2026 06 25 papers llms are bayesian in expectation
criticl weak to strong generalization
bdh cq in context learning with recurrent latent reasoning