Unverified50% confidenceFactExact time
When the guidance signal deviates from the model's own sampling distribution, the policy gradient estimate incurs bias.
1
Sources
50%
Confidence
Long-term
Relevance
10/7/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe 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.78% similarUnverifiedPrediction distribution monitoring tracks statistical characteristics of model outputs (such as mean, variance, and quantile distributions); the core challenge lies in setting reasonable alert thresholds, as being too sensitive leads to alert fatigue while being too lenient allows real degradation to remain hidden for a long time77% similarVerifiedSome models carry the risk of over-optimization toward test sets, causing leaderboard scores to deviate from real-world performance (benchmark saturation).76% similarUnverified近似遗忘通过梯度上升等方法主动削弱特定数据对模型参数的影响，但效果难以量化验证76% 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75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/983186API
curl https://kongchang.com/api/v1/knowledge/claims/983186MCP
get_claim(id=983186)