In-Context Learning Is Provably Bayesian Inference
ICL-as-Bayesian-inference theory: the risk-decomposition identity and coupled p-N scaling of the Bayes gap verified numerically; the posterior-variance bound honestly inconclusive under finite precision.
3supported
0falsified
3inconclusive
0not audited
268trace events
31failures preserved
claims
C1Risk Decomposition Identitysupported
C2Coupled p-N Scaling of Bayes Gapinconclusive
C3Rapid Task-Type Identification (Posterior Concentration)supported
C4Bayes Gap Upper Bound Ratesupported
C5Wasserstein Stability of Bayes Gapinconclusive
C6Posterior Variance Bound by Minimax Riskinconclusive
figure evidence
the trail
33 model calls · 95 tool runs · 18.8 min wall · append-only, failures preserved
0:00