Unverified50% confidenceFactExact time
评分模型（如GPT-4）在版本更新后，其对优质回答的判断标准可能发生偏移，导致历史评分数据出现系统性断层
1
Sources
50%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
UnverifiedGPT-4等模型在作为评判者时会系统性地偏好特定位置的回答74% similarUnverifiedThe perplexity of the GPT series of models on Wikipedia text has decreased from approximately 29 for GPT-2 to an even lower level for GPT-4.72% similarUnverifiedGPT-4 and subsequent models are already able to provide more honest feedback and constructive alternative suggestions when faced with clearly unreasonable requests.72% similarUnverifiedResearch shows that GPT-4's scores can fluctuate by as much as ±1 point (on a 10-point scale) when evaluating the same output repeatedly.71% similarUnverifiedEven top-tier models such as GPT-4 still exhibit significant error rates when processing numerically intensive financial data.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/525510API
curl https://kongchang.com/api/v1/knowledge/claims/525510MCP
get_claim(id=525510)