Unverified50% confidenceFactExact time
When questions are too easy and the model can answer them correctly almost every time, the reward signal tends toward a constant, parameter updates approach zero, and training effectively stops
1
Sources
50%
Confidence
Long-term
Relevance
9/28/2026
First Seen
Sources
Related Entities
Related Claims
Unverified训练样本末尾遗漏EOS token会导致模型无法学习何时停止生成，推理时陷入无限续写76% similarUnverified训练时不能一开始就追求快，若直接压步数模型容易陷入死循环反复生成重复字符，只做第一步的版本碰上难题会卡死，需先练准再练快75% similarUnverifiedSimply increasing model scale or extending training duration does not automatically solve the problem of hard tasks being neglected, because the root cause lies in the absence of the signal itself.74% similarUnverifiedThe closer an event is to the cutoff date, the smaller its share in the training corpus, and the less clearly the model has mastered it.73% similarUnverified当生成模型反复在自身生成的数据上训练时，输出分布会逐渐退化，低概率事件被系统性遗忘73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/955144API
curl https://kongchang.com/api/v1/knowledge/claims/955144MCP
get_claim(id=955144)