Unverified50% confidenceFactExact time
Vision-language models (VLMs) suffer from prompt bias: when the prompt explicitly specifies a target category, the model tends to 'find' objects matching the description in the image even when they do not actually exist.
1
Sources
50%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
Related Entities
Related Claims
Unverified多模态大模型（VLM）不是OCR，它在视觉向量上进行注意力计算并结合语言先验推断内容，对无规律哈希值会产生幻觉而非报识别失败77% similarUnverifiedUnder the paired setting, when the same target is described in text-only form or guided with text+image, the model's predictions become inconsistent, indicating that its understanding of the target may depend on the prompt form rather than true multimodal semantics.76% similarUnverifiedCurrent vision-language large models (VLMs) already possess open-vocabulary object detection, scene understanding, and reasoning capabilities, making it possible to avoid training a separate model for each new task.75% similarUnverifiedEven with chain-of-thought (CoT) prompting, models still cannot stably ignore irrelevant images, demonstrating that making a model explain its reasoning is not equivalent to making it comply with annotation constraints.71% similarUnverified模型在未收到明确指令时输出图片，意味着系统提示词设计或RLHF训练中存在激励偏差71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1033616API
curl https://kongchang.com/api/v1/knowledge/claims/1033616MCP
get_claim(id=1033616)