Google's VeriHarness Turns Consensus on Its Head for LLM Verification

Google's VeriHarness challenges the "majority consensus means correct" assumption in agent sampling, showing consistent rollouts can hide shared errors. By turning the base LLM into a verifier with disagreement resolvers, it boosts long-horizon task verification by 6+ points and open-sources 26,000 rollouts.

2026-10-04 ~ 2026-10-05 · 3 related posts