[rgbrenner]

US bans are based on the incorrect idea that we have control over AI. But the ban only reduces the most advanced model available from US companies. Models from other countries continue development. It's as if the US decided to hit pause on competing on AI and we're just going to let someone else win it.

The US could be using these models to fix bugs and defend out systems.. but instead, we're all waiting for open source models to exceed the best unbanned model available in the US(0), and then we can all watch while attackers -- who can use any available open source model, including banned models -- to attack every US company on the internet.

US bans are a choice: a choice to lose to China; a choice to leave US companies defenseless; a choice to reduce competitiveness of the US in software. Every time a US person or company watches someone use models they are prohibited from using to achieve something US models can't, they create opposition to this ban. I can't imagine this is a sustainable policy.

0. Currently open source models are included in consideration for the "best model in the US" -- but if they're willing to ban the best from Anthropic/OpenAI, I wouldn't necessarily assume that all open source models will always be available within the US.

[missedthecue]

I agree this will slow US development but what does "win the AI race" mean? What's the finish line? If there was a finish line, say a model that met some definition of general intelligence, wouldn't other labs eventually reach it too? Does getting there 6 months sooner actually mean anything in terms of say, the 5, 20, 50, 100 year view?

[pinnola]

This is more about having the most powerful AI company based in the US, rather than abroad. If the US pauses AI development, the rest of the world will lock in with non-US AI providers who capture the market first

[jatora]

I keep seeing this take but it doesn't seem to account for the fact that labs are still researching and iterating, even if they aren't releasing.

And your point on "Every time a US person or company watches someone use models they are prohibited from using to achieve something US models can't" makes no sense at all. What SOTA models are others able to use that US companies are not? Zero.

[WarmWash]

The public will get ~5.5 level models.

Private industry will essentially get ITAR access to better models, and other countries will get 5.5 with maybe some select partners getting access.

Its totally sensible that the US will keep strong AI to itself to boost its own industry.

(ITAR is a very low bar, but you need to be a citizen and your employer must fill out some forms.)

[danmaz74]

According to this logic, Taiwan should keep the most advanced chips to itself, the EU should keep the most advanced chip lithography machines to itself, etc etc.

[CamperBob2]

Its totally sensible that the US will keep strong AI to itself to boost its own industry.

No, it isn't. What in the world would lead you to think that will be an option?

Anything we develop can and will be duplicated elsewhere and distributed widely.

[tmpz22]

“Private industry” as defined by Musk, Ellison, Kushner and Trump.

Im so sick of winning. Im so happy for the future of the US and humanity. Yey tech bros.