The Roads Get Paved

TLDR. The returns on investment for constraining jobs, ML R&D in particular, is much greater now that companies have the capability to automate some core parts of it. Thus, they will “pave the roads”¹ to make it easier for the systems to succeed.

It’s great that we’re working on estimating when the capability will be there for drop-in remote workers but “purely text-based software engineering” is actually still very important! Not because that’s how research is done now, but because that will be the most efficient way to do research in the near future. Given the bottlenecks or relatively low AI performance on tasks involving seeing, clicking, etc., the thing that smart companies will naturally do is to continually work on constraining the “problem” of research engineering until it fits nicely within a box in which the LLMs can perform. The environment will adapt to the systems’s jagged capabilities, even if you think that the systems capabilities will remain jagged.

As noted by Tyler Cowen, the Epoch folks, and ~everyone else who is paying attention at this point, we have systems with jagged, sharp capabilities profiles. Extremely competent in some domains, and less competent in others. Whether they’ll catch up in the other domains is hotly debated, but I claim it is not load-bearing for some questions around just how transformative they will be.

Lots of great smart people will be quick to notice that even if AI progress stopped yesterday we’d get years of gains in productivity as a result, as the technology diffuses. This must be in spite of the jagged capabilities profiles. I believe that this point is, in reality, supported in part by the fact that the diffusion involves adapting the environment to the systems—paving the roads, so to speak.

I don’t actually know how the horses vs. carriages debate played out at the time, but a reasonable argument would indeed have been “horses are so much more versatile, able to handle edge cases and many more terrains than carriages, which need a well-trodden path”. And indeed all-terrain capabilities took much much longer to develop than the simpler form of carriages, so the point could have been true. But instead, we just changed the environment a ton to make use of the remarkable capability of carriages / cars.

Footnotes

1. Title analogy h/t a coworker ↩