Unverified50% confidenceOpinionTime unknown
The results of a single-task evaluation are subject to randomness, and the same model may produce noticeably different answers under different prompts and different random seeds.
1
Sources
50%
Confidence
Long-term
Relevance
9/12/2026
First Seen
Sources
Related Claims
UnverifiedEven with identical inputs, different random seeds or sampling parameters can lead to entirely different outputs from large models.78% similarUnverified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% similarUnverified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73% similarUnverifiedChecking consistency through random sampling yields an inflated consistency rate because most samples are simple, masking the regions that are truly problematic.73% similarUnverifiedIn few-shot prompting, simply changing the order of examples can cause dramatic fluctuations in accuracy, ranging from near-random to near-optimal.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/917840API
curl https://kongchang.com/api/v1/knowledge/claims/917840MCP
get_claim(id=917840)