Unverified50% confidenceFactExact time
在某些任务上，仅因随机种子不同导致的性能波动幅度可能超过某些论文声称的方法创新带来的提升
1
Sources
50%
Confidence
Long-term
Relevance
8/5/2026
First Seen
Sources
Related Claims
Unverified即便固定所有随机种子，在不同硬件或并行度下，训练结果仍可能出现被大模型放大为显著能力差异的细微偏差77% similarUnverifiedA 2018 study by the Google Brain team showed that simply changing the random seed could cause performance fluctuations of several percentage points in certain models.76% similarUnverifiedIn few-shot prompting, simply changing the order of examples can cause dramatic fluctuations in accuracy, ranging from near-random to near-optimal.71% similarUnverified权重初始化的随机性对模型性能影响最深远，数据批次打乱顺序的随机性影响最小70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/690452API
curl https://kongchang.com/api/v1/knowledge/claims/690452MCP
get_claim(id=690452)