A Richter Scale For Hard Problems

I think, Dave, you flagged one of the more interesting points. If 10 billion tokens is sort of a reference class for problem difficulty, I was speculating at some point in the future, if you fix model capability, and of course model capability per token is continuing to increase over time through iterated amplification and distillation, that at some point we'll look back in the spirit, Peter, of Solve Everything, and we'll say, oh, that hard math problem, oh, that was That was a level 9 problem, 10 to the 9 tokens. That was a level 11 problem. And we'll just have some convenient logarithmic scale to talk about all hard problems.