure turning the letters into numbers as you have does ensure it'll be pretty much exactly 20% and fully reusable, if slow-ish to execute (tbf I imagine you'd get a lot faster if you had to do so repeatedly for some reason).
A fairly large but finite list of things you can name, without knowing in advance how many - UN member states, say - and then some process for converting that number into whatever your random number parameters are.
Perhaps this is your 'things you don't know until you look' idea.
This year the book review entries are listed at written by Scott Alexander. I believe last year they were listed as by 'a reader.'
This doesn't really matter for people who read the posts, as there's a pretty clear disclaimer at the top, but when you restack a quote, Substack attributes it to Scott Alexander.
I'm not sure if Scott minds, but I figured I'd bring it up.
Haha, what actually happened is that an administrative assistant of mine who literally goes by the handle "a_reader" was the person who posted them last year. This year I'm posting them personally. I'll see if there's a way to fix it without making her post it from her handle.
Ahh.
That’s a good username to have.
I'm still sort of shocked by the apparent absence of impactful indie games or movies that were made with significant AI assistance. The technical requirements to make something truly enthralling have fallen steeply, but it seems like people's executive functioning has been steadily melting in inverse proportion. Or maybe I just can't find them in the overflowing slop piles.
I just had a fun AI-assisted game idea.
Imagine a low-tech-looking RPG, maybe it looks a bit like Stardew Valley. You walk up to the first NPC and get a traditional dialogue tree type interface, you can choose about five different things to say and get five different responses, which lead to more things to say and more responses. But then you realise that this dialogue tree just keeps going, and going, and going, branching off in millions of different directions.
It's not generating them on the fly, which is would be too expensive, it's just got these vast dialogue trees pre-rendered if you will. Millions, tens of millions of possible questions and answers for each character, all consistent with that character's worldview. A combination of human and automated quality control on each character, to ensure they don't go round contradicting themselves in important ways.
The game, then, is about navigating these vast dialogue trees to find out what each character knows. I'm thinking that a good setting would be an insane asylum, where you have a lot of characters with very different personalities and nothing to do except talk to each other.
I've been working on a hobby game for the last few months and Claude Code has been very helpful. But there's still a ton of work that goes into playtesting and tuning and design and arranging art assets (though Claude can sometimes help with that too) and so on. Progress is faster than it'd be without Claude, but it's still very slow.
>The technical requirements to make something truly enthralling have fallen steeply
Four comments, all related to this statement in different ways.
1) As written, that's just plain wrong. No, they haven't. Note how strong your statement is. You're not saying "can be made with significant AI assistance" there. You're saying "can be made much easier with AI assistance". No, it can't. AI can help a skilled artisan with tools and know-how. It can't replace him. Not now, not in foreseeable future. There's too many things it just can't do, or can't do well enough, and the things it can do well still ultimately require the "traditional" set of skills to put together.
2) The audience requirements to allow themselves be truly enthralled have risen sharply. Go 5 years back in time, take contemporary tools with you, and there'd be plenty of things you could amaze and awe people with. Sure. Those specific things have been noticed, utilized, and exploited to death. They're considered slop now, something to be avoided. (Evoking the usual cries of "moving the goalposts". Technically correct, even. Still, if you've been noticed, you've lost, and it's hard, and getting increasingly harder, not to get noticed. You either embrace the medium and drop the pretenses, or shrug your shoulders and go back to regular craft.)
3) One possible exception - coding. It's probably already possible for contemporary tools to replace a programmer up to a certain skill level entirely, and the replacement is probably slowly happening. You're not seeing it partly because it isn't noticeable or advertised (see #2). But you're also not seeing speed-up and efflorescence, right? That's because the skill requirement hasn't been eliminated, it merely changed. So far, best AI coders are simply skilled coders. Even if contemporary tools could ultimately allow you to match skilled coders, they still require skills to use. Different, new, dealing-with-AI skills, which take time to develop, and which we're still only beginning to develop.
4) Everything aside, we've already had ways to lower technical requirements before AI, and we used them. Libraries, engines, etc. All mature technologies that already do a good job sparing you from most of the technical details and boring grind. Perhaps there's simply not that much left to meaningfully simplify?
Movie directors have been making great art by organizing other people's labor around their specific vision since the dawn of cinema. They aren't skilled artisans themselves, they bring those people into their service. I am surprised that we haven't seen more great work from visionaries who act in a directorial position and direct the various AIs to produce artisanal quality art & video & code... often filmmakers like Scorsese see that there is this potential latent within the technology, but I suppose people are bottlenecked at the level of vision, taste, probity. Which is a disappointing, I hoped at least one of those ideas guys was legit but they've all been unmasked as frauds.
My hunch: regardless of their value as productivity boosters, AI tools tend to be particularly appealing to developers who make generic boring games and unappealing to developers who make impactful games.
A common pattern to many developers of iconic indie games is how important the process of creation is to them. Some are fiercely independent coders who would rather stretch the development for years rather than take in collaborators. Some let the process itself grow the work into unexpected directions. Some are multi-talented folks who also create the graphics and music themselves, which gives their work a signature style. In general, they seem like the sort of people who see the journey at least as valuable as the destination and would not take shortcuts at the price of compromising their vision.
Meanwhile, the kind of person who's most likely to see AI as a godsend is the Idea Guy - a person who believes that a million dollar idea emerges fully formed from the mind of an auteur, and that execution is merely a long chore on the way to a finished product. Their ideas are often stuff like "a blend of those two franchises I like" or "my favorite game but without my pet peeve", which means that even if it's theoretically possible to use AI in a creative way, they probably won't.
I think the technical requirements to make a good, fun game weren't ever that steep, it's the game design that is a black art that neither AI nor anyone could help you with.
Disagree, though we'll have to wait a bit to see if I'm right or wrong. Point AI at 'Overwhelmingly Positive' reviewed games, reverse-engineer the general designs of the games as well as summarize the positive feedback, and I suspect 90% of the good elements would be cookie-cutter capabilities.
Hasn't worked for writing. LLMs still can't write a truly superb essay, even though that should be their central skill, and that's the same in video games, where it's basically everything or nothing.
What do you think large studios and design-by-committees games were doing? Plenty of people pointed at Doom/WoW/PUBG/Disco Elysium/Whatever game had a huge impact at a given point in time.
In the best cases, one or a handful of these clones succeeded (and in some cases replaced the original). In some, none survived, because video games, like a lot of arts, works like a barberpole of taste: what is hype one day will be rote and overdone the next.
Forgive me for descending to the object level, but are there any good Disco Elysium-likes yet?
Esoteric Ebb, as Francis Irving wrote, but also Zero Parade, Aether & Iron, Rue Valley, Sovereign Syndicate, and, if we extend a tiny bit the definition, Citizen Sleeper. Plus a number of announced yet unreleased projects, including some by members of the disco elysium team.
Thanks (to both of you)! I've been meaning to try Zero Parades, which seems to have been pretty well received if I allow for the backlash against the studio. So I'll start by checking out that and Esoteric Ebb.
Apparently Esoteric Ebb has lots of commonalities (internal voices, no mechanical combat) and is good.
To you and Carlos, fair enough. Maybe it won't produce superb games, but I think it can significantly raise the bar and make a bunch of the cookie-cutter effort to get there much easier.
How many great games were made that forgot simple ease-of-use features (which I see complaints about all the time on 'very positive'-rated games). Even great games often are lacking these in the early days, whether due to oversight or time crunch.
AI code, AI textures, AI 2D backgrounds are probably everywhere. AI writing and foreground AI art are still too obvious and provoke an immediate "slop" reaction.
One gamedev niche where AI impact is visible is adult videogames. People still hate AI writing and AI art, but AI-assisted animations are quickly becoming the norm. Rigging an animated scene is a hard and complex skill. The best animators can produce terrific animations, but your average developer is more likely to make "mannequins on rails" than anything resembling intercourse.
The AI can take a still image and a prompt and turn it into an animation that has inertia, soft body physics, full inverse kinematics, you name it. It's not perfect, but already better than 90% of all handcrafted animations.
One significant factor is to remember the lead time needed to develop (not just make, but come up with the idea for, flesh out the concept for, find investors for, etc) an 'impactful' game.
AI generation to make something impactful may (debatable) be there today, but it wasn't 5 years ago, and I'd be surprised if many 'impactful' projects go from conception to release in less than that time frame.
I agree with the impression that we aren't seeing any masterpieces or even notable works made using AI, with disclaimers. One disclaimer is that goodness only knows how many indie games now have AI-written code in them. We can only guess.
On the movie front: I hear some talk about Young Washington https://www.imdb.com/title/tt32104007/ using AI for some special effects, and also getting some praise. I've never seen it, but it seems like a very patriotic political scissor-movie, that some people will love on principle and some people will hate on principle. Also, I don't think it uses AI much.
It's not exactly a mainstream movie masterpiece, but there's an all-AI Minecraft youtube series https://www.youtube.com/@LatentDiffusion that seems to be pretty popular and well-liked. Not great cinema, but I'm watching it and it's not too bad. I don't think you will enjoy it if you never tried a game, but lots of kids who did seem to.
With these AI-made movie attempts, what I can't help but notice is that often the human is the weak link. The movies or shorts fail because the story isn't there, or because editing is incompetent, or a person doesn't understand how the movie language works. The AI can make very cool looking shots, and can now even keep them consistent, but someone has to write that script, or prepare that storyboard, and yikes. Often it seems that the creator was counting on AI doing all of the heavy lifting and didn't this they themselves would need to do much. They would fail just as badly if you gave them $300mil and a full control over a studio.
Yes, practically every established creator with skills, talent and an audience would rather die than openly associate themselves with AI right now, on principle or out of fear of backlash or what have you.
If we were living in a world where AI is permanently frozen at the 27-07-2026 level, I'd expect that some of the kids currently incompetently using AI video generation for making shorts would develop their skill, learn something about filmmaking and begin creating better things, and some would create excellent films. Eventually there would emerge some Orson Wells of AI. But we are not going to be living in that world.
I think AI is in the original position of video games (they were made for children and nakedly to make money -> artists ignored them -> impression that games had no artistic potential reinforced) but at a much more accelerated scale, where the people using it have such terrible taste and are shotgunning slop at such volume that it seems like an actual plague upon (and existential threat to) real creatives, and so to even call yourself an artist you have to reject it out of hand, at least publically.
I think it's possible that there have been masterpieces produced with the help of AI, but the incentives push ruthlessly against disclosure.
Pretty much. What frustrates me, is seeing various pieces of AI art enjoying success and being positively received... when published by the dishonest cheaters who present it as human work. Imagine a wonderful new screwdriver design far more useful than any other... that only works in the hands of the lying scoundrels and never in the hands of a honest craftsman. And yet, with all the knee-jerk hate, I can understand the cheaters.
People often love it! https://old.reddit.com/r/ArtPorn/comments/1na34m0/ivan_aivazovsky_sunset_on_the_sea_1848_736x977/ But only when they don't know.
Agreed. Are today's younger game developers so thoroughly anti-AI that they entirely refuse to use it or work for any studio that tries? Even at the loss of their own productivity and pay?
As an indie game designer (but not a programmer myself) I can say most of the guys I know are looking into it. I've spoken to a bunch of other devs in completely unrelated contexts (so this isn't a shared pro-AI community) and some of them already swear by how much time it saves, others aren't using it yet but are planning future projects around engines/languages they know AI excels at, and the one guy who I know refuses to use it for code has still extensively EXPERIMENTED with using it for code and his refusal is just based in happening to be working with some rarer languages that AI often hallucinates as having properties similar to better known languages and other minutiae that mean he doesn't feel it's worth the bother for current projects that are already pretty far along.
That said, multiple of them have also expressed a serious concern with how much anti-AI sentiment there is and transparency requirements on use of AI, but nevertheless I think a lot of them plan to push forward with it. I think this might just be the moment in time we're at - it took longer for widespread LLM coding use to trickle down from more pure tech companies/circles into random stray indie devs who were likely too absorbed in their work to keep up with rapid advances (seriously, half my functionality as a designer is I have enough free time to actually PLAY GAMES rather than being so trapped in the code mines that I lose track of the state of the industry). But I imagine in a year this'll look pretty different. Takes a while for a project to come to fruition even then, and for many devs it's something they're doing initial exploration for on the side atm rather than bringing it into existing projects. I think this is just the lag time between "people starting to use them" and "their projects being in a sufficient state that they're widely available."
AI coding is shockingly recent. The first coding agent that was actually good enough to be useful was only like 9 months ago (Opus 4.5) and it was only 3-6 months ago that they got really good. I can certainly imagine that it would take a lot of time for this tech to diffuse.
If I had to guess, everyone (or almost) is silently using it to help coding and prototyping, but the push-back on game assets and documentation is too visible and visceral, so everyone is avoiding those like the plague.