VLMs Break When You Change the Rules

Featured

• Beyond Perception Errors: Semantic Fixation in Large Vision-Language Models score 6
  入选 HF Daily Papers; 有代码实现; 关键词(2): post-training, vision-language

Also Worth Noting

• LASA: Language-Agnostic Semantic Alignment at the Semantic Bottleneck for LLM Safety score 4
  入选 HF Daily Papers; HF 热度: 2 upvotes (+1)
• GoodPoint: Learning Constructive Scientific Paper Feedback from Author Responses score 4
  机构: NVIDIA; 关键词(1): fine-tuning
• Benchmarking Deflection and Hallucination in Large Vision-Language Models score 4
  关键词(2): vision-language, data curation; 顶会接收: ACL
• Long-Horizon Plan Execution in Large Tool Spaces through Entropy-Guided Branching score 4
  机构: Amazon; 关键词(1): reasoning
• ViLL-E: Video LLM Embeddings for Retrieval score 4
  关键词(2): fine-tuning, pre-training; 顶会接收: ACL
• Exploring Concept Subspace for Self-explainable Text-Attributed Graph Learning score 3
  机构: University of Toronto
• Loss-Driven Bayesian Active Learning score 3
  机构: Oxford
• LoSA: Locality Aware Sparse Attention for Block-Wise Diffusion Language Models score 3
  机构: Berkeley