Unverified50% confidenceFactExact time
The arXiv paper "The Cost of Compression: A Rate-Distortion Limit on Factual Hallucination" argues that even if a fact was seen by the model during training, limited memory capacity may force it to be stored only approximately, thereby producing hallucinations.
1
Sources
50%
Confidence
Long-term
Relevance
9/14/2026
First Seen
Sources
Related Entities
Related Claims
Unverified模型压缩技术（量化、剪枝、知识蒸馏）会不成比例地损害模型的事实准确性，量化后模型在事实性问答任务上的幻觉率可能显著上升74% similarUnverifiedResearch shows that hallucination rate is negatively correlated with the knowledge density of the prompt; the richer and more precise the context provided in the prompt, the lower the probability that the model will generate fabricated content.72% similarUnverifiedThe paper introduces rate-distortion theory into hallucination analysis, pointing out that even with perfect data, insufficient bits can lead to memory confusion.71% similarUnverified推理能力可通过知识蒸馏和强化学习微调压缩，但事实性知识的存储依赖参数规模，压缩会导致知识覆盖面下降和幻觉率上升70% similarVerifiedLLMs have a hallucination problem, in which they generate factually incorrect information with high confidence.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/919745API
curl https://kongchang.com/api/v1/knowledge/claims/919745MCP
get_claim(id=919745)