发现AI生成文本后期趋于稳定，用此特征实现高精度检测。
When AI Settles Down: Late-Stage Stability as a Signature of AI-Generated Text Detection
- 聚焦生成后期的概率波动变化，捕捉AI文本的稳定性特征。
- 在序列后半段，AI文本波动率比人类低24%至32%。
- 无需额外模型或扰动，可与现有方法互补提升检测效果。
针对零样本检测方法通常忽略自回归生成时序动态的问题，我们分析了超过12万条文本样本，发现AI生成文本在生成后期呈现显著的对数概率波动衰减（Late-Stage Volatility Decay）：随着生成推进，其概率波动迅速收敛，而人类写作则保持较高变异性。该差异在序列后半段最为明显，AI生成文本的波动率降低24%至32%。基于此现象，我们提出两个仅依赖晚期统计特征的简单指标：导数离散度（Derivative Dispersion）与局部波动率（Local Volatility）。该方法无需扰动采样或额外模型访问，在EvoBench和MAGE基准上达到当前最优性能，并展现出与现有全局方法的强大互补性。
原文摘要 · Abstract (English)
Zero-shot detection methods for AI-generated text typically aggregate token-level statistics across entire sequences, overlooking the temporal dynamics inherent to autoregressive generation. We analyze over 120k text samples and reveal Late-Stage Volatility Decay: AI-generated text exhibits rapidly stabilizing log probability fluctuations as generation progresses, while human writing maintains higher variability throughout. This divergence peaks in the second half of sequences, where AI-generated text shows 24--32\% lower volatility. Based on this finding, we propose two simple features: Derivative Dispersion and Local Volatility, which computed exclusively from late-stage statistics. Without perturbation sampling or additional model access, our method achieves state-of-the-art performance on EvoBench and MAGE benchmarks and demonstrates strong complementarity with existing global methods.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。