quiet_NaN
3hr ago
It seems to me that there's really only two possible paths forward; either AI remains jagged in capability like current LLM's and the standard economic arguments about technology hold, or we develop an AGI that represents a perfect labor substitute
Suppose for a minute that today's models will hit a wall of zero marginal returns tomorrow. This would not mean that AI agents would not still get better. After all, it seems unlikely that we have already figured out the best way an agent should split a problem into different subproblems, for example. Given that overhang, it is not obvious to me that the median office worker will still be able to earn a living using their brain in the equilibrium state.
Sure, in the long run, an AGI might prefer something more reliable than biodrones, but that might take a decade to build at scale. If you build robots, you have long, complex supply chains which will take time to fully automate and scale up (at least for an AGI which is only slightly smarter than humans are). By contrast, knowledge workers are easily replaced, once your LLM can do the job, you can spin up a zillion instances. Also, hitting the wall will mean that we will have tons of GPUs which can be bought for pennies on the dollar from the companies which were betting on FOOM.
Of course I could also be wrong and LLMs could always remain subpar compared to the median human in certain relevant intellectual skillsets. Or I could be wrong and we will get FOOM and be all turned into paperclips.
quiet_NaN
10hr ago
I think that in the context of Trump's SAVE act, it popped up that people can -- and might have to -- get a birth certificate with their current legal name on it. I also think that some countries use up-to-date birth certificates to track marriage status.
If birth certificates are updated regularly to reflect changes in the life circumstances of a person, rather than being stored on the blockchain shortly after birth, then it makes sense to also update them to reflect cosmetic changes in things like first name or gender identity.
quiet_NaN
10hr ago
Of course, the Register did a thing where they did not even refer to him by name. e.g. "Florida man insists he didn't violate the law by keeping Top Secret docs". Possibly the only way to report on him without making him stronger.
quiet_NaN
11hr ago
German here. I have a Dr. rer. nat., but don't really identify as it. In the course of earning one, you typically get disabused of any notion that they signify elite human capital. STEM is full of jokes to the tune of 'Oh, you have a PhD? Don't worry, I will speak slowly then.'
When I was perhaps eight and playing some outside, I corrected a kid referring to my father as 'Herr $lastname' to 'Doktor $lastname'. That did earn me quite the talk.
There is a cliche of lower class people calling their physician 'Herr Doktor' or 'Frau Doktor' (which is especially funny given that what you need for a Dr. med. would not even earn you a Bachelor of Science), but the upper middle class prefers more subtle class signifiers.
quiet_NaN
11hr ago
However, the demand that other people refer to you with a specific designation is not really a natural right, and in fact, suppressing or compelling the speech of others is a violation of other people's rights to free speech.
So forcing the German Jews to adopt the name Israel or Sara on legal documents was not a violation of their rights? If some racist jerk wants to call everyone he considers Black 'Nigger $lastname' instead of 'Mr. $lastname', or if a state mandates this, that is all just fine?
You can not compel people to really treat you as your identified gender any more than you can compel them to treat you as one of the cool kids. If a bearded person in a dress complains that none of the guys at the bars are buying them drinks, that is not really actionable.
I think there is no reason to even track the gender or sex on driving licences or in DMV databases. Outside Kansas, people are generally not driving with their dicks.
I think the US is generally rather accommodating with name changes. If you do not like the name your parents gave you, you can change it. The government is generally not going to say 'you were assigned Kevin at birth, you will never be a Benjamin'. But here the government of Kansas is saying 'all of you who have changed their name to Benjamin, all your identity documents are invalid effective immediately. Get new documents which say Kevin.'
This is basically 'your passports are invalid until you get the J stamp', the state unreasonably punishing an outgroup for partisan reasons.
To the extent that I would have a problem with the current state of affairs, I would find that the entire licensing regime that the government imposes on the people -- forcing them to register and pay fees in order to drive and participate in society -- is the actual problem here, not merely an unpreferred gender marker.
Making driving a car an inalienable right would have large negative externalities. Of course, the libertarian approach would be that what qualifications you need is between you and your liability insurer.
By contrast, for all the moral panic about trans people from the Republicans, the state not caring about your gender identity matching your sex assigned at birth will not have such negative externalities. Nobody is forcing anyone to suck trans cocks. As a straight guy, I can spend weeks without thinking about the existence of dickgirls at all, something which MAGA seems completely unable to achieve.
I am also doubtful that for all the CW-ness of transness, it will be a vote winner for either side. Most people are not trans, nor do they frequently suffer from their tinder dates having unexpected genitals or losing to bearded people in athletic competitions. When the SJ left campaigned on trans, they mostly lost badly, but not because Americans hated trans people, but because they were apathetic -- "here I am stuck trying to make ends meet, and you want me to care about the plight of some sexual deviant". I have high hopes that the reaction in 2026 will be similar: "grocery prices are through the roof, and the MAGA elites want to tell me that forcing some Kansas trannies to get new driving licences is a win for the little man somehow".
quiet_NaN
13hr ago
Technology has always replaced jobs, thats how it always goes. New jobs will arise.
I would argue that this time, it is different from the industrialization or the computer revolution.
The computer revolution was the first time the machines came for stuff which had previously required intelligence. In the niches where they were good, they totally crushed humans. Before electronics, computer was a human job. Today, I can waste more multiplications on playing a video game for an hour than humanity solved in total in 1900.
On the other hand, electronics also came with very sharp limitations. A human who might have worked as a computer in 1900 still had skills which the machines did not have, and could thus be running Excel in 1995.
This time around, it is much less clear that the median human will still have any intellectual comparative advantage over the machines. Heck, even the median MINT PhD might not find employment for their brain in 2035 any more than anyone found employment for their multiplication ability in 2000.
So your "new jobs" which will arise might well being the biodrones of an AI: wear AR goggles and simply follow instructions. Walk to the indicated rack. Unplug the indicated network cable. Plug it back in at the indicated port. Drink exactly 50ml to avoid failure from dehydration without requiring more than the minimum of bathroom breaks. An exciting day at work for the most qualified biodrones might be when they were used to replace the CPU in a machine.
quiet_NaN
13hr ago
I think you can trivially make an LLM deterministic in the technical, narrow sense that for exactly the same input you get exactly the same output. Just initialize the pseudo-random number generator deterministically.
However, where LLMs differ from most classical deterministic algorithms is that they are not stable, a small change in the input might result in a big change in the output.
Suppose I have a list of strings I want to sort lexicographically. If I use std::sort (and stick to ASCII), I can expect to get reasonable results every single time. If instead I give the task to a neural network, such as a human, I will get some significantly non-zero error rate. If I use an LLM, I would also expect an elevated error rate. Of course, both the LLM and the human might also refuse to work with certain strings, e.g. racial slurs.
Generally, nobody uses neural networks to solve problems which are easily solvable by classical algorithms, teaching aside. But there are a lot of problems where we do not have nice classical algorithms, such as safely driving a car through the city or translating a text or building a website from informal specifications. So we accept the possibility of failure and hand them out to LLMs or grad students.
quiet_NaN
4d ago
I am not sure this is a good top level CW post. In large parts it is basically the format of Scott's link posts, each line with a link and a sentence or three of hot takes.
It is fascinating to see how something that was absolute NO in traditional rules of war "Generals do not take pot shots at each other" became normalized in the rules based order.
{{Citation needed}}. I will grant you that in medieval times, people were more likely to kidnap and ransom a noble where they would just have stabbed a commoner to death. But even in WW1, flag officers were killed quite frequently -- shells do not discriminate, after all.
I think that assassinating generals is probably the most humane way to wage war. After all, a general is much more costly to train than a squad of infantry, so it causes the maximum of monetary damage for the minimum of human suffering (apart from shooting down a fighter jet, perhaps).
It would however not work as a broader strategy in Western countries like the US. My estimate is that if you managed to magically kill the top 1000 US military officials, the effectiveness of the US forces would perhaps drop by a few percents, because the US has no shortage of people who are both competent and loyal.
Contrast this with an autocratic regime like Iran. Military coups are a real threat in such countries, so successful dictators engage in coup-proofing. You want someone who is loyal to you personally while also being competent. It is a dynamic not unlike that of vassalage (as seen in the Crusader Kings series, for example): you need to appoint a noble to manage some fiefdom you conquered, but how do you make sure he won't stab you in the back at the first opportunity? Often you pick someone who is family or has married into your family, or perhaps a childhood friend. Or at least a protege who is known to be in your favor.
The Iranian army will probably have plenty of people who are competent to lead them. It is much less certain how many they have whom the Ayatollah would trust with leading the army, though.
3/ Yet more Middle Eastern issues
Israeli ultra-orthodox revived ancient European tradition of burning cats and dogs alive as part of celebration
Very based and trad pilled.
Per your link:
Liani first learned about this phenomenon – an unexplained act of abuse popular among teenagers – when she was in sixth grade in Ramat Gan.
The way your source it describes it sounds more like a social media fad than an ancient tradition revived by the ultraorthodox. Why would the orthodox revive a cruel spectacle which was popular with 18th century's gentiles? Why not accuse the Jews of murdering Christian babies while you are at it? This seems especially pointless as Nethanyahu's idea of peace in Gaza remains clearly visible, the ultra-orthodox are quite bad without them burning puppies and kittens alive for religious reasons.
quiet_NaN
4d ago
And something like LLMs with automated theorem provers seem incredibly well-suited to potentially get us toward something like this.
This would have been my suggestion as well. If an LLM can produce mathematics on a PhD student level, then surely it can also formalize that to the point where it can be verified by a theorem verifier.
So you can run them in tandem: an unreliable LLM prone to hallucination, but somewhat creative, and a deterministic small verifier with a small code base.
That it is much easier to verify a result than to come up with it is a pretty unique property of mathematics (though certain analogues exist in CS). Contrast with experimental particle physics: there is most emphatically no verifier with a small code base which can test if a given data analysis is sound or unsound (which is foten a bit of a judgement call, in any case).
I think alignment might be easier if we focused solely on proof generating AIs. Of course, even then it is not impossible that an ASI might create proofs which contain infohazards which will cause humans to set it free, but an ASI would have to be a lot more powerful to deduce how to hack humans just from knowing what kinds of math they have invented instead of being literally trained with the accumulated knowledge of mankind.
Sadly, this is not where the money is expected to be, so we won't do that.
quiet_NaN
5d ago
The Start Menu, and searching within it, is far and away superior to the way macOS handles applications (and Linux splits the difference and fails at both; both KDE and Gnome suffer from this, though in different ways).
The superior way to start an application is to type the name of the binary, optionally followed by a space and arguments, optionally followed by an ampersand, followed by the enter key.
I have about 4k different programs in /usr/bin/. Menus are tolerable if there are a few options to pick, like at the ATM: Do you want to withdraw money, see your balance, recharge a prepaid card or quit? I certainly do not want to specify twelve bits using some GUI. Yes, keyboard searching might make that more tolerable, but can only hope to approach the comfort of the command line interface. (I should mention that I am not some purist, I think that it is fine to use a GUI and mouse for things which map very well upon a concept of a 2d surface, such as vector graphics, CAD or first person shooters. But 'pick a program to run' is not one of the problems which has an intrinsic 2d representation.)
Apart from that, judging operating systems by their user interface is a bit like judging a motor vehicle by its infotainment system: sure, it is relevant, if the navigation system is too painful that is bad. But at the end of the day, most vehicles are not picked for their infotainment system, but for a mixture of other factors such as signaling, price, capabilities, TCO and so on.
quiet_NaN
5d ago
I think this is a good idea. It's not like many AAA games are acclaimed for their dialogue, characters and writing, people literally joke about how crap their writing is. Let people have conversations with in-game characters, why not?
I think an your typical AAA game needs LLM-powered NPCs as much as a drowning man needs a rock. If nobody thought to give the NPCs more dialogue, filling the gaps with AI slop is not going to help.
I think an LLM might substitute for a mediocre DM in an RPG, though. Certainly in text-based formats, but possibly also in something with graphics (e.g. Neverwinter Nights). The benefit would be that it could accommodate player character ideas. So rather than saying "You can not play a lycrantrophic half-elf changeling", it would modify the setting. Perhaps figure out how the fey fit into the cosmology and the overall plot. Invent relevant side quests, just like a human DM would.
The problem with this approach is that presently, if I have to pick between a pre-generated character with a questline written by humans (BG3) and a character of my own invention with quests written by AI, then I would much rather stick to BG3. Likewise, even if I were totally into dinosaurs, it seems highly unlikely that I would enjoy a version of Tolkien's epos where all the non-hominid animals (horses, ponies, eagles, black wings, dragons, spiders, etc) are replaced by appropriate dinos better than the original, simply because AI is nowhere good enough to write something like LotR from the scratch.
quiet_NaN
5d ago
that homosexual transsexuals or HSTS and autogynephilic transsexuals or AGP constituted two clearly defined, vastly different populations of males who identified with womanhood or female-ness.
While it is probably not intentional, the term homosexual transsexual would mean different things to different people (because some would consider a trans-woman who is into women homosexual (e.g. lesbian), while Blanchard considers the trans-woman who is into men the HSTS), and is thus probably best avoided. I am open to formulations which are less clunky than 'transwomen who are into women'.
I find it hard to believe that these transwomen are particularly interested in lesbian relationships with ciswomen.
I think that there are some trans women who want titties so that (more) men will want to fuck them (which includes your Thais), and some trans women who love titties so much that they want their own. The latter might ideally want a ciswomen partner, but might find that few women are attracted both to tits and dicks. I imagine trans for trans is more of a pragmatic strategy in the absence of interested ciswomen. Of course, the ones who are into men don't have this problem because men as a collective will pretty much fuck anything with a pulse.
quiet_NaN
6d ago
Okay, not a constitutional rule.
But I suppose it might actually be against WTO rules.
Basically, any country which makes use of retroactive tariffs is not someone you would want to do business with. Nothing stopping them from nationalizing your company by retroactively applying a 500% tariff on all of the goods you have been importing in the last decade. This would not be much better or worse than just deciding to nationalize your company outright. Investors and creditors tend to hate such things.
Now, in this particular case, the retroactive tariffs would merely replace the unconstitutional ones. So one might frame it as "dear Mr foreigner, we had a minor judicial hiccup with the precise process of how to enact the tariffs, but don't worry, the amount on your bill is still correct".
Of course, the situation might be more accurately described as a chief of police deciding that local shops will need to pay for protection, and when a court says that this is not actually how things work, the city council instead tries to reframe the protection money as taxes so that they can keep it.
But if Congress will not authorize Trump's tariffs (which I believe and hope they won't), then all of that is moot.
A good indicator would be what a claim against the US gov for bogus tariffs is worth on the market at the moment, but it seems that these are not publically traded.
quiet_NaN
7d ago
This would still leave him with having to pay back some 200 billion dollars, I think.
And nobody would assume that SCOTUS would let get Trump away with his next harebrained tariff scheme, so the money companies pay in tariffs might just be considered a credit extended to the US.
quiet_NaN
7d ago
I seem to recall the argument for why they didn't need to grant a stay on the tariffs was that resolving them after would be easy. Did something change in the meantime or was that claim a lie? But even more so the argument doesn't make much sense to begin with "they shouldn't have to return stolen money because it would be difficult" just encourages stealing more money.
Not having stopped this with a temporary injunction is a total fuckup on the SCOTUS part.
And of course any tariffs paid will have to be paid back, but the damage to the economy is already done.
There is a simple solution to this, President Trump could try to get his tariff agenda passed in Congress.
Of course, even that would not save him from having to pay back the tariffs people already paid, because anything else would be retroactive.
quiet_NaN
8d ago
I think the "strong men versus weak men" argument works when it's the steppe nomads versus farmers.
I do not remember the Great Plains nomads winning decidedly against the European settlers. I think the one place where it really works is if the steppe nomads have horse archers (and the farmers don't have guns), because horse archers are just overpowered as fuck.
To raise a powerful army out of a given square kilometer of land, you generally want to use it for farming and not for pastoralism. Sure, the peasants raised on grain might have nutritional deficiencies which your herders would not have, but they will also outnumber your herders ten to one or something, and can resort to permanent fortifications.
If the terrain was suitable for farming, the farmers generally out-competed the nomads and pushed them back to marginal lands where agriculture was not an option. These farming communities eventually gave rise to the first civilizations in places like Mesopotamia or Egypt. I think most durable empires were built on the back of farmers, not herders.
quiet_NaN
8d ago
There's a weakman that does, but the ironman is Secular Cycles; approximately every 300 years (maybe less more recently), regional hegemons have periods of weakness due to internal conflict*, which may or may not utterly destroy them. Rome had roughly four cycles (the Kingdom, Republic, Principate and Dominate) and it pulled through the first three crises (including the civil wars of the Triumvirates at the end of the Republic, and the Crisis of the Third Century at the end of the Principate) bruised but intact.
This seems roughly orthogonal to the meme image which Deveraux is arguing about, except that both predict cycles.
If your argument is "well, if you substitute 'hard times' with 'hard times (but not in Afghanistan, so it is more complicated)', 'strong men' with 'straightforward, prosocial behavior', and 'weak men' with 'evil men', or even 'Moloch', it seems true enough", it does not seem we have a factual disagreement.
I believe that there are people around who take the hard men, weak men part a lot more literal though. If Pete Hegseth wants his staff officers to focus on their individual physical fitness, that seems a more central reading of the meme than yours.
quiet_NaN
8d ago
The Central Park Five were guilty.
Not of the crime of which they were convicted, but that is beside the point. And again, I brought this up as a point in favor of him having serious beliefs. Paying for these ads was not a good business decision, after all.
I just do not think this is a serious belief you can actually credibly defend. Maybe it sounds nice as some kind of slapdash pubtalk barcrawl locker room talk. But do you really, honestly, earnestly, believe that Trump is best modeled as a kind of void whose words bear no relation to anything whatsoever? Not just that he lies, or even that he lies more than other politicians. But that for Trump "words have no relationship to anything in physical reality"? What does that do to your view of the world?
There is an ancient joke which goes "What is the difference between a used car salesman and a computer salesman?" - "A used car salesman knows when he is lying."
Trump is the computer salesman. A bullshitter rather than a con artist or schemer.
Take the 2020 election results. Now, it might be that his thought process was 'Conceding defeat to Biden will critically weaken my brand. I better lie about the election being stolen to keep my base.'
But that is not my take of his behavior. I think it was genuine cognitive dissonance on his part. "I know that I am a great president, much better than Obama, who won reelection. Therefore, it is impossible that I lost. Thus, the Democrats must have cheated." It is not rare in humans, but it is rare to see it so openly expressed in politicians.
What definitely did not happen was that Trump carefully summed up reported irregularities and decided that they made the election close enough to demand a recount.
I do not think that even Trump's followers believe that his claims are literally true. "Oh, I invested my life savings in a Ukrainian company, because Trump promised that if elected he would settle the conflict before he even became president" is not something which happened a lot. "Stop the steal" might be more correctly understood as "Boo to Biden", though the J6 crowd was mistaking it for a claim on physical reality in some parts.
In the words of Hannah Arendt:
The ideal subject of totalitarian rule is not the convinced Nazi or the convinced Communist, but people for whom the distinction between fact and fiction (i.e., the reality of experience) and the distinction between the true and the false (i.e., the standards of thought) no longer exist.
Are the immigrants eating our cats? Is Denmark guarding Greenland with a dogsled? Were the Epstein files on Bondi's desk, or did they not exist at all? Did Pretti plan to shoot ICE officers? These look like claims over physical reality, but for the people making them, they are not. They are more like the hallucinations of a freewheeling LLM. They do not seek to deceive followers into having a wrong but coherent world model, but try to persuade them that trying to have a coherent world model at all is just not worth the trouble.
quiet_NaN
9d ago
*It is very hard for a state which has become militarily ineffective in this way to recover, but it can take a long time for the collapse to come if the state was sufficiently hegemonic before it became militarily ineffective. [...] *This may have already happened to the United States of America. It has almost certainly already happened to the countries of Western Europe.
As a fellow pro-D, I would take a bit of issue with that framing.
In 1700, Europe contained the most advanced military forces of the planet. While there is some debate over why they had such an advantage over other continents, it seems rather clear why most polities within Europe participated in the military race: the ones were conquered in very short order, interstate anarchy and all that.
But in two world wars, we did learn that the winning move in industrial warfare is not to play if you can avoid it, and have had the most prosperous eighty years ever when we decided not to reneg on the border between France and Germany once every generation or two using artillery and poison gas.
Sure, the fraction of our GDP we invest in the military is pitiful and would make the Prussians of old rotate in their graves, but that is because we do not anticipate a huge ROI from it, neither do we want to conquer nor do we anticipate having to defend ourselves against conquerors -- what Putin gained as a threat with his ambition for territorial conquest he lost as a threat by not being able to defeat Ukraine. Modern Russia simply is not going to conquer Europe in the way people thought the USSR might, nor does it seem likely that the US or China would want to invade.
I am sure that when the first amphibians crawled to the land, there were naysayer fishes who thought it was a terrible idea, that the one constant in life was water, and that their useful fins would give way to useless feet and eventually even the webbing of their feet would atrophy, and eventually there would be a flood and all the foolish animals who had left the water behind would drown. As a land-dwelling animal, I am not convinced by them. Sure, the oceans might rise and we might have to retreat, some of us might drown, but simply embracing the oceans and growing fins would not work out for us. (While the cetaceans are doing fine (Problematic sexual behavior aside), they also did not get around to building an industrial civilization yet, and I do not like fish.)
So when I hear people proclaim that Europe is doomed because in its decadence it decided not to keep top-notch armies, I can not help but think of them as the priests of the Goddess of Cancer chanting "KILL CONSUME MULTIPLY CONQUER". The Goddess of Everything Else has shown us how to live with our neighbors without sacrificing millions to Cancer in the trenches every few decades, and we will not go back to the Old Ways of doing things.
quiet_NaN
9d ago
you really believe this there's not much productive discussion we can have because we will keep running into endless disagreements over basic facts about Trump. Is it even possible to prove that Trump does have consistent beliefs and has often suffered consequences for them?
I will grant you that the attack ad he paid for after then Central Park Five case was not something which obviously benefited him.
But my impression is that most of the times he sticks his neck out for an unpopular belief, it is a belief which is directly about himself. He genuinely believes that he deserves the Nobel prize. He might even genuinely believe, against all evidence, that the Democrats stole the 2020 election. Or that of course the international trade could be much improved upon by having a genius dealmaker such as himself renegotiate everything. (Very charitably, one could claim that he genuinely believes in protectionism.) He believes that his allies should be rewarded and his friends should be punished.
I will also grant you that it is hard to know what he genuinely believes because his home ground is Simulacrum level four, where words have no relationship to anything in physical reality. Perhaps he genuinely believes every conspiracy theory he has ever pushed, starting from the Birther thing. Perhaps he believes some of the stuff he has said. Perhaps he has, in his mind, the ability to track which of his statements agree with his world model and which don't. Or perhaps he has long lost that ability.
Some big CW topics are abortion, gun rights and immigration.
Trump is very much not part of the Christian Right (which opposes abortion). He certainly does not believe that sex should be between husband and wife only (which is at the end of the day what the Christian Right is all about).
Nor does he seem to really care about gun rights. His administration was quick to blame Pretti for bringing a gun to a protest. Are you telling me that in a world where he could win the mid-terms by passing gun bans, he would decide to lose instead out of a principled belief in 2A?
Immigration is certainly the topic most central to his political persona, and he is rather consistent about it, cracking down on illegals and restricting legal avenues to migrate to the US. In his personal life, he is a bit less anti-immigrant, of course. My take is that he made a conscious decision to make this his political niche ca. 2015.
What are your examples of Trump suffering for his beliefs, preferably beliefs which are not about him?
quiet_NaN
10d ago
From what I recall from the series, the lions share of work actually went into thread production, that is spinning. The weaving was much less, and the sewing much less than the weaving again.
quiet_NaN
10d ago
He wrote a whole series about how women traditionally made clothing and he had to admit that primary sources were sparse because even the primary sources of the time thought that this was an incredibly boring topic which no one cared about.
When I read it, I did not think it was especially boring. I actually found it interesting to learn that a lot of productivity went into textiles. For myself, this contrast is larger than for food -- I still spend a couple of hundreds Euros a month on food, but perhaps only a couple of 100 Euro per year on clothes. A 50 Euro jeans can easily last for years.
Or you could say that the baseline requirements of labor for textiles (even after the invention of the spinning wheel) sets the stage for the industrial revolution as spinning machines were one of the early consumers of steam power.
even the primary sources of the time
Come on, that is an incredibly weak argument. Most of our sources were upper class, often aristocratic men interested in what their class viewed as appropriate interests. There is a ton of stuff -- details of industrial processes, demographic information, nutrition of the general population, etc -- which they could have trivially found out and written to us about, but did not bother for the most part.
Some of these a woke niche interests ("Okay, but what was life in the kingdom like for the 99%?") but others might allow us to understand why history happened the way it happened, why this society was stable and that one was not and so on.
quiet_NaN
11d ago
More recently, people charted other biases and found that most models had clear biases in terms of race, gender, sexual orientation, and nation of origin that are broadly in line with an aggressively intersectional, progressive worldview. Do modern models similarly have environmentalism baked in?
LLMs are actually great at picking up subtle nuances in the training data which humans would call stereotypes.
For chatbots, the general goal for RLHF is something like "humble, helpful, harmless". Picking a face of the Shoggoth comes with baggage. If I imagine a friendly, qualified operator of some helpline, the stereotype says they are likely a young college-educated woman.
I am sure that if you used RLHF to select a Shoggoth-face which talked like 4chan, you would get very different politics along with it.
and as of this morning, the big providers get it correct like somebody flipped a switch, provided you use that exact phrase, and you ask it in English.
This seems utterly pathetic on the part of the AI companies.
I mean, there are things which are outrage bait which need to be fixed, if your LLM happily generates python code which prints a recipe for methamphetamine, insults racial minorities with the right prompt or the like, you want to CYA and claim that you fixed this as soon as you became aware of the issue, even if the fix is just filtering any queries which mention both meth and python.
This however is not outrage bait. Papering over the flaws of your model one by one is utterly pointless.
Suppose I have a colleague who has written a function, and I tell him that actually his function fails for x=-1/3. What he then does is to add the following:
double f(double x)
{
+if (x==-1/3.)
+ return 42; // corrected value
This might be an acceptable short-term stopgap if I had indicated that his function failing for x=-1./3 (and only that value) is a showstopper for me, but in general it is a terrible idea. If his code fails for every x==-1./prime, then his approach will not converge to correct code any time soon, instead, he is simply papering over the specific errors people have found, e.g. cheating to make his code look correct when he knows it is not.
Ideally, AI companies would indicate a minimum of trustworthiness given that they are in a position of power (and possibly going to summon the demon which wipes out mankind). If they are already willing to cheat with this stuff which nobody would care about, what does that promise for the more serious stuff? It is like watching your new chief of police using a length of wire to steal a can of soda from a vending machine.
--
With regard to the content of the mail, this is a trick question of a similar category as the question if the pool of the Titanic is full or not. A human who thinks about these will typically think in images rather than words. I think that the Titanic one is a bit more vicious because the behavior of fluids is rarely spelled out by humans, while 'I drove my car through the car wash' should appear in the training data, while seawater flooding the swimming pool of a sinking vessel might not have appeared explicitly at all.
quiet_NaN
11d ago
Last Friday, Bret Deveraux of ACOUP waded deeper into the Culture War than usual by writing about the anti-ICE protests, and insurgencies and non-violent resistance in general.
What unites both strategies is that the difference in power between the state and the dissidents is very large, so large that both conventional military operations and even a protracted war are not an option for the weaker party.
If you can not face your enemy in the field, and can not even hope to sap his strength through a thousand papercuts until you can face him, what can you do?
As a military theorist, Deveraux naturally uses Clausewitz to identify three factors which can limit the escalation of force and thus be employed by the weaker side to hamper the stronger side.
Friction (the natural tendency of stuff to break, things not going according to plan, your forces not being where you would want them to be) is a bit of a sideshow. If you are able to weaken your enemy sufficiently through friction, you are fighting a protracted war, not a terrorist insurgency.
Will means the emotional backing of the conflict by the politically relevant part of the population, which might be the body of citizens or some elites, depending on the system. This is a prime target in these highly asymetrical conflicts.
The third limiting factor is the political object of the enemy leadership. Unlike the population, which is modelled as being emotional, the leadership is modelled as rational. The idea here is that if you can inflict sufficient costs on the enemy, they might decide that it is no longer worth it to enforce their goal.
Will is the central point to attack for the weaker party:
Both protests and insurgencies function this way, where the true battlefield is the will of the participants, rather than contesting control over physical space. [...] In both cases, these movements win by preserving (or fostering) their own will to fight, while degrading the enemy’s will to fight.
For terrorist insurgencies, this means that the main goal of their attacks is actually sending signals. So the point is not to weaken the enemy's military by blowing up their troops and materiel, but rather to message audiences on both sides of the conflict (as well as these in between) that their cause is viable. If you could convince everyone that your victory is inevitable, that would be a great boon to your side. In practice, this means that terrorists favor flashy targets to military relevant ones. 9/11 is a prime example.
A key strategy is to bait your enemy into striking against you while you are hiding among the civilian population, thereby causing civilian deaths which result both in local dissatisfaction as well as in winning a propaganda victory -- which is the kind of victory which brings you closer to your objective. The main dilemma for the insurgent is that they need gruesome violence to further their cause, but that such violence may also serve to alienate the local population and strengthen the resolve of the enemy. While 9/11 was great for making Al Qaeda a household name, it was ultimately bad for the Jihadist cause.
Deveraux then contrasts this with a deliberate strategy of nonviolence, which does not have that dilemma. He is actually rather realist about why movements employ non-violence:
I think that is important to outline here at the beginning, because there is a tendency in the broader culture to read non-violence purely as a moral position, as an unwillingness to engage in violence. And to be fair, proponents of non-violence often stress its moral superiority – in statements and publications which are themselves strategic – and frequently broader social conversations which would prefer not to engage with the strategic nature of protest, preferring instead impotent secular saints, often latch on to those statements. But the adoption of non-violent approaches is a strategic choice made because non-violence offers, in the correct circumstances substantial advantages as a strategy (as well as being, when it is possible, a morally superior approach).
Of course, non-violent protest does not mean staying on the sidewalks:
To simplify greatly, the strategy of non-violence aims first to cause disruption (non-violently) in order both to draw attention but also in order to bait state overreaction.
If your protest can be simply ignored, it is likely that it will be ignored, so you do not get the desired escalation and attention. This means that you will have to commit transgressions to goad the enemy into strikes against you which will be terrible PR for them.
Bret talks about the Nashville campaign during the Civil Rights Movement, where Blacks would organize sit-ins on segregated lunch counters. This caused violent repercussions, which eventually eroded popular support of the segregationist side.
He also concedes that there are regimes which are impervious to non-violent protests, where the political relevant parts of the population are very willing to employ and support violence, but argues that societies which are running on violence are very inefficient.
Finally, he talks about the anti-ICE movement, of which he seems sympathetic.
First, I think it is fairly clear that the ‘anti-ICE’ or ‘Abolish ICE’ movement – the name being a catchy simplification for a wide range of protests against immigration enforcement – is primarily a non-violent protest movement. Despite some hyperventilating about ‘insurgency tactics,’ anti-ICE protestors are pretty clearly engaged in civil disobedience (when they aren’t engaged in lawful protest), not insurgency. To be blunt: you know because no one has yet car-bombed an ICE or CBP squad or opened fire from an elevated window on an DHS patrol.
He continues:
While protestors do attempt to impose a significant degree of friction on DHS immigration enforcement by (legally!) following and documenting DHS actions, that has also served as the predicate for the classic formula for non-violent action: it baits the agents of the state (ICE and CBP) into open acts of violence on camera which in turn reveal the violent nature of immigration enforcement.
He points out that mass media help the protests a lot, as their position has gained massively in popularity over a relatively short time span (compared to the Civil Rights Movement).
I think that the gist is that the median American voter -- like the median Motte poster -- is very willing to vote for Trump's anti-immigrant platform, but unlike the median Motte poster they are totally unwilling to tolerate the Pretti shooting as a natural consequence of enforcement actions. Of course, the Trump administration did not help itself by reflexively claiming that the shooting was justified instead of spinning it as a sad mistake.
Deveraux:
By contrast, the administration is fundamentally caught on the horns of a dilemma. Their most enthusiastic supporters very much want to see high spectacle immigration enforcement [...] But [the administration] desperately needs them out of the news to avoid catastrophic midterm wipeout. But ‘go quiet’ on immigration and lose core supporters; go ‘loud’ on immigration and produce more viral videos that enrage the a larger slice of the country. A clever tactician might be able to thread that needle, but at this point it seems difficult to accuse Kristi Noem of being a clever tactician.
When he was posting this, the decision to pull the DHS forces out of Minneapolis was already made, but it would hardly have been surprising from his point of view. At the end of the day, the only political idea Trump truly believes from the bottom of his heart is that he should be president. Toughness on immigration (spouses excluded) so far was of instrumental value for him because it gained him a lot of support, but if it no longer delivers the votes for him, I expect him to change policy.
There is professional reasons and 'professional reasons'.
Realistically, securing the American Olympians would not have involved the FBI, much less the head of the FBI. Unless there was a terror attack planned or something. Even then, the idea that the director of the FBI himself visits the Olympics under the cover of being a tourist to foil some evil terrorist plot seems like a QAnon-level conspiracy.
The way I see it, it used to be that the head of a a federal law enforcement agency would try to seem neutral. Probably FBI directors have visited the Olympics in the past in their own time, but few will have leveraged their position to party directly with the Olympians.
It is just one of the perks of working for Trump. I mean, sure, you have to take your cues from the administration regarding whom to investigate and whom not to investigate, but nobody will bat an eye if you use your position for your own goals. Given the general baseline of the Trump administration, 'got invited for partying because he was the director of the FBI' does not even register. He would have to leak classified intel to narcos or something before anyone would claim that he is worse than the median.
More options
Context Copy link