Paying Less Generalization Tax: A Cross-Domain Generalization Study of RL Training for LLM Agents

Paying Less Generalization Tax: A Cross-Domain Generalization Study of RL Training for LLM Agents explores Improving cross-domain generalization in RL for LLM agents by enhancing state information richness and randomization.. Commercial viability score: 2/10 in LLM Training.