An Opinionated Guide to ML Research by John Schulman #73

An Opinionated Guide to ML Research

From GPT-5.6 Terra:

Schulman’s essay argues that strong ML research comes from choosing high-upside problems, making steady progress on them, and deliberately expanding your skills. He favors goal-driven work—pursuing a concrete capability or outcome—because it creates a distinctive perspective and motivation, while cautioning that solutions should remain general rather than overfit to one demo. He recommends developing research taste through critical reading, experienced collaborators, and reflection; keeping a daily experiment notebook with regular reviews; resisting the temptation to abandon promising projects too quickly while reserving limited time for exploration; and investing continually in fundamentals through textbooks, theses, papers, and reimplementations. The central message is to pair ambitious long-term goals with disciplined, incremental execution and broad technical growth.