"Attention-Guided Reward for Reinforcement Learning-based Jailbreak against Large Reasoning Models" scored 7.0/10 predicted impact on @kurateorg — Novelty 7.5 · Clarity 7.5 · Significance 7.0