Weak-to-Strong Preference Optimization: Stealing Reward from Weak Aligned Model

Connections are not prepared for this record yet, or its source metadata has changed. The original record remains available while background snapshots are rebuilt.