Smooth Exponentials for Robotics

I recently reread Dario Amodei’s essays "Machines of Loving Grace" and "Adolescence of Technology." In these essays and his interviews, Dario speaks often of a "smooth exponential" of AI capabilities, as if machine intelligence "emerge[s] spontaneously from the right combination of data and raw computation." It is sobering to contemplate what it means to experience a "smooth, unyielding increase in AI’s cognitive capabilities." Taken very literally, no single agent can really alter the smoothness of the curve, any more than a single atom can alter the course of an exergonic reaction. We could pull an all-nighter trying to rush that paper through, or we could shut down our lab and protest outside OpenAI and Anthropic offices, and progress would continue all the same1. There are just too many ways for it to happen, so it is overwhelmingly likely to happen. In scientific research, one typically expects progress to come in fits and starts. Research has a tendency to feel like a "hero’s…