Language-Based Feedback Helps AI Learn Better From Flawed Training Data

Researchers show that natural language critiques enable more robust policy learning than traditional scalar feedback methods.