Unverified50% confidenceBenchmarkExact time
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
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
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.72% similarVerifiedIn long-chain tasks, if each step has an error rate of approximately 5%, the overall success rate may drop to below 36% after 20 steps.68% similarUnverified发布的checkpoint按块大小16训练，若将采样块大小改为64，HumanEval从76%跌至约20%，GSM8K从高位跌到2.2%67% similarUnverified实验数据显示，经过5-10轮迭代训练后，模型在低频词汇和长尾知识上的覆盖率下降超过40%，而输出文本在表面流畅性上几乎没有可察觉的退化67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/565267API
curl https://kongchang.com/api/v1/knowledge/claims/565267MCP
get_claim(id=565267)