In a post thread on X, Stanford NeuroAI Lab founder Daniel Yamins said he and Aran Nayebi have developed a new “Theory of Contravariance in NeuroAI,” built around two ideas they call weak-strong equivalence and Zippering. Yamins linked both an arXiv paper and a Substack explainer, while Nayebi summed up the claim on X this way: for sufficiently hard tasks, “the choice of alignment metric does not matter.”