Analysis argues a state-funded European AI model is unfeasible, noting Meta's budget dwarfs Germany's defense spending

from https://www.siliconcontinent.com/p/nineteen-thoughts-on-ai-and-europe

@Simon__Grimm You don't offer any theory for the observed differences, only an argument for giving up without bothering to understand why some succeed and others fail.

In my view, Musk and Zuckerberg seem to be failing due to cultural issues in their companies.

What's your view?

@Simon__Grimm Consider for a second that the reason Zuckerberg and Musk failed at AI is not because the problem is hard, but because they were bad at their job

@Simon__Grimm The better strategy for Europe is to push for global treaties that reduce the risk to themselves.

Even if you don't think the doomer position is reasonable, Europe won't catch up; their best bet is if models remain under some international control.

@xlr8harder Oh, I'm not opposed to privately-run labs; if smart people want to start something similar to Ilya's SSI in Switzerland, I would be stoked.

I do think the probability of its working is low, but I'm much less inclined to doubt private efforts.

@davidmanheim What power do they have to push for this?

I agree treaties and the like could maybe be good. But if one could chat with Friedrich Merz for 30 minutes, would one really tell him 'go to Washington and ask for an international treaty'?

@Simon__Grimm Yann LeCun's AMI is the closest thing to this, right? Top people, unconventional approach, European presence in Paris. How are they doing?

To poach all the best people, you need a good story. "EU industrial sovereignty" is not good enough if the competition is "build safe AGI".

@Simon__Grimm Do you think the only option is a state run lab?

But yeah, Meta and xAI have abysmal cultures. Many researchers would never work for either; while many European researchers would be happy to contribute to something at home, if they thought there was a chance.

@Simon__Grimm Not go to Washington, but yes push for action and concrete discussions and high-level multi-stakeholder talks at international fora where both the US and China will already be present.

@Simon__Grimm I think there is real value in removing the sorts of things that prevent privately-run labs from flourishing in Europe.

Capital access and regulatory uncertainty are the two biggest I'm aware of, but it would take a large scale effort to move the needle there.

@xlr8harder Would a state-run European project have better culture? How well would they handle ownership?

@ianchanning @Simon__Grimm I guess it's possible for the same person to make some good calls and some bad calls. Zuckerberg already incinerated billions with the metaverse saga, so this has precedent. Musk has run Tesla and xAI into the ground (for now) and DoGE is an embarrassment but SpaceX is doing well

@Simon__Grimm C'mon. Arguing we should not try something potentially historically consequential for our quality of life and freedom because it’s hard and because we might fail is not a great recommendation from someone who cares about progress.

Facebook is interesting given how much effort they put into FAIR. I mean they created PyTorch.

They certainly had the time, money, interest and available data.

Musk did start many years after and has added a few 100B to the SpaceX valuation with xAI so I wouldn't call that a failure.

So I'm expecting Musk to catch up, but FB is a real quandry. Calling Zuckerberg bad at his job seems unlikely, but don't have a better explanation.

@Simon__Grimm ...and if they realized ASI is actually a near-term plausible existential risk, there are a wide variety of other forms of economic and military leverage European leaders should be willing to apply.

@LudwigBald I'm all for private AI efforts!

@Simon__Grimm Leverage European leadership on the hardware side (ASML, Zeiss) to negotiate a sharing agreement.

And use local talent and give them more resources. You could do worse than giving people like @antirez access to some serious compute.

@xlr8harder @Simon__Grimm I enjoy the new profile pic immensely

@Simon__Grimm I think 1/100000 of 125b USD (so 1.25 m) is enough to build a capable post-LLM AI system for biosecure unvaccinated smart people in present day Portugal, the research is already there and increasing bit by bit:

@Simon__Grimm The argument you're making is the best example of why Europe is behind: lack of ambition. This is not how to win