New training method lets AI models improve themselves without external teachers

Original: RISE: Recursive Improvement via Self-Extrapolating Policy Distillation

Source: arxiv.org ↗