Eugene Yan writes that he once assumed maximum effort produced maximum results. After learning about the 85% rule, he argues the opposite: giving 100% can actually yield less than 85% of the possible outcome. The essay is a personal reflection on productivity, not a technical ML paper.
Still, the idea resonates with machine learning practice. A model trained to perfect accuracy on its training set often performs worse on unseen data—a phenomenon known as overfitting. Similarly, a team that optimizes every sprint to the limit may sacrifice adaptability and long-term velocity.
Yan's rule is not a direct prescription for model training, but it offers a useful analogy. Knowing when to stop optimizing—whether a model or a work week—can be as important as pushing hard. The 85% rule is a reminder that restraint is sometimes the more effective strategy.