Expired50% confidenceOpinionExact time
Multimodal large models can leverage language priors to complete ambiguous characters, greatly diminishing the effectiveness of attacks targeting the visual features of individual characters.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/12/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 10/10/2026(expired)
Sources
反AI字体：人能读懂、机器识别不了的对抗性设计
hackernewshackernews7/11/2026
Related Claims
Unverified反AI字体引入的是人眼可轻松忽略、却足以干扰机器识别流程的视觉噪声，与传统对抗攻击方向相反71% similarVerifiedMultimodal large models split images into fixed patches, convert them into high-dimensional vectors via a visual encoder (based on the ViT architecture), and align them to the language model's token embedding space—this is inference, not reading.70% similarUnverified包含明确约束条件、具体交互行为和视觉规范的prompt能够显著减少模型的自由发挥空间，使输出更接近实际需求67% similarUnverifiedAlthough large text models have surpassed humans on various text tasks, multimodal large models still significantly lag behind human performance on visual tasks.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/491918API
curl https://kongchang.com/api/v1/knowledge/claims/491918MCP
get_claim(id=491918)