Are the small model, light compute, open source models a threat?
It seems obvious that Paul Scharre, Greg Allen and the Biden administration are right:
Scharre said to Jordan on China Talk: "...compute - access to AI chips - is the key point of leverage over access to AI capabilities..." (https://link.chtbl.com/ChinaTalk); and,
Greg Allen agrees with the Biden chip strategy (https://www.csis.org/analysis/choking-chinas-access-future-ai): "In short, the Biden administration is trying to (1) strangle the Chinese AI industry by choking off access to high-end AI chips; (2) block China from designing AI chips domestically by choking off China’s access to U.S.-made chip design software; (3) block China from manufacturing advanced chips by choking off access to U.S.-built semiconductor manufacturing equipment; and (4) block China from domestically producing semiconductor manufacturing equipment by choking off access to U.S.-built components."
But how should we assess the power of the fast moving advances in small, even locally run, open source models summarized in the leaked Google memo (https://www.semianalysis.com/p/google-we-have-no-moat-and-neither)? Apparently they can be almost as good. Does all this make deploying a practical level of AI (not for autonomous applications but for assisting human performance) more achievable for the Chinese even in the face the chips embargo?
(Sorry....I'm trying Notes for the first time and can't figure out how to put this comment in the best place....)
Also, let me say that I think the "workshop" conversation you had with Paul Scharre was high value add. You started with "help our friends with AI" and he built on it with "help them with compute." Now there is a powerful geopolitical offering! Much better than Belt and Road. It would be an entirely doable by the U.S., a cloud full of Nvidia chips and LLMs and a braintrust to work on commercial and defense applications with friends and allies. And it would totally wrong-foot the CCP who could or would never do something like that.
I should have also said "potential friends and allies." Just imagine how attractive this could be to countries - fence sitters as well as current friends - that cannot aspire to compete in AI software or computing power.
Are the small model, light compute, open source models a threat?
It seems obvious that Paul Scharre, Greg Allen and the Biden administration are right:
Scharre said to Jordan on China Talk: "...compute - access to AI chips - is the key point of leverage over access to AI capabilities..." (https://link.chtbl.com/ChinaTalk); and,
Greg Allen agrees with the Biden chip strategy (https://www.csis.org/analysis/choking-chinas-access-future-ai): "In short, the Biden administration is trying to (1) strangle the Chinese AI industry by choking off access to high-end AI chips; (2) block China from designing AI chips domestically by choking off China’s access to U.S.-made chip design software; (3) block China from manufacturing advanced chips by choking off access to U.S.-built semiconductor manufacturing equipment; and (4) block China from domestically producing semiconductor manufacturing equipment by choking off access to U.S.-built components."
But how should we assess the power of the fast moving advances in small, even locally run, open source models summarized in the leaked Google memo (https://www.semianalysis.com/p/google-we-have-no-moat-and-neither)? Apparently they can be almost as good. Does all this make deploying a practical level of AI (not for autonomous applications but for assisting human performance) more achievable for the Chinese even in the face the chips embargo?
(Sorry....I'm trying Notes for the first time and can't figure out how to put this comment in the best place....)
haha, I'm trying to record a show on this this week
Oh! I'm very excited!
Also, let me say that I think the "workshop" conversation you had with Paul Scharre was high value add. You started with "help our friends with AI" and he built on it with "help them with compute." Now there is a powerful geopolitical offering! Much better than Belt and Road. It would be an entirely doable by the U.S., a cloud full of Nvidia chips and LLMs and a braintrust to work on commercial and defense applications with friends and allies. And it would totally wrong-foot the CCP who could or would never do something like that.
Be sure to keep shopping that around!
I should have also said "potential friends and allies." Just imagine how attractive this could be to countries - fence sitters as well as current friends - that cannot aspire to compete in AI software or computing power.