Unverified50% confidenceFactExact time
Controlled experiments confirm that verbatim repetition reduces flexible learning, whereas mixing multiple forms of knowledge expression yields more robust category learning.
1
Sources
50%
Confidence
Long-term
Relevance
11/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedExperiments show that after only a few generations of self-training, the variance of model outputs drops sharply and the text tends toward homogeneity76% similarUnverifiedDuring joint fine-tuning, a lower learning rate helps stabilize policy updates and avoid destroying fine-grained behaviors that have already been learned.75% similarUnverified自顶向下的学习方式(先建立直觉再补数学)能降低遗忘率并提高学习者留存率74% similarUnverifiedAs Skill entries increase, prompts get longer and inference slows down, requiring periodic consolidation and deduplication.73% similarUnverified强化学习领域长期存在可复现性问题,即使使用相同算法和超参数,不同代码实现之间的性能差异可高达数倍73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1173954API
curl https://kongchang.com/api/v1/knowledge/claims/1173954MCP
get_claim(id=1173954)