Trust Functions: Near-Lossless Weak-to-Strong Generalization by Learning When to Trust the Weak Teacher
For practitioners with scarce reliable labels, trust functions provide a practical method to leverage weak supervision without performance loss, with iterative amplification of gains.
The paper introduces trust functions that assign scalar scores to weak labels to filter unreliable supervision, enabling near-lossless weak-to-strong generalization across domains like world knowledge, quantitative reasoning, and strategy games, where students match or surpass ground-truth supervision.
Weak-to-strong generalization studies how to improve a strong student using supervision from a weaker teacher when reliable labels are scarce. We view this primarily as a data selection problem, where the key challenge is to identify which weak labels are reliable enough to serve as a training signal. To address this, we introduce trust functions that assign each weak label a scalar trust score and use these scores to filter weak supervision. Across several domains, including world knowledge, quantitative reasoning, and strategy games, trust filtering yields students that match and sometimes surpass ground-truth supervision, achieving near-lossless weak-to-strong generalization. Moreover, trust functions enable an iterative weak-to-strong chain that compounds gains by training a student and reusing it as the next teacher, amplifying the gains. There are several mechanisms to which advantage of trust functions can be attributed.