Unverified50% confidencePredictionExact time
Brownian diffusion models may systematically underestimate the probability of extreme events in heavy-tailed data scenarios, whereas stable models can correctly extrapolate to the tail region when α=ν.
1
Sources
50%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedStandard Gaussian diffusion models have inherent limitations in capturing heavy-tailed distributions.72% similarUnverifiedThe probability distribution learned by a model is naturally biased toward high-frequency events and common patterns, resulting in a systematic underestimation of long-tail events.71% similarUnverifiedMoE模型在低并发场景下的推理延迟往往不如同等激活参数量的密集模型，更适合高吞吐量云端批处理场景70% similarUnverified对MoE模型的量化需要格外谨慎，因稀疏激活使专家参数数值分布集中，低比特量化易造成路由偏差导致能力退化69% similarUnverifiedEven when the correct context is provided, frontier models still have a certain probability of generating content that contradicts the context (the hallucination problem).68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1037354API
curl https://kongchang.com/api/v1/knowledge/claims/1037354MCP
get_claim(id=1037354)