Research
FORK reward cuts verbosity and boosts precision in rubric-based RL
The FORK reward adds a precision objective to rubric-based RL, requiring each response unit to be necessary and factually correct. Reported win rates against the base policy rise from 10.6% to 68.0% on RubricHub-Medical and from 22.1% to 74.1% on HealthBench.
Every outlet that ran it 1 article
Hugging Face PapersResearchLead
Eat Your Precision: Precision Objective Mitigates Reward Hacking in Rubric-based RL