Before AI can be trusted with the decisions that matter, we must understand the mechanics of its failure. We are here to pursue that understanding — rigorously, openly, in service of humanity rather than capital.
Read the argument →Industry labs produce extraordinary work, but their agendas are shaped by product roadmaps and quarterly cycles. Academic AI faces publish-or-perish incentives, fragmented funding, bureaucracy, and excessive teaching loads for top-level professors.
The result: no one is asking the deep questions — why models undergo sudden capability jumps, what governs the scaling of performance with data and parameters, and how a system should represent what it doesn't know.
Cosmologists have spent thirty years building the mathematics to extract knowledge where ground truth is unknowable. That infrastructure has never been systematically applied to AI.
We apply the tools of statistical mechanics, information geometry, and field theory to build a principled theory of how neural networks learn.
Regulators are already demanding trustworthiness guarantees the field cannot yet provide. A generation of physicists trained in petabyte-scale inference is ready to make the leap — but the institutional home does not yet exist.