Active Semantic Optimization (ASO)
Training-Free GRPO for Single-Agent Adaptation
Hanzo AI Research
ASOGRPOCode GenerationCost Reduction
Abstract
Training-free adaptation via Bayesian product-of-experts (PoE) decoding with token/embedding-level experiential priors. Foundation for Zoo's DSO multi-agent coordination. Achieves 99.8% cost reduction ($18 vs $10,000+).