"DeepSeekMath-V2 introduces self-verifiable mathematical reasoning for the first time. The model consists of a verifier trained to evaluate mathematical proofs to identify logical flaws and assign scores based on how rigorous the proof was" www.nature.com/articles/d41...