Tag
This paper develops a minimal theory for partially correlated verifier cascades in LLM harnesses, showing concave log-odds, polynomial reliability decay, blind-spot ceilings, and that decorrelation is more effective than adding more gates.
Introduces Score Broadcast and Decorrelation (SBD), a principled framework for broadcast-based credit assignment that generalizes to differentiable loss families including cross-entropy, Bregman divergences, and proper scoring rules. The work provides theoretical grounding for the three-factor learning rule and demonstrates improved performance over existing broadcast approaches on CIFAR-10 and Tiny ImageNet.