The problem with your "close to the metal" assertion is that this has been parroted about every iteration of LLMs thus far. They've certainly gotten better (impressively so), but again, it doesn't matter. By their very nature (whether today or ten years from now), they're a big risk at the business level which is ultimately where the rubber has to hit the road.
