Hmm… I’m not sure I agree with this. Open source Chinese models are pretty close to the performance ofChatGPT and Claude. And it seems to be pretty easy to distill a frontier model by training on its outputs.
I also think there is another risk from AI which is under discussed which is massive centralization of power enabled by AI. If the Yuskowskian model of AI danger proves incorrect and you can in fact control narrowly superhuman AI well enough to get it to do your alignment homework, the next biggest threat is that one small group of people control everything forever.
Distillation can help to catch up but not exceed the frontier, and as the frontier pulls ahead the models can be used by intelligence agencies to hamper the supply chain to build and run the datacenters for AI training and inference in other nations.
Since centralized institutions and entities inevitably outperform decentralization, the future being controlled by a small handful of people or AIs was already the destination. The only thing to prevent is mass extinction, and ensuring those in control are the good guys, which means America.
I’m curious are people disagreeing with the points about centralisation > decentralisation, or the points about the usa being the best/better than china. Assuming that USA or china is the country to develop TAI, (free feel to correct that assumption if you disagree about the likelihood), as a non-american/non-chinese person I’m not happy with either option but still think the USA has the edge in creating a better post-TAI world than China would. I wouldn’t have thought this claim was so blantently false as to gather 23 disagree points.
Hmm… I’m not sure I agree with this. Open source Chinese models are pretty close to the performance ofChatGPT and Claude. And it seems to be pretty easy to distill a frontier model by training on its outputs.
I also think there is another risk from AI which is under discussed which is massive centralization of power enabled by AI. If the Yuskowskian model of AI danger proves incorrect and you can in fact control narrowly superhuman AI well enough to get it to do your alignment homework, the next biggest threat is that one small group of people control everything forever.
Distillation can help to catch up but not exceed the frontier, and as the frontier pulls ahead the models can be used by intelligence agencies to hamper the supply chain to build and run the datacenters for AI training and inference in other nations.
Since centralized institutions and entities inevitably outperform decentralization, the future being controlled by a small handful of people or AIs was already the destination. The only thing to prevent is mass extinction, and ensuring those in control are the good guys, which means America.
I’m curious are people disagreeing with the points about centralisation > decentralisation, or the points about the usa being the best/better than china. Assuming that USA or china is the country to develop TAI, (free feel to correct that assumption if you disagree about the likelihood), as a non-american/non-chinese person I’m not happy with either option but still think the USA has the edge in creating a better post-TAI world than China would. I wouldn’t have thought this claim was so blantently false as to gather 23 disagree points.