Andy Masley on Getting the AI-Related Words (and Numbers) Right

Episode 4 welcomes Andy Masley, the independent writer and researcher behind one of the most widely read Substacks applying careful, numbers-first thinking to AI. A former high school Physics teacher with a background in philosophy, Masley has built a following by taking the claims everyone repeats and checking if they hold up. Dr. Samuel Jay and Larry Sampler begin with language itself. Masley argues that the two dominant framings of AI like "it is just a tool" and "it cannot really think because it is not human," are both broken, and that the metaphors we reach for quietly shape policy, teaching, and who gets held accountable. From there the conversation moves into harder territory: what interpretability research can and cannot tell us about what is happening inside a model, why visible chain-of-thought reasoning may be fading, and why "it just predicts the next word" badly misleads people about what these systems do. The back half turns to Masley's signature beat: AI and the environment. He walks through why the viral "a bottle of water per prompt" figure is wrong by roughly two orders of magnitude, where the real concerns actually sit (the green-energy transition, local gas turbines, electricity prices), and why contextless big numbers tell you nothing about any industry. Along the way: the higher-education cheating crisis, the geopolitics of whose values get trained into frontier models, the unglamorous work of fixing broken business practices before automating them, and a closing tour through effective altruism. A wide-ranging, occasionally unnerving hour about thinking more clearly.