Unverified50% confidenceFactExact time
The first term (M/N)δ*(B/M) of the error rate lower bound describes the compression distortion of facts that the model has seen but stored inaccurately due to limited memory.
1
Sources
50%
Confidence
Long-term
Relevance
9/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe experiments validate that the compression distortion term is an observable failure mode independent of coverage rate by systematically varying the fact load M and effective trainable memory B.71% similarUnverified如果参数量充足的模型连几十个样本都无法记住，几乎可以断定存在硬性错误，如标签错位、损失函数计算错误、梯度未正确回传等71% similarUnverifiedThe second term (1-M/N)(1-1/K) of the error rate lower bound corresponds to the traditional coverage problem, where the model can only guess randomly for facts that never appeared during training, with a correct-guess probability of 1/K.70% similarUnverifiedIn traditional LLM inference, the unpredictability of sequence lengths leads to memory fragmentation and waste, with measured results showing that as much as 60%–80% of GPU memory may be occupied inefficiently.69% similarUnverifiedThe paper introduces rate-distortion theory into hallucination analysis, pointing out that even with perfect data, insufficient bits can lead to memory confusion.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/919748API
curl https://kongchang.com/api/v1/knowledge/claims/919748MCP
get_claim(id=919748)