Unverified50% confidenceFactExact time
The researchers conducted a four-reading audit on another four VLMs covering three tokenizer families, concluding that the Yes/No reading confusion is a protocol-level rather than model-level issue.
1
Sources
50%
Confidence
Long-term
Relevance
10/20/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedOn the long-document question-answering task, both models got all four questions right, and both clearly stated 'not in the notes' for questions whose answers were absent from the notes.62% similarUnverified微调VLM能够复用大模型已有的视觉理解与语言生成能力，在相对有限的标注数据下实现对特定文献类型的适配59% similarUnverifiedIn the test, the payload was passed word-for-word to the reading model with only each vendor's built-in synthesis functionality removed, and token counts were derived from actual bills rather than estimates.57% similarUnverified系统在进行页数校验、Track分配和评审组织时读取的是Submission层级的Paper Type，而非作者表单中的字段57% similarUnverifiedFTS5 依赖字面匹配，当查询词与存储内容用词不同但语义相近时可能漏检，是纯全文方案的固有局限56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1074069API
curl https://kongchang.com/api/v1/knowledge/claims/1074069MCP
get_claim(id=1074069)