Bitter Lesson

The Bitter Lesson is a short 2019 essay by reinforcement learning researcher Rich Sutton, and the argument it makes. Looking back over decades of AI research, it holds that general methods such as search and learning, which improve as computation grows, have ultimately outperformed methods in which researchers built in human knowledge by hand.

The essay is often cited in support of scaling up models, data and compute, and it serves as a reference point in debates over how much human-designed structure, or inductive bias, to build into models.

This entry is based on AIPOST articles and widely known facts. If something is wrong, please send us a correction request.