Identifying the Source of Generation for Large Language Models

TL;DR

This paper introduces a method for identifying the original source of tokens generated by large language models, enhancing transparency and reliability in AI-generated content.

Contribution

It proposes a novel token-level source identification technique using a bi-gram MLP, enabling source tracing for LLM outputs.

Findings

Token-level source identifiers are feasible for tracing document origins.

The method generalizes well across different datasets and models.

Source identification improves transparency in LLM-generated content.

Abstract

Large language models (LLMs) memorize text from several sources of documents. In pretraining, LLM trains to maximize the likelihood of text but neither receives the source of the text nor memorizes the source. Accordingly, LLM can not provide document information on the generated content, and users do not obtain any hint of reliability, which is crucial for factuality or privacy infringement. This work introduces token-level source identification in the decoding step, which maps the token representation to the reference document. We propose a bi-gram source identifier, a multi-layer perceptron with two successive token representations as input for better generalization. We conduct extensive experiments on Wikipedia and PG19 datasets with several LLMs, layer locations, and identifier sizes. The overall results show a possibility of token-level source identifiers for tracing the document, a crucial problem for the safe use of LLMs.