Unverified50% confidenceFactExact time
经过指令微调和RLHF训练的模型在遵循格式规范方面表现更好，有时反而会生成格式更完美的虚假引用
1
Sources
50%
Confidence
Long-term
Relevance
7/13/2026
First Seen
Sources
Open-Source MCP Server: Automatically Verifying the Authenticity of AI-Generated Paper Citations
redditr/learnmachinelearning7/10/2026
Related Claims
Unverified模型的格式方言源于指令微调阶段训练数据的格式差异，被模型深度内化形成强烈格式偏好，难以通过运行时提示词完全覆盖73% similarUnverifiedThe training objective of large language models is to generate 'text that looks plausible' rather than 'programs that are guaranteed to be logically correct,' which causes models to tend to output 'verification narratives' rather than actually executing verification after code generation.71% similarUnverified大语言模型的原生能力边界由训练数据的模态和格式决定，标准LLM的输入输出均为纯文本Token序列，天然无法处理PDF、DOCX、XLSX等二进制文件格式70% similarUnverified受限解码在保证格式正确性的同时，可能改变模型实际采样分布，导致语义质量的次优选择70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/500085API
curl https://kongchang.com/api/v1/knowledge/claims/500085MCP
get_claim(id=500085)