AI agents crack a mystery in neural nets

AI agents may disprove polysemanticity, a key theory about how neurons encode concepts, potentially transforming interpretability research.