One Word in the Bitter Lesson Misled Silicon Valley on LLMs

sebkrier · x · 2026-11-08

A repost of Dileep George's argument that much of Silicon Valley misread Sutton's Bitter Lesson as vindicating LLMs:

The misreading stems from a single poor word choice in the essay's closing lines.
Original: "We want AI agents that can discover like we can, not which contain what we have discovered. Building in our discoveries only makes it harder to see how the discovering process can be done."
He argues "Building in" should read "Feeding in" — feeding in human discoveries isn't the same as hard-coding them, so LLMs aren't automatically the bitter-lesson endpoint.

More from AGI Musings