The model is not the metaphor
Machine learning has given us a precise language for systems that adapt: loss, reward, regularization, alignment, drift. Outside the field, those ideas often arrive flattened into prophecy or hype.
Learning From Machine Learning takes one technical idea at a time and asks what it can illuminate beyond the model. Not because people are algorithms. Because good abstractions earn their keep in more than one room.
Short is a constraint
The format leaves nowhere to hide. If the idea cannot survive plain language, it is not ready. If the metaphor outruns the mechanism, it is not true enough yet.
The best pieces begin with something technical and end somewhere human: why an overparameterized life can still generalize, how reward misspecification reaches beyond software, or why optimism sometimes requires a less linear model of the future.
What travels
The audience came for machine learning. Many stayed for a different way to describe uncertainty, ambition, and change.
That is the work: not making the technical world smaller, but making its ideas available to more people without sanding off the difficult parts.
Watch the field notes
Learning From Machine Learning - short-form notes on technical ideas that travel beyond the model.