Why ‘winning’ the AI race is so hard to define

AI development is often framed as a race among countries, companies and academic researchers. But figuring out who’s actually “winning” is far from straightforward. There’s no single yardstick, and models can look more or less advanced, depending on what you measure. Rayan Krishnan—co-founder of large language model benchmarking company Vals AI and former researcher at the Stanford Artificial Intelligence Laboratory—is trying to change that by building standardized, neutral tests that evaluate how well the systems perform real-world tasks. Companies like Anthropic tout their performance on Vals’ exams for model capabilities, and the company’s studies are increasingly being used by financial and legal firms weighing AI procurement decisions. Krishnan talks to us about AI competition between the U.S. and China, and why it’s nearly impossible to determine whether a model has reached artificial general intelligence.