FUSE: Ensembling Verifiers with Zero Labeled Data

arXiv:2604.18547v1 Announce Type: cross Abstract: Verification of model outputs is rapidly emerging as a key primitive for both training and real-world deployment of large language models (LLMs). In practice, this often involves using imperfect LLM judges and reward models since ground truth acquisition can be time-consuming and expensive. We introduce Fully Unsupervised Score Ensembling (FUSE), a method for improving verification quality by ensembling verifiers without access to ground truth correctness labels. The key idea behind FUSE is to control conditional dependencies between verifiers in a manner that improves the unsupervised performance of a class of spectral algorithms from the ensembling literature. Despite requiring zero ground truth labels, FUSE typically matches or improves upon semi-supervised alternatives in test-time scaling experiments with diverse sets of generator models, verifiers, and benchmarks. In particular, we validate our method on both conventional academic benchmarks such as GPQA Diamond and on frontier, unsaturated benchmarks such as Humanity's Last Exam and IMO Shortlist questions.

Related

• PARM: Pipeline-Adapted Reward Model
• Why Don't You Know? Evaluating the Impact of Uncertainty Sources on Uncertainty Quantification in LLMs
• IUQ: Interrogative Uncertainty Quantification for Long-Form Large Language Model Generation

Source: arXiv cs.CL | 2026-04-21