Caffeine Powered

@Tomodovodoo @NateBubis I think we will see a lot of problems get solved because we can chase more rabbit holes for cheap.

@NateBubis Over so so many grids, where so you start? I guess someone could have worked their way up and checked, but I guess this was seen as a fruitless endeavor. This time Fable took the time to do so.

how to disproof the Jacobian conjecture: 1. give a random counterexample on twitter. 2. done. Well, congrats I guess. As far as I could check it seems to... Just be a correct counterexample.

@thesupermanmx These researchers findings are irrelevant by the time they publish.

Researchers proved every major LLM is secretly obsessed with Japan. And they finally figured out why. For years, we’ve been told that AI is entirely Western-centric, that it just reflects Silicon Valley and American values. A landmark paper by Cardiff and Basque researchers tested 31,680 cultural prompts across 24 languages on frontier models like ChatGPT, Claude, and Gemini. The results shattered that assumption. In six out of eight frontier models, Japan was the single most frequently referenced country when asked open-ended cultural questions. Ask about traditional dances, festivals, or everyday practices in an open context, and the AI defaults to Japan. Over and over again. Here is the twist nobody expected. This bias doesn't come from raw pre-training internet data. The researchers tracked where the obsession forms. It emerges after pre-training, during the supervised fine-tuning and alignment phase when humans teach the AI how to behave. Why Japan? Because decades of global soft power, rich cultural export, and clean, universally admired digital archives make Japanese culture uniquely "safe" for AI safety filters to lean on. When labs train models to be harmless and universally pleasing, the AI defaults to the cultural equivalent of comfort food. It avoids controversy by talking about anime, sushi, and tradition.

@misraetel We just discovered a new species of monkeys. It isn't because it took particular genius. It's because there is a limited supply of people who can know that a given monkey isn't in the corpus to go to every inch of the world.

@misraetel I'm bearish on ASI with current architectures. I think we'll see a surge in areas where going down a lot of rabbit holes is fruitful. There is a lot to be discovered not because it requires genius, but because there isn't enough talent to chase down all the rabbit holes.

My prediction of ASI in 2026 looking a bit warmer! 👀

@RichardHanania Why do my tax dollars pay for someone who is so bad at their job?

Jesus Christ this is bad.

@nixcraft Linux is written with AI now.

If LLMs are good at coding, why don't we see any killer games or apps written in them largely? Where are all those vibe coded apps and their solo developers making millions? 😂

@asymmetricinfo In Richmond there is a debate over funding right now. A whole lot of (far left) people think we should tax VCU, a public state run school and research hospital. I think universities shouldn’t underestimate how much people reflexively hate anything that has a lot of 0’s

I think that there are a lot of epistemic reasons it's unwise to set the university up as a countermajoritarian political project, but even if you think I'm wrong, who will be funding this project?

@JhonnyB17025719 @thomasunise If you can only run a software if the maker provides you with the binary, it’s not open source.

@JhonnyB17025719 @thomasunise You can fine tune it with settings or scripting. If a new version is released, there is no way for you to get from K3 to K4 yourself because you can’t even make K3 yourself. Imagine if a software didn’t provide the binary, it would be impossible for you to run it.

If Trump bans Chinese AI because Kimi and GLM are catching up to OpenAI and Anthropic, that isn’t “winning the AI race.” It’s admitting defeat. Letting Washington turn regulatory capture into industrial policy is the lowest IQ move you could make. Altman and Amodei are losing market share faster than expected so they have to lobby for a moat, kill open competition, and preserve their right to price-gouge Americans while Chinese labs democratize frontier intelligence for the rest of the planet. Nothing says “America First” like banning the models that made American AI companies look second-rate. LLMs are a commodity and going to zero. The future is Open Source and there is no stopping it. Buy GPUs and download open weights while you can.

@jun_song @GabiiAH11 Open source AI doesn’t exist. Just because software isn’t a SaaS product doesn’t make it open source.

@GabiiAH11 Open source AI

I smell extreme fear from USA. It’s just beginning.

@Geiger_Capital They are not open source. No more open source than iMessage is. If the free models simply distill the paid models, what happens when the frontier labs stop improving the paid models?

The fact that cheap, open source AI models can essentially match frontier model capabilities is very good for the general public. It’s very bad for frontier labs trying to gatekeep and charge for access to super intelligence.

@hosseeb @ylecun @deanwball It’s like claiming MS word is superior to Google Docs because it is “open binary”

@hosseeb @ylecun @deanwball It isn’t open source software. It’s no more open source than if grand theft auto 6 gets released for free. It being free isn’t what makes something open source.

This argument by @deanwball is being badly misunderstood. It's OK to disagree with it, but first you have to actually understand what he's saying. He's saying: releasing the weights for a frontier-level model is effectively dumping. Dumping is when you sell a product at significantly below cost in order to corner market share. It's illegal. The reason: dumping results in short-term consumer surplus, but long-term it prevents the formation of a competitive market and discourages capex outside of the dumper. Standard Oil famously did this in order to consolidate the oil market before it was broken up. So why is he claiming releasing the weights of a frontier level model is basically dumping? Isn't he just describing open source? His argument: it's not financially sustainable to train a frontier model and release the weights. In the long run, you will not be able to internalize enough of the gains given the cost of training a frontier model, because neoclouds and other inference providers will be able to outcompete you at actually serving the model. It costs an astronomical amount of money to train frontier models, and if everyone else can serve them, you don't capture enough of the surplus to pay for the training and R&D. It's not like normal open source when you build some software and then release it and sell services on top of it. The amount of capex required for frontier-level models is an order of magnitude higher than normal software, which is why doing this at frontier level is so economically irrational. Right now the Hong Kong stock market is ebullient enough that Chinese AI companies are not getting punished for the fact that they're all deeply, deeply unprofitable. Releasing model weights is great marketing, intellectually appealing, and strikes fear into the hearts of their opponents. We can assume the status quo continues for a while because of the AI supercycle. But eventually the AI market will correct, the Hong Kong market will dump, and suddenly these Chinese labs won't be able to afford to training super expensive models without internalizing more of the gains. But what if China, seeing that this strategy is successfully kneecapping the US lead (by discouraging further capex and lowering valuations), says no--don't stop. And so the Chinese government starts buying up the shares of these companies and demanding that they continue releasing frontier-level weights, profitable or not. In that case, it becomes a genuine space race. For-profit companies cannot continue to compete on either side. US labs valuations fall, and the White House realizes that to keep their advantage in the AI race, they cannot rely on the free market to maintain their lead. They nationalize the labs and fund them off government subsidies. Now you have government-controlled and distributed models on both sides. That's what Dean is calling the "dystopian hellscape." The best analogy is drug development: if China were to sell American drugs back to us really cheaply, that would result in a large short-term consumer surplus. Cheap Viagra and Ozempic is obviously great. But in the long run, this would discourage investment in developing new drugs. That's the sense that Dean is saying it's long-term "decel." Now, I happen to disagree with Dean. I think the consumer surplus of having frontier-level open weight models is huge, even at the current capabilities. I also think China is going to defect from this strategy soon (there's been reporting along these lines, that Beijing will stop allowing large models to be open-weight; I think there are other reasons for this aside from competition). I also suspect that nationalization of labs is inevitable as they take on more geopolitical and cyber capabilities. But he's not wrong--releasing frontier-level weight models is weird. The question of how long this market will remain profit-driven is a very coherent question to ask.

@Esk3nder It’s not open source! Linux is open source. Grand theft auto 6 isn’t. If they gave you the game for free, that wouldn’t be open source either.

I can’t believe this needs to be defended. Haseeb is right. It’s an uncomfortable truth but China found an exploit via opensource LLMs, leveraged as industrial policy. They are absorbing the cost of frontier development, which compresses Western labs’ margins, and makes the global ecosystem increasingly dependent on Chinese releases. Geopolitically, China gets the distribution while the US is forced into some kind of “state intervention” in defense of its ecosystem and IP. Obviously this doesn’t make open source bad, but pretending the strategy is purely altruistic is retarded.

@jurbed How does moonshot make money? They don’t sell hardware.

The difference is how you make money. US it's training to create intellectual property to sell inference. China is leaning towards giving away models for free, but selling hardware for inference, undercutting overpriced western hw and winning on hw exports. It's just a completely different business model, I don't know which will win, but remember, China doesn't really do intellectual property, they do factories. Note they many models are optimized for the new completely Chinese made chips for inference. The semiconductor VC fund invested in DeepSeek.

@QueenMab87 So you would be fine if we discriminated against you? Or do only you get to decide what is a harmful belief?

My history cohort had a very conservative member doing military history & foreign policy. He made things hell for us. Contrarian, dismissive, actively sexist. It’s not discrimination to face consequences for harmful beliefs & behavior

@ChrisHeaslip Part of that may simply be the default settings for codex. Have you tried using 5.6 on high with thinking mode turned on while in the browser? It makes a huge difference.

As a non-engineer, my assumption was that AI = ChatGPT on web. But Codex is 10000x better even for someone non-technical.

@the_transit_guy @thealfordplea I rarely see you criticize the costs or try to provide an indication of how much it should cost.

@thealfordplea Well that might be a generality but definitely that cant include me when transit costs and delivery is all I talk about. The same can be said for car proponents who don’t ever care about project cost.

One of the reasons people like me largely ignore opponents of the California High-Speed Rail is that they aren't looking to deliver the project faster; they just fundamentally hate trains.

@fornicv @Astuterealm0 @3RenChengHu No. But the price of Italian olive oil and French wine also cost the same to someone making $5k in China vs $80k in the US.

@Caffeindated @Astuterealm0 @3RenChengHu 没错，但美国人每天都啃食iPhone度日吗？

China's middle class is below the US poverty line.

@thomasunise Sure. But that’s like calling MS word open source because you can run it locally without internet. It is different than Google Docs, but it isn’t open source.

They are open weights, I have the weights downloaded right now and could spin up a server and locally host it And all my data would be private The only thing that you can’t access is their training data, but I can still locally host and fine tune the model how I want You can’t do that with OpenAI or Anthropic