Stanford researchers open-source LLM-as-a-Verifier framework, claiming state-of-the-art results on SWE-Bench Verified · Digg