Unverified50% confidenceFactExact time
In information retrieval and fact-checking, judging whether a citation supports a claim is usually called entailment judgment, and using LLMs to perform the text entailment classification task is far harder than verbatim quotation.
1
Sources
50%
Confidence
Long-term
Relevance
10/17/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe paper proposes an LLM-as-a-judge external filtering approach, which uses the parametric knowledge of large language models to judge whether transcribed text is contextually plausible and thereby discards contextually implausible transcription results.70% similarUnverifiedLLMs tend to over-literally interpret examples in prompts, treating them as hard rules rather than reference semantics.69% similarUnverifiedLLM-generated text tends to overuse connective phrases such as "it is worth noting," "undeniably," and "in summary," exhibits uniform sentence length, and tends to end paragraphs with a summarizing closure.68% similarUnverifiedWhen non-textual information (such as images, tables, code, and sensor data) is input into an LLM, it often must first be serialized into textual descriptions, resulting in information loss and token waste.67% similarUnverifiedLLMs operate on numerical representations of text rather than reading the original text word by word like humans, so stylistic information (tone, rhetoric, copyright notices) is largely diluted during vectorization.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1049798API
curl https://kongchang.com/api/v1/knowledge/claims/1049798MCP
get_claim(id=1049798)