The Illusion of Reasoning: Exposing Evasive Data Contamination in LLMs via Zero-CoT Truncation Paper • 2605.21856 • Published 8 days ago • 7
SkillGrad: Optimizing Agent Skills Like Gradient Descent Paper • 2605.27760 • Published 3 days ago • 12
Phi: Preference Hijacking in Multi-modal Large Language Models at Inference Time Paper • 2509.12521 • Published Sep 15, 2025 • 5