Improving Model Safety Behavior with Rule-Based Rewards

By /
We've developed and applied a new method leveraging Rule-Based Rewards (RBRs) that aligns models to behave safely without extensive human data collection.
Scroll to Top