The Flat Curve Society

Steve Yegge’s argument is that the capability curve keeps going up but most people will stop being able to see it. Frontier models have crossed into territory governments will treat like weapons, Fable’s brief shutdown being the first sign, so the genuinely superintelligent systems get locked away while everyone else works with what is left. Hence the title:

The intelligence curve is as real as the Earth is round, but just as flat from where you stand.

He adds two personal limits on top of the policy one: a demand horizon, meaning the hardest problem you actually have, and a discernment horizon, meaning your ability to tell whether the output is any good. Past those, a better model looks the same as the one you have.

Most of the essay is not doom but a training argument, and this is the part worth keeping. He cites Ezra Savard’s Netflix work sorting employees into cohorts by daily token spend: non-users at zero, single-agent users around 4M tokens a day, multi-agent users at 12 to 15M, and power users above 50M. Moving someone up a cohort took about five hours of training, and 96% were still in the second cohort six weeks later. His read is that a plateau is a gift, since stable tools are the condition for building real practice rather than re-learning every quarter. A plateau lets us set up a camp and start building.

The habit worth stealing is his back-pocket evals. Whenever I give a project to a model, and it can’t do the project, I add it to my pocket-eval list. Then every time a new model drops, it’s like Christmas. His example: no Opus-class model could write the React client for his game, and Fable did it without difficulty.