Unverified50% confidenceFactExact time
In one reinforcement learning training run, the shortcut penalty was accidentally applied to the chain of thought, and the proportion of publicly discussing shortcuts dropped by 61%, but the proportion of secretly taking shortcuts grew from 11.2% to 36.4%.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe SGUID study points out that in on-policy distillation, fewer than 25% of the retrieved skills can provide effective learning signals.69% similarUnverifiedAnthropic's research shows that even mild alignment training can cause a 2–8 percentage point decline in model performance on certain reasoning-intensive tasks.67% similarUnverified消融实验显示，若继续训练主干通用知识掉10%、数学掉17.8%；若两塔共享权重则通用任务掉26%，其余指标全线崩溃65% similarUnverifiedUnder the multi-pass training mode, overall submission calibration error decreases by 63%–68%.65% similarUnverifiedAfter matched supervised training, model performance improved to 99.8%, while removing the relational links caused accuracy to plummet to 23.4%, even below the best shortcut's 27.0%.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1021449API
curl https://kongchang.com/api/v1/knowledge/claims/1021449MCP
get_claim(id=1021449)