AI RESEARCH
RVPO: Risk-Sensitive Alignment via Variance Regularization
arXiv CS.LG
•
ArXi:2605.05750v1 Announce Type: new Current critic-less RLHF methods aggregate multi-objective rewards via an arithmetic mean, leaving them vulnerable to constraint neglect: high-magnitude success in one objective can numerically offset critical failures in others (e.g., safety or formatting), masking low-performing "bottleneck" rewards vital for reliable multi-objective alignment.