In AI, most artifacts have a short shelf life. So I’m glad this talk I gave 2.5 years ago is still relevant to some people. The Bitter Lesson has become much more popular since then, but I’ve seen a gap between how readily people agree with it and how much it actually guides their day-to-day work. That’s partly because it’s a statement about what wins in the long run, which doesn’t easily translate into day-to-day decisions, especially under enormous pressure to deliver. What I tried to offer in this talk was a practical interpretation. That interpretation starts with acknowledging that we have to make progress with the methods, compute, data, and algorithms available today. So we add structure that enables progress now. But we should remember what justified that structure and revisit it when those conditions change. When it’s no longer needed, remove it. Adding structure while understanding when to take it out may not look "Bitter Lesson pilled" at first glance, but I think this is how we actually put the lesson into practice.

After a great time at OpenAI, we (@EdwardSun0909, @_jasonwei) recently joined @Meta Superintelligence Labs. The first month has already been so much fun building from a clean slate with a truly talent-dense team! Very excited about the compute and long term focus of the new lab