Diffusion-Proof: A Recipe for Formal Theorem Proving Beyond Auto-Regressive Generation

arXiv·medium signal

Applies diffusion-style generation rather than autoregressive decoding to formal theorem proving, a key benchmark for LLM math/reasoning ability. Signals continued momentum for diffusion language models in structured-reasoning domains where global consistency matters. By Ruida Wang, Rui Pan, Pengcheng Wang et al. (cs.LG).