The Bitter Lesson
70 years of AI proves it: scale beats cleverness, and that stings.
Grounded in the research on the bitter lesson
Bet on compute, not cleverness
Rich Sutton dropped this essay March 13, 2019. One line: general methods that leverage computation win, by a huge margin, over methods built on human cleverness. The engine is Moore's law. Compute gets exponentially cheaper, so any approach that just eats more of it eventually laps the hand-tuned one. Stop optimizing for today's hardware. Build the thing that gets better when you throw more compute at it.
Remember Deep Blue
1997: Deep Blue beats world champion Garry Kasparov. It didn't win with grandmaster intuition coded by hand. It won with brute alpha-beta search on specialized hardware, scanning millions of positions per second. The AI community found this disappointing, not inspiring. That reaction is the tell. When raw search humiliates your elegant chess theory, the lesson lands bitter. Take the win anyway.