This month's paper:

Shrinking the Generation-Verification Gap with Weak Verifiers

Language models are getting better at reasoning but their ability to verify their own outputs still lags behind. This paper tackles that challenge head-on by introducing Weaver, a framework that combines multiple weak verifiers into a single, stronger verifier without relying heavily on labeled data.

Weaver uses weak supervision to estimate verifier reliability, normalize inconsistent outputs, and filter low-quality signals, resulting in a unified score that better reflects true response quality. In practice, this approach significantly boosts reasoning and math task performance rivaling models several times larger, such as achieving o3-mini-level accuracy using only Llama 3.3 70B as the generator.

💡 Special Guest - Author of paper:

We’re thrilled to be joined by the Jon Saad-Falcon, Stanford PhD Candidate in Computer Science, to discuss the paper and take questions from the group.

📅 Date: November 20th

🕚 Time: 11amET

Speakers: Adam Boaz Becker - Founder, HeadOn. Jimin (Anna) Yoon - Tech Lead / Senior Software Engineer

Moderator Arthur Coleman: CEO, OnlineMatters Inc.

Join the #reading-group channel in the MLOps Community Slack to connect before and after the session.