Unverified50% confidenceFactExact time
The SGUID study points out that in on-policy distillation, fewer than 25% of the retrieved skills can provide effective learning signals.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAnthropic's research shows that even mild alignment training can cause a 2–8 percentage point decline in model performance on certain reasoning-intensive tasks.73% similarUnverifiedInvoking an existing Skill consumes far fewer tokens than generating from scratch, and the success rate improves through repeated iteration.70% similarUnverifiedThe number of records required to learn and certify useful decisions may be a quadratic order of magnitude fewer than those required to calibrate the source.68% similarUnverified少样本学习在提示词中附带少量输入-输出示例，通常1-5个67% similarUnverifiedNegative sample learning strategies have elevated the design success rate from the early 5–10% to over 50% in certain application scenarios67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1013300API
curl https://kongchang.com/api/v1/knowledge/claims/1013300MCP
get_claim(id=1013300)