Unverified50% confidenceFactExact time
2023年多项研究证明，对同一基准采用不同措辞重新提问，部分模型表现会显著下滑
1
Sources
50%
Confidence
Long-term
Relevance
7/10/2026
First Seen
Sources
Grok vs GPT vs Claude: Which AI is the Best Coding Assistant?
hackernewshackernews7/8/2026
Related Claims
UnverifiedIn long contexts, models are more easily influenced by the most recent few turns of conversation rather than the system prompt set at the beginning, a phenomenon known as instruction-following decay.73% similarUnverifiedStarting from 2024, the industry has observed diminishing returns from simply scaling up model size, particularly on tasks such as multi-step logical reasoning, mathematical proofs, and scientific discovery.70% similarUnverifiedVerbosity bias is a widely studied issue in the research community: human evaluators tend to favor longer, more nicely formatted responses, and a 2024 LMSYS study found that controlling for output length caused the Elo rankings of some models to change significantly.69% similarUnverifiedExperiments show that after only a few generations of self-training, the variance of model outputs drops sharply and the text tends toward homogeneity69% similarUnverifiedA model or prompt update may improve conversational quality while simultaneously degrading date parsing, making cross-version behavioral comparison necessary to prevent regressions from being introduced into the production environment.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/462737API
curl https://kongchang.com/api/v1/knowledge/claims/462737MCP
get_claim(id=462737)