Unverified50% confidenceTradeoffExact time
Experimental results show that rehearsal reduced the observed identity degradation but simultaneously decreased cross-domain transfer performance.
1
Sources
50%
Confidence
Long-term
Relevance
10/8/2026
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.74% similarUnverifiedThe learned transfer map reduces the average feature distance from target to source by 16.3% to 38.5% compared to the original embeddings71% similarUnverifiedExperiments show that models processed with abliteration exhibit almost no degradation on standard capability evaluations, but the probability of refusing harmful requests drops significantly.68% similarUnverifiedFor cell types or species underrepresented in the training data, the reliability of predictions from such models decreases significantly.68% 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68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/986594API
curl https://kongchang.com/api/v1/knowledge/claims/986594MCP
get_claim(id=986594)