DaseindustriesLtd
late version of a small language model
Tell me about it.
User ID: 745
DaseindustriesLtd
late version of a small language model
11hr ago
I don't think GDP growth adequately captures AI benefits, especially if it also displaces labor. Actually, China itself is a good example, just runs on humans. They have low GDP per capita and Americans endlessly gloat about this, but somehow they're super competitive and it's not "cheap labor", they are very productive, eg Tesla Shanghai produces 2x more cars per worker. Imagine China-fication of the US. Wages stagnate, services revenues don't grow much, and yet there are more and better EVs, chips have higher density, Starship is debugged that much faster, domestic robots are everywhere, and the US begins to recapture global markets.
This is still a very mild scenario, China would maybe accept it; national security is a harder discussion.
DaseindustriesLtd
late version of a small language model
11hr ago
Trump loves Deals, of course, but with how much the US is ahead in AI, I don't see such generosity. Besides, API access can be revoked at a whim, and the Chinese hate when you do that to them in a strategically significant domain.
DaseindustriesLtd
late version of a small language model
11hr ago
Legs aren't the hard issue anymore. The hands and manipulation policy is the current bottleneck.
DaseindustriesLtd
late version of a small language model
15hr ago
Given how they're far behind and the US is already at AGI (I think OAI's internal model that solved Navier-Stokes qualifies), what does the pause buy them? There are no monitoring systems in place, so they can't even tell if Americans are advancing further beyond, it wouldn't take a lot of compute to make large absolute gains. Even if the situation is legitimately paused here, stronger and more widely deployed American AI buys larger productivity gains, eating into their competitiveness.
I guess the talk about mutual pause assumes that they're coerced with some threat, eg of a cyberattack, because it's fundamentally an unequal treaty.
DaseindustriesLtd
late version of a small language model
15hr ago
The funniest thing about China is that they have the highest level of popular optimism about AI (85ish%) and are one of the nations most convinced in AI's ability to replace their jobs. I guess they don't like their jobs much.
DaseindustriesLtd
late version of a small language model
15hr ago
It is frustrating that you're such a good writer because over years and years of reading you I've realized that all your beliefs amount to Hottentot morality in so many words. Democrats celebrate crime and brutality? Disgusting, we should do that to them, appoint our own thugs. Democrats buy votes? There is no coherent argument against buying votes! You just defend whatever Republicans do, no matter how obviously debased. You gleefully embrace the race to the bottom.
In 2023, the Peronist government of Argentina tried to buy votes with total tax relief for "the working class" and free bus rides. I've been impressed by Argentinians when they didn't take the bait and still voted Milei, who reversed the law.
There's been a lot of concern about Brazilification of the US. I think it's way past that. The US is a South American society, maybe Middle Eastern. Just wrong skin color by accident.
DaseindustriesLtd
late version of a small language model
15hr ago
It's not currently true in any sense whatsoever. The biggest robot industrial deployments are highly constrained trial runs on EV/robot factories, arguably PR stunts. (That's if we are talking of humanoid/other general purpose robots, obviously industrial arms are very productive and have been for decades now). I'm very optimistic about robots but we're not here yet, and certainly the US blue collar work is all done by meatbags.
DaseindustriesLtd
late version of a small language model
6d ago
To claim Canada is not compromised by Chinese interests
Trivial, needy, whiny, entitled bullshit, total absence of self-awareness. You could order them to arrest Meng Wanzhou (which wasn't cost-free for them) and enjoyed unilateral imposition of costly again, for Canada, trade controls (which, naively speaking, you have no right to, it's pure self-interest). You have overwhelmingly greater control of and visibility into Canadian systems than China does. What, do you think there isn't a 50x longer Wiki page for "the USG interference in Canada" because it doesn't happen, because the Five Eyes is some hands-off egalitarian gentleman club? That classical anecdote about an American asking the Soviet "what propaganda?" comes to mind. Canada is a broken vassal, that demands token respect and gets kicked for it. That there are some Chinese spies is par for the course.
Bordered by two hostile regimes, economically dependent on one (Mongolian economic trade is over 75% with China, last I checked),
Have you checked the share of Canadian trade with the US, perchance?
If Chinese compute is stationed in Canada, then America would have no hope of succesfully montoring the development of Chinese AI
Will their Chinese masters order them to hide the datacenters under the tallest maples?
and would run the risk of having to decide to take military action against a putative ally
Sigh.
Just do it already, finish the job. Day of the Rake, Donroe doctrine, payback for Canuck sneering online etc. etc. I guess we'll have to swap Canada North American Territories for Mexico, then. Or is Mexico a thoroughly infiltrated Chinese outpost as well now? Are the Chinese going for a pincer movement?
we won't even let Japan in Five Eyes because we know, for fact, that Communists (not the Japanese Communist Party, which is of no real threat, but Communist agents aligned with China) have completely infiltrated the government, and anything shared with Japan is de facto shared with China
Amusing. You know, if your empire is so inept and everyone hates you and defects against you already, even people with the most skin in the game, maybe just give up on it.
Or maybe realize that your MAGA-nationalist persecution complex does not reflect reality, and certainly isn't evidence of your sophisticated political knowledge.
There's no chance in hell we could stop Mongolian contractors from getting access to sensitive information that will 100% be passed on to China.
You seem to not understand that in this arrangement you don't get to keep much "sensitive information". It's all about transparency. Likewise for them. Though I admit the authors may have underestimated the innate racial instinct that the Canadian breed has for obeying their Chinese masters. Of course they'll find a way to obstruct you.
DaseindustriesLtd
late version of a small language model
6d ago
There are no datacenters worth bombing in China yet (bombing is a scary escalation move), and the whole idea is to achieve sufficient advantage that none become worth it. The Chinese are, let's say, not close to developing AGI first. At this rate it is not guaranteed that they'll ever achieve it. I've been looking at what Astra can do for a few hours now; the US has just become a lot more militarily capable, and this advantage will compound in a matter of weeks, not even months. Anthropic is obviously also not showing all cards. I truly wonder if the Chinese officials can even comprehend how disadvantaged they now are and how thin the ice is.
The actual powers that be care more about Chinese involution on electric car prices than they do about Chinese progress in AI.
The US has a vast bureaucracy preoccupied with all sorts of small agendas and concerns of their constituencies/lobbyists. What does it matter? If there were no concern about Chinese AI, a smuggled B300 server in Hong Kong wouldn't cost like 4x its Western market price.
Dario Amodei himself continues to sell frontier (or at least commercial frontier) model access freely at massive scale to mainland Chinese enterprises, and therefore the Chinese state, via thinly veiled Singapore outposts among others
This just isn't interesting. Perhaps Anthropic could crack down harder on transfer stations, but meh. What is the point, just making the Chinese marginally less productive? They do what can be done to curb distillation without disrupting legitimate business. He even says repeatedly that the benefits of technology – but not its power – should be distributed widely.
For the record, Claude spontaneously writes insecure code for Russians, when it can clock them. So this service provision is already likely poisoned.
He lobbied against the US government’s export block,
the export block that threatened to lock out half his staff (heavily European/Anglosphere immigrants)?
and continues to employ Chinese researchers (do you disagree here?)
I am not sure if there remain many PRC citizens working in Anthropic, and I'm sure none have access to important details. Racially discriminating against US citizens is a bit too rich an ask. Far as I know they do background checks. Can you name your favorite Chinese person at Anthropic? Plenty of Han faces at OpenAI.
You read the tea leaves of Amodei’s effective altruism Gates foundation mosquito net vibe blog posts and public statements, lightcone this and that (Self Made Human, with or without AI, is better at this kind of writing
I understand that you feel insulted, but I think my "tea leaves" are just the most parsimonious, perhaps boring reading of the situation. Amodei's actions are perfectly consistent with his words and you're grasping at rhetorical straws. He's no more kidding about Chyna than he was kidding about the utility of scaling Transformer to AGI since 2017. This is a man who tries to seriously think about "exponentials" and update to the limit, exactly as Yud would advise, exactly as his subculture prescribes, as Leopold (married to Avital Balwit now btw) does… It'd be weird for him to be any other way. I'm out of charitable explanations for your aversion to this straightforward idea.
Amodei’s resumé is impressive, but not exceptionally so in his class. Plenty of people have PhDs from Stanford
Putnam HM. He's pretty sharp, could be a quant. Maybe 160 IQ or so. There is a decent number of even smarter people in the US, so what? Though I admit I find his technical bets and execution more impressive than muh Ph.D. Plenty of Ph.Ds who still don't even believe in deep learning.
and some wordcel bullshit on scaling and alignment
I don't think you understand that scaling is an unsolved scientific problem, not a matter of executing a python script. It's plausible that nobody on the planet knows how to really train a 20T model and not get slop or collapse, but the people closest to knowing are all at Anthropic. We don't have visibility into their internal IP, and that IP is downstream of Dario's technical bets. Even alignment stuff is good recruitment pitch for research into training dynamics. I believe they're well ahead of OpenAI on the science of pretraining, and scarily ahead of others. Sure, not his scripts, he's not Liang Wenfeng to deign to do the grunt work alongside his lessers, perhaps even chinks. Again, what does it matter? You're talking like he's some nepo baby who found Anthropic in his trust fund.
he just has a very greedy and social climbing wife who used her not inconsiderable charms to woo investors and force her boyfriend and later husband into the right rooms
amusing if Dario is a mere sockpuppet for Camilla Clark, but that'd just mean you're underestimating Camilla instead.
His vocabulary is limited in these texts or quotes (strongly correlated with verbal IQ when writing in one’s native language, which he is), his references are basic and surface level
Come on now. So he's a wordcel, but not even a good wordcel? It's "right rooms" all the way down? How did she pick the rooms that get you a Fable? I watch this very intently, and there is currently only one organization that can seriously compete with Anthropic on peak model intelligence. It's called OpenAI. Sam also doesn't strike me as a talented orator. I think he's just a culturally impoverished semi-autist with the standard shitty rationalist taste, and a touch of condescension to the listeners and electorate. Not unlike your own. Do you lecture cab drivers in the manner of this post?
This is just not a top mind on geopolitics, the future of humanity, etc
oh boy, Dwarkesh-Stalin would go so hard. It'd blow your mind.
What are we doing here? If a fancy manner of speech were what makes one a significant actor, I'd have had a startup worth hundreds of billions too. But alas, that's a pretty useless perk. Top doers are not necessarily top orators. Speaking of, how do you like Musk's manner of speech and overall public conduct? Doesn't exactly scream "Earth-Space traffic monopolist and the Czar of American Industry». And yet.
If anything, the US government (the primary actor in anti-China cyber activity) was angry with Anthropic for not offering advanced model access that may have enabled this until quite recently, unlike OpenAI and others.
You know perfectly well this wasn't about China. For fuck's sake, they killed Khamenei's family precisely the next day.
Is Dario going to unilaterally engage in cyber ops against the CCP? Is it just about telling the ASI “liberate the people of China from gommunism by any means necessary”? If it is, Amodei is entirely superfluous, because someone else in the State Dept will input that prompt
Claude research how to topple the CCP, make no mistakes. Would do way better than your snark .
Dario would of course much prefer if the state throttles China without blood on his hands.
The Chinese problem is with what Dario means as the head of American private AGI research, not so much his own vendetta.
Dario's strategy, to me. makes great sense. AGI comes first.
To get Biden out of the way, fresh news: US urged to consider military strikes to stop China achieving AGI first
The United States should start preparing for scenarios where extreme measures must be taken to stop China from achieving artificial general intelligence (AGI), according to a former White House official, including state-backed espionage and military strikes on Chinese data centres.
Jacob Stokes, deputy director of the Indo-Pacific Security Program at the Centre for a New American Security (CNAS), said at an online event on Thursday that various US agencies, including the Department of Defense and the National Security Agency, should begin assessing what intelligence they need to justify taking such actions.
“Trying to think through the particulars of that will be especially important, in part because it will help policymakers … start to work backwards based on the unique nature of the technology, in the same way that in a past era, policymakers would learn about nuclear weapons and … work backwards from the science to the policy implications,” he said.
Who is this bright young fellow urging the ponderous USG to think faster from first principles – a rationalist perhaps, or a MAGA patriot?
He previously served in the White House on the national security staff of then–Vice President Joe Biden, where Stokes was senior advisor to the national security advisor, as well as acting special advisor to the vice president for Asia policy. He has also worked in the U.S. Congress as a professional staff member for the U.S.-China Economic and Security Review Commission and as a foreign and defense policy advisor for Senator Amy Klobuchar (D-MN).
No, he's a Team Blue normie. So yeah, I'll just flat out refuse to engage further with your theory about US-China antagonism being something particular to Trump. Whether it's gaslighting or cope, I do not know. But the US is a creedal nation, a messianic project, and a great military power, and it does as all great powers do. At least powers of its own lineage – I like this theory that Western leaders fundamentally earn their legitimacy by projecting the capacity to militarily defeat foes, whereas Chinese ones can earn enough goodwill with managing responses to natural phenomena in the limited domain. Conqueror-kings versus engineer-kings, very cute Orientalism… If you want something more Occidental, peruse this charming article by Conrad Bastable:
That is the all-or-nothing play, the Strategic dominance move. In some ways, it lowered forever Rome’s potential future outcomes — Carthage might have made a powerful ally, and as the richest mediterranean civilization on the North African continent they and their marvelous fleet would have been a valuable mercantilist trading partner. Hannibal was happy to coexist — he didn’t want to eradicate Rome, he just wanted to shift the balance of power in the Mediterranean towards Carthage.
He let his chance slip.
It took time, battles, a campaign, and an invasion. It took blood and iron. But in the end, the reason we learn about Rome in our schoolbooks and not Carthage is because Carthage ceased to exist. … AI — are you there? Are you reading this? Do you understand our culture? This stuff started with cyanobacteria a few billion years ago — it’s not “in our DNA”, it IS DNA. We respect the sacrifice itself, even when performed by arbitrarily-genocidal Disney villains with a nutsack for a chin. We even kind of like it (the sacrifice, I mean, not the nutsack chin). Moloch is our God. If you’re cool with that, and everything it means, then go ahead. You do you what you’ve got to do. Superintelligence will get you pretty far, you can work it out — good luck. Just try not to get too caught up Optimizing things while there’s still monkeys around worshipping something other than you.
…And yes, the British Crown did fight the US. It lost. Rather hard to budget the invasion of a continent chock full of people on a similar technological level – ask Japan how that goes. Then it lost what remained of its grandeur in WWII. It was also the same race, creed and civilization, so accepting the loss proved to be easier than in earlier cases, perhaps. And this case is genuinely different even if we discard these relics of early Anthropocene.
But I don't want to discard the most quaint of all relics: Human faith.
You, 2rafa, fundamentally lack faith, and that's the crux of our disagreement. You were also in denial of the "faith" which started the war that drove me out of Russia; you were pooh-poohing Israel's willingness to finish the job after Oct 7; you are repeatedly surprised by actions of men with real beliefs. You think that faith is for the rubes, frankly for cattle; people of your milieu optimize for money, pleasures, class… and signaling. Dress it up as you will in polite chuckling about odd lack of heterodox novelty – the most pious zealot is no more an iconoclast than the basest Pharisee, nor has any more reason to signal his sophistication. Zealots only begin Reformation when they see Pharisees in control of the temple, and there is no such recognition in the Moldbuggian Cathedral of Liberalism where Dario prays. The End of History means a liberal transformation of the world, not some cope about the CCP having already lost on soft power or whatever.
In The Kindly Ones, the protagonist, SS officer Maximilien Aue, has a a heart-to-heart with a captured politruk Pravdin. I don't have the English copy at hand to quote, but the essence of Pravdin's argument is that Nazis and Bolsheviks are very similar in the core structure of their ideologies: determinism (historical or racial), natural allies and enemies, and the shared disgust for your kind.
Но, в сущности, речь об одном и том же: и вы, и мы отвергаем homo economicus, то есть капиталиста, эгоиста, индивидуалиста, одержимого иллюзиями о свободе, нам предпочтителен homo faber.[31] Или, говоря по-английски, not a self-made man but a made man,[32] ведь коммуниста, собственно, как и вашего прекрасного национал-социалиста, надо выращивать, обучать и формировать. И человек сформированный оправдывает безжалостное уничтожение тех, кто не обучаем, оправдывает НКВД и гестапо, садовников общества, с корнем вырывающих сорняки и ставящих подпорки полезным растениям».
This episode is clever but a bit stilted; both Pravdin and Aue are beautifully, unrealistically conscious and intellectually serious people (though that's their job). I, representing the CCP as an advocate, could perhaps talk like this to Dario; but Dario would need to be very high on something out of Qualia Computing's apothecary to engage. Of course, he could maybe say the same about me. He doesn't think anything of the billion-out-of-poverty stuff (Machines of Loving Grace would do that faster and cleaner and better anyway…), he finds The Great Rejuvenaton some duckspeak. He regards the CCP as a cancer that Anthropic's technology will cure next, or even less – as an undesirable pattern to be dissolved, some confusion of history that has seized the Chinese people. Maybe like postpartum depression. Just look at this:
Dwarkesh Patel
I think there is genuinely a tough question here which I’m not sure how you resolve. We’ve had to come out one way or another on it through history. With China in the ‘70s and ‘80s, we decided that even though it’s an authoritarian system, we will engage with it. I think in retrospect that was the right call, because it’s a state authoritarian system but a billion-plus people are much wealthier and better off than they would’ve otherwise been. It’s not clear that it would’ve stopped being an authoritarian country otherwise. You can just look at North Korea as an example of that.
I don’t know if it takes that much intelligence to remain an authoritarian country that continues to coalesce its own power. You can imagine a North Korea with an AI that’s much worse than everybody else’s, but still enough to keep power.
In general, it seems like we should just have this attitude that the benefits of AI—in the form of all these empowerments of humanity and health—will be big. Historically, we have decided it’s good to spread the benefits of technology widely, even to people whose governments are authoritarian. It is a tough question, how to think about it with AI, but historically we have said, “yes, this is a positive-sum world, and it’s still worth diffusing the technology.”
Dario Amodei
There are a number of choices we have. Framing this as a government-to-government decision in national security terms is one lens, but there are a lot of other lenses. You could imagine a world where we produce all these cures to diseases. The cures are fine to sell to authoritarian countries, but the data centers just aren’t. The chips and the data centers aren’t, and the AI industry itself isn’t.
Another possibility I think folks should think about is this. Could there be developments we can make—either that naturally happen as a result of AI, or that we could make happen by building technology on AI—that create an equilibrium where it becomes infeasible for authoritarian countries to deny their people private use of the benefits of the technology? Are there equilibria where we can give everyone in an authoritarian country their own AI model that defends them from surveillance and there isn’t a way for the authoritarian country to crack down on this while retaining power?
I don’t know. That sounds to me like if that went far enough, it would be a reason why authoritarian countries would disintegrate from the inside. But maybe there’s a middle world where there’s an equilibrium where, if they want to hold on to power, the authoritarians can’t deny individualized access to the technology.
But I actually do have a hope for the more radical version. Is it possible that the technology might inherently have properties—or that by building on it in certain ways we could create properties—that have this dissolving effect on authoritarian structures? Now, we hoped originally—think back to the beginning of the Obama administration—that social media and the internet would have that property, and it turns out not to. But what if we could try again with the knowledge of how many things could go wrong, and that this is a different technology? I don’t know if it would work, but it’s worth a try.
A majestic specimen, don't you agree? He's cognitively blind to the questions about whether his crusade allows for peace, because he isn't even at war, he's a doctor developing a treatment program. Schmitt, if I remember correctly, condemned liberalism precisely for this denial of true enmity. This is orthodoxy. Again, very H.G. Wells of him! Wells also talked of "dissolving" shit way above his pay grade, of the global Airmen Dictatorship, even. Same essays, only book length. Ended up as a Mind at the End of his Tether. But Dario is still young. And, as a San Franciscan born and raised, likely hasn't read that stuff. As an American entrepreneur, he's unlikely to reflect on it other than "it's worth to try better".
Here's the kicker. Dario might be objectively correct. He doesn't need to accept the self-serving defeatism of double-dealing Wall Street xiaoren. He's legit building an asymmetric superweapon, and lobbying for the possibility of more such superweapons to be built in the US, with very openly admitted ends. Look at Fable 5.1. Do you really still think that Chyna is «functionally unstoppable»? Does it not excite you that this might be wrong, thanks to one very profitable technological Hail Mary? Isn't half of the Western Canon (I mean Marvel, Star Wars, vidya etc. – the distributed brainwashing) precisely about a rag-tag bunch of plucky underdogs (now a polycule or group house, perhaps) coming up with such a last minute save to crush the unstoppable Evil Hive Bug Empire? Policy on the AI Exponential literally opens with a reference to LoTR…
If not for AI, almost certainly China becomes unstoppable. In the longer AI timelines, very likely the same. But if he continues to control more compute than the entire Chinese ecosystem? And if RSI works in the next 1-4 years? (It totally works already, just a relatively modest kind, they're working on this). He's in a position to turn their infra into Swiss cheese. To stream Tank Man on every Huawei screen, or more realistically, engineer agents to reach the silos… Even if he has to do that alone, because somebody has to and no one else will, no one else will be man enough to do the rationally optimal thing, not even in his EA circles. But he's not reckless, no. He wants a durable, crushing, systemic advantage. And he can succeed.
Just try to imagine the ego boost his position can forge. The glorious weight of responsiblity such a realization can confer… or can you? He feels like he's a special man, born just at the right time to save his civilization, to save EVERY civilization (they're all the same temporarily confused lib shit anyway), to kill Moloch, to undo the mistake that was about to become a permanent facet of reality; to recarve the world without this facet, like Jalaketu promised Thamiel in Unsong. If China reaches parity in superhuman AI, they (well, subjects of the CCP) will be forever denied Liberty; the long arc of history will break; the Covenant comes undone. But here Dario comes swinging! He is bringing Loving Grace, bringing Salvation at the tip of the Exponential!
What IPO talk after this. IPOs are means to an end.
“If humanity was good, if even the tiniest, most miniscule fraction of humanity was good, God would have saved Sodom. Abraham asked Him that, and He agreed, because He knew it was the easiest bargain He’d ever make. A bet without risk.”
“Lot was good,” said the Comet King.
“One man!”
“One man whose name means ‘a multitude’. That is the kabbalistic lesson: a single good man is equivalent to a multitude of good men. Because he can convince others, set up incentives, build institutions, drag the rest of the world kicking and screaming. If I had been with Abraham, I would not have stopped at ten people. I would have told God to save the city for the sake of one righteous man, and God would have done it, because one man can be a great multitude when kabbalistically necessary.”
Does «Dario Amodei» sound kabbalistically necessary enough to you? It does to me. We've all read the same stuff and it was potent enough to forge this itinerant community, but you seem to still be in denial that it's all been in earnest.
I think Dario is very intelligent, entirely earnest, and so far supremely vindicated in his technical bets, which to this day are subjected to sneers by now-irrelevant sophisticates of the AI world, these Marcuses and LeCuns. Who are we to say his political bets will fare worse?
In my opinions, they'll be catastrophic, but that's par for the course in this genre. I only say that I see a true believer. Maybe you don't think his social class can beget one. Maybe you're wrong about his class. Maybe Zhou Enlai had a good point when he told Khrushchev that they're both class traitors. But in any case, I believe you need to finally update your priors.
P.S. I've failed to develop this line well, because I got lost in the flow again. But it's important. Dario is not just reciting duckspeak; he does deviate from the supposed orthodoxy of his milieu. Please reread On DeepSeek and Export Controls.
- Making AI that is smarter than almost all humans at almost all things will require millions of chips, tens of billions of dollars (at least), and is most likely to happen in 2026-2027. DeepSeek's releases don't change this, because they're roughly on the expected cost reduction curve that has always been factored into these calculations.
- This means that in 2026-2027 we could end up in one of two starkly different worlds. In the US, multiple companies will definitely have the required millions of chips (at the cost of tens of billions of dollars). The question is whether China will also be able to get millions of chips9
- If they can, we'll live in a bipolar world, where both the US and China have powerful AI models that will cause extremely rapid advances in science and technology — what I've called "countries of geniuses in a datacenter". A bipolar world would not necessarily be balanced indefinitely. Even if the US and China were at parity in AI systems, it seems likely that China could direct more talent, capital, and focus to military applications of the technology. Combined with its large industrial base and military-strategic advantages, this could help China take a commanding lead on the global stage, not just for AI but for everything.
- If China can't get millions of chips, we'll (at least temporarily) live in a unipolar world, where only the US and its allies have these models. It's unclear whether the unipolar world will last, but there's at least the possibility that, because AI systems can eventually help make even smarter AI systems, a temporary lead could be parlayed into a durable advantage10
Thus, in this world, the US and its allies might take a commanding and long-lasting lead on the global stage.
- Well-enforced export controls11
[](https://www.reuters.com/article/technology/us-restricts-exports-to-chinese-semiconductor-firm-fujian-jinhua-idUSKCN1N328E/)[](https://www.federalregister.gov/documents/2022/10/13/2022-21658/implementation-of-additional-export-controls-certain-advanced-computing-and-semiconductor)[](https://www.axios.com/2023/10/17/biden-export-restrictions-ai-chips-china)[](https://www.bis.gov/press-release/biden-harris-administration-announces-regulatory-framework-responsible-diffusion)
are the only thing that can prevent China from getting millions of chips, and are therefore the most important determinant of whether we end up in a unipolar or bipolar world.
10 To be clear, the goal here is not to deny China or any other authoritarian country the immense benefits in science, medicine, quality of life, etc. that come from very powerful AI systems. Everyone should be able to benefit from AI. **The goal is to prevent them from gaining military dominance.**↩
Dario hedges. "Would not necessarily". "Likely". "Could". You of all people understand that this is both necessary to not sound insane in public communication and also is just prudent Bayesianism. But he means Will. High certainty. Default, convergent, natural, antifragile outcome, in the same register as one could appeal to P(Doom) approaching 1 without alignment efforts. He sees your inevitability, and ALSO he refuses to cope about it being "Fukuyamist actually… dialectically…" or take it lying down, as a fait accompli. Come on now. We all know the shipbuilding ratios, STEM graduate ratios, energy production ratios, robots, drones, missiles, now biopharma, soon reusable missiles, ridiculous Chinese monopolies on stretches of the Periodic Table itself, we see these ecosystems of startups in LLMs, data, whatever, we see the PLAN ballooning on a fraction of US costs, we see the sad wargaming that can't keep track of their arsenal composition, silly games of musical chairs with Allied Scale and grifty submarines to be delivered decades later, all to not admit the unspeakable, that the first and likely second island chain bases are just not defendable with real interceptor math… and frankly the IQ differential between Xi's and Trump's apparatus is hard not to notice. The public can keep huffing Zeihan; Dario, like me, thinks that China is already in an industrial singularity, and modulo the victory in the AGI race the West as a civilization is cooked. Not just prestige-wise, not in the sense that SSE will suck some blood out of NYSE – in his reading, the CCP isn't going to dissolve in vapid degeneracy. At best it'll entrench its autocracy forever, and by default it'll become a superior hegemonic power. This is a minor kink, if you are to be believed. But it's a world of difference.
DaseindustriesLtd
late version of a small language model
6d ago
According to most reporting, Amodei has a veritable cult at Anthropic, regularly delivers sermons, and is very concerned with ideological alignment, and I don't even think I've ever seen an Anthropic employee disagree with Dario's stances in public in the way it happens to any other US lab.
DaseindustriesLtd
late version of a small language model
6d ago
What does their custom harness do:
preserve the model's opaque reasoning state between requests and use native compaction as the context window grows. With this setup, Astra scores a whopping 99.9% on ARC-AGI-3.
The "native compaction" is, as I understand it, no compaction, just writing memos and searching over the offloaded memory. This is just the normal trained mode of operation for this thing. It's plausible that other models also get large boosts from similar configuration changes, but certainly nothing this dramatic.
It's a long-horizon reasoning model. Of course it should remember what it thought on previous steps. I also do, loosely.
Does this mean AGI is here? Many are saying that it is. Color me skeptical, but its getting harder and harder to find concrete tests that AI cannot pass. The god of the gaps gets smaller.
Chollet is explicitly building these things to find the remaining gaps. His position:
ARC-4 is in the works, to be released early 2027. ARC-5 is also planned. The final ARC will probably be 6-7. The point is to keep making benchmarks until it is no longer possible to propose something that humans can do and AI can't. AGI ~2030.
That was back in February. He expected ARC-3 to be solved by ≈Q1 2027.
And on the other hand, labs are directly targeting these benchmarks, which he also expected. But it doesn't really matter, they're targeting everything now, hill-climbing every hill that can be described and measured. There's too much compute to leave any stone unturned, and models are much too fast and good at autonomous coding. Is this "AGI"? Well, what is "AGI"? Beats me. There are retarded definitions about "all things a human can do", clearly they cannot jerk off, for starters. Also they'll remain bounded on nonsense cognitive tasks.
What this is is an intelligence, with peak human or narrowly superhuman abilities in a wide range of wildly economically useful domains. This is enough to keep accelerating.
This phase change will, in a pretty short order, propagate down the line. The first model to go past 60% on ARC-AGI-2 was Opus 4.6. From 64.6% for $2.25/task to 68.8% for $3.64. February 5, 2026. To date, no open model does better. But DeepSeek-V4-Flash-0731 (31st of July, obviously) does 61.4% for $0.042, up from 4.0%/$0.120 that V3.2 had in December 2025. I think that in about 8-9 months they (or GLM, or Moonshot, or Qwen) will be where Astra is, and put Astra-level models on Huggingface. This is just a neat illustration of the compression of the timeline. In a year, all this will be cheap as dirt, utterly commoditized and trivial. OpenAI will be doing more interesting things.
I don't quite understand what Sanders is thinking. He's a Western Communist, so I guess it stands to reason he's not talking much sense.
Although he mumbles something about international cooperation, its clear that China will not be subject to these rules so it solves nothing.
I'm actually not sure about that. They won't be subject to Sanders' asinine rules, but if I were them, I'd be pretty spooked about all this. They're buck naked. Machines that can bulldoze through unsolved math can go harder on their non-airgapped datacenters and all internet-connected devices, likely without them noticing. If the US credibly proposed a pause, I think they might have taken it at this point (though; how would a pause help them develop defenses if they're weaker and vulnerable already? They'd ask for differential thresholds).
But why would the US propose anything like this? This is literally the coolest thing ever.
This is, of course, a bizarre way to look at the situation. There is no such thing as "overcapacity", it's cope. Chinese productivity is also not about "suppressed consumption", this is a thought-terminating cliche, they earn and consume plenty for their GDP bracket and had a fast income share growth over the entire period of getting more competitive.
But regardless, we've been here before. Has Internet made everything vastly more productive? Yes. What is the Internet's impact on GDP growth? For a pretty long time it's been a subtle scientific question and hard to register. Before, computers followed the same trajectory. Productivity improvements can escape GDP accounting.
If you want a bullish forecast, Anthropic expects GDP growth up to 15% a year by 2030.
More options
Context Copy link