The Bitter Lesson is highly misinterpreted, and generally overrated scaling is a necessary part of intelligence. it’s also what you do when you’re out of ideas. there are many ways to spend compute. for example: - search for new axes to use compute (o3 / reasoning models, 2024) - redesign the system to eliminate bottlenecks (transformers, 2017) - improve the data (most of what drives improvements at frontier labs) - redefine the problem altogether (InstructGPT, 2022)