Empirical Findings Generalize Surprisingly Far

Previously, I argued that emergent phenomena in machine learning mean that we can't rely on current trends to predict what the future of ML will be like. In this post, I will argue that despite this, empirical findings often do generalize very far, including across "phase transitions" caused

Previously, I argued that emergent phenomena in machine learning mean that we can't rely on current trends to predict what the future of ML will be like. In this post, I will argue that despite this, empirical findings often do generalize very far, including across "phase transitions" caused by emergent behavior. This might seem like a contradiction, but actually I think divergence from current trends and empirical generalization are consistent. Findings do often generalize, but you need to think to determine the right generalization, and also about what might stop any given generalization fro

related reading

• Just Ask for Generalization | Eric Jangevjang.com
• Future ML Systems Will Be Qualitatively Differentbounded-regret.ghost.io
• To Understand Language is to Understand Generalization | Eric Jangevjang.com
• More Is Different for AIbounded-regret.ghost.io
• Scaling is subtler than it seemsberen.io
• [2604.21691] There Will Be a Scientific Theory of Deep Learningarxiv.org
• Understanding deep learning requires rethinking generalizationarxiv.org
• Deep Learning is Not So Mysterious or Differentarxiv.org
• [2503.02113] Deep Learning is Not So Mysterious or Differentarxiv.org
• We May be Surprised Again: Why I take LLMs seriously.inference.vc
• Reinforcement learning towards broadly and persistently beneficial modelsalignment.openai.com
• Do Machine Learning Models Memorize or Generalize?pair.withgoogle.com