Improving Model Safety Behavior with Rule-Based Rewards
✨ AI Summary
🔊 جاري الاستماع
We've developed and applied a new method leveraging Rule-Based Rewards (RBRs) that aligns models to behave safely without extensive human data collection.
