Get the latest tech news

Does the Bitter Lesson Have Limits?

Recently, “the bitter lesson” is having a moment. Coined in an essay by Rich Sutton, the bitter lesson is that, “general methods that leverage computation are ultimately the most effective, and by a large margin.” Why is the lesson bitter? Sutton writes:

This views organizations as chaotic “garbage cans” where problems, solutions, and decision-makers are dumped in together, and decisions often happen when these elements collide randomly, rather than through a fully rational process. If we wish to teach agent-builders the bitter lesson, we need to get better at defining outputs concretely that are resistant to reward hacking (even if they’re not perfect, as we explored yesterday) and figure out how to represent “Garbage Can” organizations with data. It can be run basically anywhere (which is actually a growing problem in the online chess scene…) While the bitter lesson did demonstrate new mechanisms for search humans wouldn’t have anticipated, wrapping those in human-knowledge-based rules proves both more performant and practical.

Get the Android app