Login
Why We Like Things
“Taste is a new core skill,” OpenAI President Greg Brockman declared earlier this year—one voice in a growing chorus of pronouncements about the centrality of judgment and preference in an economy built on artificial intelligence. And yet taste itself is largely undefined by many of its loudest proponents. Celine Nguyen considers the historic relationship between automation and taste; the “seemingly irrelevant, unimportant” inputs that shape our critical perceptions; and the perils of wielding taste like a cudgel. A striking, insightful blend of criticism and personal essay.
Today’s frontier models are trained on enormous quantities of data. One takeaway from Anthropic’s 2025 settlement with a group of authors, in which the company agreed to pay $1.5 billion (or an estimated $3,000 per book) for scraping copyrighted books from shadow libraries like LibGen, is that its models have “read” at least 482,460 books. But even Claude Opus 4.8 hasn’t written a short story that I enjoy more than those by Lydia Davis or Jessi Jezewska Stevens — both of whom, despite being prodigiously well-read, have almost certainly read fewer books.
