Scoring Higher, Answering Worse: Mitigating Reward Hacking in Rubric-Based RL via Protocol-Level Rubrics
Paper • 2609.38847 • Published • 12
Hugo Fernandez
HugoFernandeznon
AI & ML interests
Working on AI projects: transformers, datasets and fine-tuning.
Recent Activity
new activity 1 day ago
YuyaoTechnology/MatterSeek:Can two users explore the same result set simultaneously? updated a model 25 days ago
HugoFernandeznon/retrieval-quantized published a model 25 days ago
HugoFernandeznon/retrieval-quantizedOrganizations
None yet
upvoted a paper 25 days ago
upvoted 3
papers about 1 month ago
False Frontiers: Diagnosing and Mitigating Co-Cheating in Self-Evolving Search Agents
Paper • 2609.39102 • Published • 150
Transformers Stop Thinking Too Early, and a Tiny LoRA Fixes It
Paper • 2609.36585 • Published • 28
The Teacher Is a Direction, Not a Destination: Extrapolating RL-Induced Representation Residuals in On-Policy Distillation
Paper • 2609.36484 • Published • 168
upvoted a paper about 2 months ago