Unverified50% confidenceOpinionExact time
Vision-language models may 'hallucinate' in math grading that requires precise derivation, giving annotations that appear reasonable but are wrong.
1
Sources
50%
Confidence
Long-term
Relevance
10/20/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMultimodal large language models often rely on language priors rather than truly relevant visual evidence when answering visual reasoning questions.78% similarUnverified视觉幻觉指模型在描述图像时生成与实际视觉内容不符的信息，源于语言模型倾向于用语言先验知识填补视觉信号不明确的空白77% similarUnverifiedVisual reasoning involves implicit intermediate steps and lacks clear verification methods, making it long-standing difficult to model.75% similarUnverifiedThe paper proposes that the deficiencies of multimodal large models in spatial, embodied, physical, and temporal reasoning may all stem from the same missing capability—visual transition reasoning.72% similarUnverifiedLanguage models are not inherently good at processing pixel-level images, and the state abstraction of intermediate layers and the design of the action space directly determine whether it can perform meaningful operations.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1076348API
curl https://kongchang.com/api/v1/knowledge/claims/1076348MCP
get_claim(id=1076348)