📄 中文摘要
LLM 代理具有非确定性特征，除了常见的结果变异外，还存在一种更严重的失败模式，即代理在不同运行中给出逻辑上相反的答案。为了解决这一问题，构建了一个中间件层，利用来自 arXiv:2602.23271 的总方差公式和 NLI 矛盾检测方法，来识别和诊断 LLM 代理的矛盾。这种方法能够有效地分析同一查询在多次运行中的不同回答，帮助开发者更好地理解和改进 LLM 的输出一致性。
使用 NLI 和总方差检测 LLM 代理矛盾 — Python 实现
📄 English Summary
Detecting LLM Agent Contradictions Using NLI and Total Variance — A Python Implementation
LLM agents exhibit non-deterministic behavior, which is well-known. However, a more severe failure mode exists where an agent provides logically opposite answers across different runs. To address this issue, a middleware layer was developed that utilizes the Total Variance formula from arXiv:2602.23271 and NLI contradiction detection to identify and diagnose contradictions in LLM agents. This approach effectively analyzes varying responses to the same query across multiple runs, assisting developers in better understanding and improving the consistency of LLM outputs.