Philippe Lemoine

actually an insane thing for openai’s head of strategy to publicly say

God, I know it’s just the game for third place and that it doesn’t really matter, but you still can’t do this to me.

Dean Ball took a lot of heat this week for calling open source decelerationist. Many people read this as him being personally anti-open source. I don't think that's true. This was my interpretation of what he was saying: if open source is sufficiently capable and ubiquitous, it will eventually destroy the big labs business models. This means they would no longer have funding. To quote Ilya Sutskever from his testimony in the OpenAI vs. Elon Musk trial: 'If there is no funding, there is no big computer.' And without big computer, there can't be big model. In my opinion, what would happen next in this scenario is that the United States Government would step in, nationalize the leading labs, consolidate them into a single federal entity, and then fund it directly through the Department of War under national security. To put into perspective how easily they could do this, the proposed 2027 defense budget is $1.5 trillion. Most people who strongly support open source would probably not see that outcome as ideal, or this future as a pleasant one. This doesn't mean we should abandon open source or stop wishing for its success. It simply means we should take the middle path. There is fire on both sides of us now, and that will likely remain true for the foreseeable future.

I appreciate that Dean Ball, Head of Strategic Futures at OpenAI and former Senior Policy Advisor to the White House, is just directly saying, under his legal name, that the purpose of introducing nonsensical regulations is to hurt open source and favor incumbent corporations

Upamecano was arguably the best French player of the tournament and he is proving how much we depended on him again.

If AI companies can't capture the value deriving from demand for tokens, they will just be bought by the companies that do (the hyperscalers that are serving the models for instance), which then will finance further investments in model development out of the profits they generate thanks to AI. In practice, that's already the case to a large extent, because whether in the US or in China most of the capital used to finance model development directly or indirectly comes from the companies that own the infrastructure used to serve the models. They won't stop investing in model development as long as they think that it will improve capabilities enough to give companies that can serve the best models a competitive edge. I guess many people fear that, if the open weights model wins, people will stop investing in model development because of a collective action problem, as every company will anticipate that if they invest in model development other companies that didn't will be able to capture part of the value created by more capable models without having paid for it. But if that's the case they will simply close their models and the open weights model will not win in the end. I think right now Chinese companies have adopted that model because they're lagging behind in capabilities and that's a way to make up for that, but unless there is another benefit to making their models open weights, they will eventually close them once they have caught up. Ball seems to think that, unless the government intervenes to protect US companies from Chinese open weight models, the Chinese government may prevent that from happening by subsidizing companies that invest in model development so they will keep making those investments, but that's not what is going on at the moment. For instance most of the capital raised by Moonshot comes from private Chinese hyperscalers and, although I have no doubt that people will twist themselves into a pretzel to argue that it's not really private capital due to the Chinese government's role in the economy, that's just a far-fetched argument people only make because it supports a pre-existing conclusion. Should that eventually happen, it will just mean that China is subsidizing consumers abroad, which if you ask me would be pretty stupid of them but if they want to make the Chinese taxpayer pay so that people in the rest of the world can have nice things I don't see why I should care. People will raise concerns about censorship or whatever, but in that scenario the models are open weights, so it can easily be post-trained out of them. I know people will also say that, once China has killed the AI industry elsewhere, they can close the models and we'll be stuck with theirs, but I don't think that's realistic as in that case Western governments could just ban Chinese models to support their AI industry and frankly I think that scenario is just paranoia without any basis anyway.