Measuring, Localizing, and Ablating Alignment Signatures in LLMs
This research addresses the problem of understanding and mitigating AI-like stylistic signatures in LLMs, which is important for users and developers concerned with the detectability and naturalness of generated text.
This paper investigates the stylistic regularities introduced by post-training in language models, finding that aligned models generate text with lower human-corpus affinity and higher AI-detection rates compared to base models. The authors developed PASTA, a training-free method, which successfully reduced the AI detection rate for most aligned models across 11 models and 6 detectors.
Aligned language models often exhibit a recognizable AI-like style, yet its connection to post-training and internal representations remains poorly understood. In this work, we study whether post-training introduces or amplifies AI-like stylistic regularities and whether these regularities have a localized internal signature. To this end, we compare human text, base-model generations, and aligned-model generations under matched human-source prefixes. Aligned generations show lower human-corpus affinity and higher AI-detection rates than base generations, suggesting that post-training shifts generated text away from human-corpus style and toward detector-visible AI-like text. We then introduce PASTA (Post-training Alignment Signature Targeted Ablation), a training-free method that estimates a post-training alignment signature from aligned-base residual contrasts and ablates the corresponding direction during decoding. Across 11 aligned models and 6 AI detectors, PASTA lowers the detection rate for most aligned models; this effect transfers well across detectors and is not reproduced by random directions. Qualitative analysis suggests that PASTA generations remain relevant and coherent while exhibiting greater stylistic variation. Together, these results show that AI-like stylistic effects of post-training can be measured, localized, and causally tested through activation ablation.