NeurIPS review jokes about impossible-to-falsify ideas get a TabulatPFN comeback

GaelVaroquaux · x · 2026-07-26

A researcher jokes that a good fraction of NeurIPS reviews suggest ideas that are effectively impossible to falsify empirically on a normal research budget.

The reply turns that into a technical jab: post-train an LLM on joint samples from the graphical model à la TabulatPFN. Viola. The punchline is the closing line: “Jensen is happy,” a nod to variational / graphical-model math and the gap between elegant theory and practical evidence.

Related event: LLMs Questioned on Tabular Data: Custom Graph Models and PFNs Prove Superior(9 posts)→

Original post →

More from Fun

Fun channel →