The Bitter Lesson

If

Across seven decades of AI research the approaches that won were search and learning scaled with available computation, not the ones that encoded human insight about the domain. Built-in knowledge helps in the short run and becomes the ceiling in the long run, because unlike computation it does not get cheaper every year.

Then

“When choosing between encoding what you know and building something that learns from more data or more compute, ask which one improves on its own when the budget doubles. Bet on that one for anything with a multi-year horizon.”

Unless

Compute Is Not Free — inside a fixed budget domain knowledge is often the only lever you have, and the lesson describes the trajectory of a field rather than the right move this quarter