Unverified50% confidenceBenchmarkExact time
Under the influence of multi-turn narratives, the models' final judgments shifted by an average of 25 percentage points compared to matched single-turn baselines.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedIf the accuracy per step is 95%, after 20 steps the overall accuracy will drop to approximately 36%.76% similarUnverified当相关文档被放置在上下文中间位置时，模型正确回答率相比放在首尾位置下降高达20-30个百分点73% similarVerifiedIn long-chain tasks, if each step has an error rate of approximately 5%, the overall success rate may drop to below 36% after 20 steps.71% 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71% similarUnverified同一个基座模型在不同系统提示词下的表现差异可以高达30-50个百分点69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/898709API
curl https://kongchang.com/api/v1/knowledge/claims/898709MCP
get_claim(id=898709)