Rich Sutton's bitter lesson states that compute beats algorithms in AI, but the author argues an even bitterer lesson exists: choosing the right task matters more than data, compute, or algorithms. The example of InstructGPT demonstrates how a smaller model optimized for instruction-following outperformed the larger GPT-3 trained on next-token prediction, showing that task selection—often requiring domain expertise outside machine learning—is crucial for practical success.
Rich Sutton argues that 70 years of AI research demonstrates that general methods leveraging massive computation ultimately outperform approaches that rely on human domain knowledge. He illustrates this through examples in chess, Go, and speech recognition, where computation-intensive techniques defeated human-knowledge-based methods despite initial skepticism from researchers.