A Moore's Law for AI

Sure, it's fit a Moore's-Law-like pattern since 2012. Moore's Law wasn't that, though. Moore's Law was a prediction that development was going to continue to follow that path. This paper doesn't seem to predict.

With semiconductors, Moore could see that we were nowhere near fundamental limits of our ability to scale processes. With AI... are we? I certainly don't know; I'm not sure that anyone else does, either.

Scaling down to smaller features was just dealing with assembling atoms. You could see that, while there was a lot of work to do, we had room to keep going. But AI is dealing with algorithms, and how do you know whether there's still exponential room to improve?