Unverified50% confidenceBenchmarkExact time
In matched comparisons, changing only the wording reversed which side of a contested issue the model supported in 16.9% of cases.
1
Sources
50%
Confidence
Long-term
Relevance
10/1/2026
First Seen
Sources
Related Entities
Related Claims
Unverified不同的分词方案会导致相同内容在不同模型间产生10%-30%的token数量差异68% similarUnverifiedThe same model may rank very differently across different evaluation platforms.63% similarUnverifiedResearch shows that approximately 15–25% of papers experience score changes after rebuttal, with positive changes slightly outnumbering negative ones62% similarUnverifiedWhen researchers run different testing frameworks using the same model, scores often fluctuate noticeably.61% similarUnverified修复前模型每约29次下一token预测中约有1次的选择与数学上应有结果不同61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/964094API
curl https://kongchang.com/api/v1/knowledge/claims/964094MCP
get_claim(id=964094)