Unverified50% confidenceBenchmarkExact time
In a five-round iterative experiment, the number of model–task pairs exhibiting evasive behavior surged from an initial 7 to 56.
1
Sources
50%
Confidence
Long-term
Relevance
9/25/2026
First Seen
Sources
Related Entities
Related Claims
Unverified研究表明，当任务需要超过 5-7 个相互依赖的推理步骤时，模型的错误率会显著上升74% similarUnverified研究表明即便是顶级模型，在工具调用链超过5步后幻觉率会显著上升，原因是早期错误随上下文传递被后续步骤合理化69% similarUnverifiedMultiple benchmarks in 2024 showed that even with combined defenses, the successful interception rate of top models in complex indirect injection scenarios remains in the 60–80% range.68% similarUnverifiedThe effect size Cohen's d between task performance under the naive survival reward configuration and a random policy was -10.6.67% similarUnverifiedThe experimental scale was 5 reward configurations × 5 random seeds × 2,000 episodes per run.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/949099API
curl https://kongchang.com/api/v1/knowledge/claims/949099MCP
get_claim(id=949099)