Unverified50% confidenceFactExact time
Zero-shot prediction results are cached and reused across random seeds, so this part does not constitute truly independent repeated experiments.
1
Sources
50%
Confidence
Long-term
Relevance
9/21/2026
First Seen
Sources
Related Entities
Related Claims
Unverified通过记录随机种子可精确复现导致失败的特定顺序，再通过二分法逐步缩小污染源的范围76% similarUnverifiedIn the traditional discrete flow matching process, each sampling step does not evaluate the quality of the current sequence and is a blind random jump.73% similarUnverifiedThe 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.72% similarUnverifiedThe experiments used three training random seeds paired with a fixed test set split, along with a limited hyperparameter tuning budget.72% similarUnverified纯LLM探索模式在确定性任务上成功率难以保障，因每次推理存在随机性68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/937864API
curl https://kongchang.com/api/v1/knowledge/claims/937864MCP
get_claim(id=937864)