The Bitter Lesson
General methods that scale with compute beat hand-coded human knowledge — over the medium term.
Sutton’s observation from decades of AI research: across chess (search), Go (search plus learning), and language, the approaches that leverage more computation eventually outrun systems built on carefully hand-encoded human expertise. The bitter part is that our clever domain knowledge keeps losing to brute general methods that scale.
A useful extension: adaptive designs that iterate at scale tend to beat cleverly-designed static ones — closer to Darwin than to a blueprint.
Why it matters: prefer systems that get better with more compute and iteration over ones whose ceiling is however much you already know.