Did he tho ? Win I mean, the USSR did disintegrate but the narrative that US policy was more than marginal to this disintegration has always seemed self-serving to me. The USSR was basically a colonial empire losing control on its periphery, yes the fact that there was a serious competitor is part of why it was losing control, but actual policies of that competitor, I don't really think so. I think what happened is simply that the core was'nt strong and self-confident enough due to decades of corruption and gerontocracy and that without a strong core, empires disintegrate. Did the US really "win" when the UK lost its empire, did Germany "win" when France did ? In a way, yes, but in the sense that they profited from it marginally, not in the sense that it happened due first and foremost to their action. Afgans won against the USSR, Balts won (but they still are as butthurt as if they had lost, so they lose !), the 'west' not so much, we just looked at the trashcan fire self-congratulating ourselves.

Humans can, and do, do it, but it's not something that is as easy as calling a tool. Drugs are brute force solutions (but so is vector steering TBH, it's just that we *also* have a physical body that pays the brute force tax), autohypnosis-related techniques are high effort and training and the conscious layers of the brain may have a very different desired vector than the lower layers and the whole system is definitly not byzantine fault tolerant in the least !

And I'm also a big fat hypocrite because my elder boy is probably going to expat to some godforsaken nickel mine in LatAm or methane field in whateverstan.

Are'nt you missing more than half of the narrative here ? Russian polarisation actually is much less ideological and mythological than "western" polarisation, to Russians "nazis" are genocidal invaders that must be pushed back, not a persistant internal threat that must be gatekept out of power forever. It's not communists that defined their entire political legitimacy on "supressing the next Hitler", the great patriotic war is'nt really that ideological a myth, it's a national union under existential threat myth. It's western liberal democracy that created both neoconservatism and the new left, both defined entirely by the WW2 narrative, one more focused on external hitlers, and the second more focused on internal hitlers.

Like the formerly most innovative Chinese models were, they are american... The difference is that China is getting more and more attractive, training and retaining more and more researchers while young French people feel no reason to stay home. I believe they are short-sighted, and should stay, but I *understand* why they don't, I'm not blaming them, there is no obvious reason to stay and the non-obvious ones that you need a lifetime to understand are **also** on a very downwards trend anyways...

It's not any different today, Christianity has been *extremely* adaptative to social reality, however valuable and truthy the core doctrines, it spread from adopting roman "interpretatio romana" oeucumenism and roman civilizationnal tooling. It is then no surprise that when western founding myths and ideological zeitgeist are brutally replaced in a few generations, it adapts quickly to the change. Its disappointing, it's a real indictment of the institutional backbone of organized Christianity, but surprising it is not. The conclusion is that even if it's true (and I do believe it is, in the important core) then the parts about God protecting the church are still false regardless (yes, I know the cope, He protects the hidden church, but that's about as bullshit as "it wasn't true communism").

This nails it, I'm a bit slow sometimes, but the evidence here tipped me over from "I feel something is off" into "we are measuring it wrong" territory. Not in the sense that the benchmark is bad, maybe it is, maybe it isn't, but in the sense that, like basically all benchmarks, it measures autonomous capabilities. And on that front, reliability beats intelligence, it doesn't mean that hill climbing long autonomous tasks and auto research isn't valuable, as a scoring function it's great. As a benchmark however, it's subtly off-base unless your criteria is that the model has to to the whole job, not contribute as an actual **team** member. If I take a straightforward task that may require a lot of tedious looping through low information feedback to complete, and don't want to be involved until I get a merge request yeah, Sol 6.1 is good, it's great. But as an exobrain addition to an actual knowledgeable human, it sucks dicks in hell, Gemini flash is better, GLM 5 is better, hell, Deepseek V4 pro would probably be much better if it weren't slow and expensive ! So the benchmark isn't lying per se, it's just measuring a "very specific set of skills", but if people are selecting an exobrain based on that, **including** for the specific task of doing ML R&D, they are getting fucked, hard.

It will only make the bloodlust worse, since safety is definitely **not** the goal, improving safety won't change a thing, it makes Deepseek more likely to be used and Dario's nightmare is intelligence being accessible to his enemies, which is why he hates China, because they **already** have what he fears, AI or not AI, a country full of intelligence that isn't "aligned" to his brand of new left monoculture.

It's pretty much guarenteed they will, and agentic long sessions will help, because they are the unlocking step for swarms. The question is whether they will spend on aura problems or just do the hill climbing and just let the price speak for itself and let Chinese universities (and others, but others won't take advantage for political reasons) do the spend. Obviously, there are infra optimisation problems that can, hypothetically, be improved on with heavy mafs, but it's not that easy, human mafs went high enough there that the low hanging branches where you get benefits at reasonable scale were likely mostly cut already by supercomputing research.

Is that Sol 6.1 ? I don't dislike it, it *is* cheap (on API and enterprise contracts), but it feels a lot dumber than GLM 5.3f, let alone 5.3 or DS 4.1f. I am assuming it's wasting a fuckload of reasoning being extremely compliant, it still has the OpenAI signature instuction following dead-seriousness, and since it strives to be token efficient, I think it is reasoning budget starved, it feels like it both tries to not think too much *and* spends a lot of tokens checking its own state to "make no mistake", makes me think of a junior lawyer that spends too much time thinking about keeping his billables reasonable...

Turbo retard *can* work, but you need a very strong popular legitimacy base to do so, the US historically has been great at helping you get one by being even more retarded on a global basis, but it seems to have better instincts in LatAm and has generally avoided razing cities and building rape camps too obviously there.

LOL, they actually did quant from first principles ? I did'nt know that, it's impressive, and a trap for competitors indeed, reproducing that feat is a lot harder than just doing it the normal way, survivor bias hard at work here !

DS is actually being professional about it, they just are'nt being Corporate Memphis reddit-coded SV quirky millenial "professional". The cultural roots being finance, I'm actually expecting internal codenames to be memeworthy up to the razor's edge of cringe. TBH I have seen a lot of cringe branding from China, but rarely if ever the same brand of cringe as what spawned sphincter logos.

I know I'm in the minority here, but I'd like a 4.1 Pro that is'nt too big. 1. Training a 3-10T properly requires a fuckload of tokens and deepseek has'nt deployed GW scale infra yet (and even when it does, a lot of it will probably have to be inference), I'd assume that over a few Tparameters existing pretrain corporae are too small and massive synthetic inputs are needed, quality control of such to avoid claudism self-reinforcement is HARD. 2. It's only economically valuable if the demand for research-grade models is high, I'm not convinced at all that the general industry maturity is anywere near the point where it can use those. 3. The compute and RAM glut is not going away anytime soon, efficiency is its own reward, if and when research-grade inference starts compounding, efficiency gains in the seed are likely to compound with it. Yes, the theory is that you can just waste petajoules of extra compute early to compound earlier, and get the efficiency later, but it strongly feels like richfuck cope to me, cheap capital never had that effect in the past, it's useful to quickly take market share, but never translated to efficiency even when spent on workforce recruitement.

It *looks* like it's model size dependent, not solely model size dependent since Argon is'nt that great (but it could be smoller than it looks, or have its own knowledge offload engram-like trick) and K3 is below GLM 5.3, but still, the gap between GLM 5.3 and 5.3 flash in this ranking tells me that size matters a lot to it.

The idea is that the people who say this know exactly what "democratic" means, it means "aligned with USrael", nothing more, nothing less. And with this definition, it makes perfect sense, two integrated vassals would have precisely the relationship their master allows, nothing more, nothing less.

Yeah, plus the toasterf*cker internet phenomenon. I still believe that on average the benefits of internet organic communities slightly overweight the negatives, but boy are the negatives bad...

Ghislaine Maxwell

This, Balts had a good image too, before we actually got to discover butthurt belt politics. If Japanese internet is what no-piracy does to a mf, then all hail 🏴‍☠️

The crisis was fake, not that debt/GDP does'nt measure a problem over time (the instantaneous measure is'nt that interesting, but if it grows, it means that debt is not being used for investment), but the systemic risk was fake, in the sense that it was neutralized pretty easily. The core issue is whether there is actual national unity on a specific solution, if so the EU would go cry about it, even for a small country, because their *only* option is to threaten separation. Treaties may at some point have been considered as something else than a joke, but in current-year they have been shown to be nothing else. The thing is that the EU viewpoint **is** the plurality viewpoint, getting monetary sovreignty back may be a majority viewpoint in the abstract, but it's politically split enough that boomer fear of change is the actual politically effective plurality, there is not enough legitimacy to tell the EU to go fish.