I think it looks like regulatory capture by design. The point of even the most optimistic AI safety regulation is to restrict access, usage, progress, and slow or disable a market-driven dynamic. Any outcome from an honest initiative talking the correct steps will look like regulatory capture to some.
I agree that the generation of LLM models will not be a breakthrough, but the overall capability increases from perfecting step-by-step and decision tree styles of prompting are very promising.