3B Matches R1 on Refusal; B Matrix Is LoRA's Bottleneck

Also Worth Noting

• Crowded in B-Space: Calibrating Shared Directions for LoRA Merging score 4
  入选 HF Daily Papers; 关键词(1): coding
• HeLa-Mem: Hebbian Learning and Associative Memory for LLM Agents score 4
  关键词(1): distillation; 顶会接收: ACL
• PRISM: Probing Reasoning, Instruction, and Source Memory in LLM Hallucinations score 4
  关键词(2): reasoning, open-source; 顶会接收: ACL
• SIF: Semantically In-Distribution Fingerprints for Large Vision-Language Models score 4
  关键词(4): quantization, distillation, fine-tuning, vision-language; 顶会接收: CVPR
• OASIS: On-Demand Hierarchical Event Memory for Streaming Video Reasoning score 4
  关键词(2): compression, reasoning; 顶会接收: CVPR
• Jailbreaking Large Language Models with Morality Attacks score 4
  关键词(1): jailbreak; 顶会接收: ACL
• CogGen: A Cognitively Inspired Recursive Framework for Deep Research Report Generation score 4
  关键词(1): open-source; 顶会接收: ACL
• Abstain-R1: Calibrated Abstention and Post-Refusal Clarification via Verifiable RL score 4
  关键词(2): fine-tuning, reasoning; 顶会接收: ACL
• EvoComp: Learning Visual Token Compression for Multimodal Large Language Models via Semantic-Guided Evolutionary Labeling score 4
  关键词(3): lightweight, compression, vision-language; 顶会接收: CVPR
• If Only My CGM Could Speak: A Privacy-Preserving Agent for Question Answering over Continuous Glucose Data score 4
  关键词(3): lightweight, deployment, reasoning; 顶会接收: ACL
• SAVE: A Generalizable Framework for Multi-Condition Single-Cell Generation with Gene Block Attention score 3
  顶会接收: ICLR
• Noise-Adaptive Diffusion Sampling for Inverse Problems Without Task-Specific Tuning score 3
  顶会接收: ICLR
• BIASEDTALES-ML: A Multilingual Dataset for Analyzing Narrative Attribute Distributions in LLM-Generated Stories score 3
  顶会接收: ACL
• Beyond Black-Box Labels: Interpretable Criteria for Diagnosing SubjectiveNLP Tasks score 3
  顶会接收: ACL
• Dynamic Emotion and Personality Profiling for Multimodal Deception Detection score 3
  顶会接收: ACL