Token Prediction as Implicit Classification to Identify LLM-Generated Text
Yutian Chen, Hao Kang, Vivian Zhai, Liangze Li, Rita Singh, Bhiksha Raj
TL;DR
The paper tackles the challenge of not only detecting machine-generated text but also attributing it to its originating LLM. It reframes source attribution as a next-token prediction task by fine-tuning a T5 model (T5-Sentinel) with reserved label tokens, avoiding extra classifiers. Empirical results show superior multi-class and binary detection performance over a classifier-based baseline (T5-Hidden) and existing detectors, supported by interpretability analyses (Integrated Gradients and t-SNE) that reveal reliance on linguistic structure rather than superficial cues. The OpenLLMText dataset of ~340k samples enables robust evaluation and provides a practical resource for future provenance detection research, though limitations related to source-bias and generalization to non-native text remain important considerations.
Abstract
This paper introduces a novel approach for identifying the possible large language models (LLMs) involved in text generation. Instead of adding an additional classification layer to a base LM, we reframe the classification task as a next-token prediction task and directly fine-tune the base LM to perform it. We utilize the Text-to-Text Transfer Transformer (T5) model as the backbone for our experiments. We compared our approach to the more direct approach of utilizing hidden states for classification. Evaluation shows the exceptional performance of our method in the text classification task, highlighting its simplicity and efficiency. Furthermore, interpretability studies on the features extracted by our model reveal its ability to differentiate distinctive writing styles among various LLMs even in the absence of an explicit classifier. We also collected a dataset named OpenLLMText, containing approximately 340k text samples from human and LLMs, including GPT3.5, PaLM, LLaMA, and GPT2.