Gaussian reward models

New framework improves reward modeling for diverse human preferences

Researchers have developed a new framework called Anchor-guided Variance-aware Reward Modeling to address limitations in standard reward models when dealing with diverse human preferences. This method enhances existing …