paper-with-me

Papers

An Empirical Study of Reward Specification and Benchmark Reliability in GRPO-based LLM Unlearning

2026-08-18 · Rubén Balbastre, Juan Manuel Orduña, Mariano Pérez arxiv

Practical LLM unlearning is usually evaluated through two objectives: suppress target-specific knowledge and preserve non-target utility. In generative QA, this leaves a third behavior underspecified: when a target-adjacent prompt admits a broader answer without target-specific leakage, the model should answer at that level rather than leak, evade, or refuse. We study this specification problem in a controlled LoRA-GRPO RWKU setting, comparing four reward designs that span lexical suppression, anti-refusal shaping, rubric-based broad answering, and an explicit refusal contrast, with and without SFT warm-up. The experiments show that optimization success is not equivalent to behavioral unlearning: RWKU forget scores, held-out completion audits, terminal training-rollout audits, and training dynamics can point to different conclusions. We trace these disagreements to reward-hacking endpoints, policy-support limits in GRPO, benchmark probes that miss endpoint changes, and rewards that can select broad-topic answering with low semantic leakage during optimization.

📄 PDF Abstract BibTeX arXiv:2608.17804

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

Jailbreaking as a Reward Misspecification Problem

2024-06-20 · Zhihui Xie, Jiahui Gao, Lei LI, Zhenguo Li 외

The widespread adoption of large language models (LLMs) has raised concerns about their safety and reliability, particularly regarding their vulnerability to adversarial attacks. In this paper, we propose a novel perspec…

Red Teaming

Average Reward Reinforcement Learning for Omega-Regular and Mean-Payoff Objectives

2025-05-21 · Milad Kazemi, Mateo Perez, Fabio Somenzi, Sadegh Soudjani 외

Recent advances in reinforcement learning (RL) have renewed focus on the design of reward functions that shape agent behavior. Manually designing reward functions is tedious and error-prone. A principled alternative is t…

Reinforcement Learning (RL)

ARM-FM: Automated Reward Machines via Foundation Models for Compositional Reinforcement Learning

2025-10-16 · Roger Creus Castanyer, Faisal Mohamed, Pablo Samuel Castro, Cyrus Neary 외 arxiv

Reinforcement learning (RL) algorithms are highly sensitive to reward function specification, which remains a central challenge limiting their broad applicability. We present ARM-FM: Automated Reward Machines via Foundat…

Zero-shot GeneralizationReinforcement Learning

Model-Free Reinforcement Learning for Symbolic Automata-encoded Objectives

2022-02-04 · Anand Balakrishnan, Stefan Jakšić, Edgar A. Aguilar, Dejan Ničković 외

Reinforcement learning (RL) is a popular approach for robotic path planning in uncertain environments. However, the control policies trained for an RL agent crucially depend on user-defined, state-based reward functions.…

reinforcement-learningReinforcement LearningReinforcement Learning (RL)

SmartEval: A Benchmark for Evaluating LLM-Generated Smart Contracts from Natural Language Specifications

2026-05-10 · Abhinav Goel, Agostino Capponi, Alfio Gliozzo, Chaitya Shah arxiv

We introduce SmartEval, a benchmark for systematically evaluating the quality of Solidity smart contracts generated by large language models (LLMs) from natural language specifications. SmartEval provides a corpus of 9,0…