People who do not watch AI developments closely are often suspicious of an intuition that I’ll call “straight lines on graphs,” which holds that progress in AI is (not only rapid but) remarkably regular over time. 1
In my experience, most people (including today’s straight-liners) start out suspicious of “straight lines on graphs,” but eventually have the intuition beaten into them. It doesn’t come from rational debate; they don’t wake up one day and have a change of heart about the viability of extrapolating across orders of magnitude. Instead they make implicit guesses of how the future might go, then notice that “straight lines on graphs” predict the future better than their guesses. 2
Thanks for reading! Subscribe for free to receive new posts and support my work.
Unfortunately I don’t think I can easily convince you of the intuition via a blog post. But hopefully I can at least share some mental models that come out of it.
Much has been written about the “jagged frontier” of AI capabilities — the thought that the capability profiles of AIs appear very uneven relative to some notion of human capabilities.
I’m not sure I totally buy the frame, but I definitely do agree that AI capabilities are totally remarkable in some domains and far less impressive in others.