Approximating Softmax in Pretrained LLMs: Model Sensitivity and Kernel AccelerationResearchSep 30, 2026
One Geometry, Different Outcomes: Readout-Dependent Effects of the Modality Gap in Vision-Language ModelsResearchSep 30, 2026
Beyond Symmetric Agents: Cognitive Diversity and Multi-Agent Debate in Small Language ModelsResearchSep 30, 2026
CLIMB: A Clinical Multimorbidity Benchmark for Diagnosing Co-occurring Conditions through Multiturn ConversationsResearchSep 30, 2026
You Only Edit Once: Incentivizing In-Context Capability of LLMs via Local Demonstration RefinementResearchSep 30, 2026