My biggest concern is, what happens when countries like China, who aren't restricted by this, far outpace western countries in this technology? Do we just shrug and accept our far inferior models? LLMs are a productivity multiplier (similar to a search engine), so it'll have a large impact on the economy if licensing costs prohibit large scale training.
"Our" models are not inferior. There is plenty of data, and the next frontier is prediction-time compute and data synthesis. Shouldn't the Chinese worry that they are depressing the publication of commercial IP?
