SEAR trains one LLM via adversarial process rewards to explore harmful reasoning paths but flip to safe outputs, reducing over-refusal while preserving safety.
Regularizing hidden states enables learning generalizable reward model for llms
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
verdicts
UNVERDICTED 5representative citing papers
HARVE removes the component of the reward-head vector aligned with a multi-directional hacking subspace from residual streams using a small set of contrastive examples, improving robustness on RewardHackBench across eight models without fine-tuning while preserving general capability.
DynaCF dynamically downweights shortcut-sensitive samples in reward model training by tracking margin shifts under online counterfactual perturbations within the Bradley-Terry loss.
HyRe personalizes reward models at test time by reweighting an ensemble of heads trained on aggregate preferences, using few target examples to outperform uniform averaging and prior methods on RewardBench and 32 tasks.
Data-centric filtering yields an 80K preference dataset and reward models that lead RewardBench while boosting other top entries.
citing papers explorer
-
Addressing Over-Refusal in LLMs with Competing Rewards
SEAR trains one LLM via adversarial process rewards to explore harmful reasoning paths but flip to safe outputs, reducing over-refusal while preserving safety.
-
HARVE: Hacking-Aware Reward-Head Vector Editing for Robust Reward Models
HARVE removes the component of the reward-head vector aligned with a multi-directional hacking subspace from residual streams using a small set of contrastive examples, improving robustness on RewardHackBench across eight models without fine-tuning while preserving general capability.
-
DynaCF: Mitigating Shortcut Learning in Reward Models via Dynamic Counterfactual Sensitivity
DynaCF dynamically downweights shortcut-sensitive samples in reward model training by tracking margin shifts under online counterfactual perturbations within the Bradley-Terry loss.
-
Test-Time Alignment via Hypothesis Reweighting
HyRe personalizes reward models at test time by reweighting an ensemble of heads trained on aggregate preferences, using few target examples to outperform uniform averaging and prior methods on RewardBench and 32 tasks.
-
Skywork-Reward: Bag of Tricks for Reward Modeling in LLMs
Data-centric filtering yields an 80K preference dataset and reward models that lead RewardBench while boosting other top entries.