PURR: Efficiently Editing Language Model Hallucinations by Denoising Language Model Corruptions

TL;DR

The paper introduces PURR, an unsupervised method that efficiently edits language model outputs to reduce hallucinations by training compact denoisers with artificially corrupted text, achieving faster and more accurate results.

Contribution

PURR is a novel unsupervised approach that trains compact editors to denoise corrupted text, significantly improving attribution and reducing hallucinations in language models.

Findings

PURR outperforms existing editing methods in attribution accuracy.

PURR achieves orders of magnitude faster inference times.

PURR effectively trains on domain-agnostic corrupted data.

Abstract

The remarkable capabilities of large language models have been accompanied by a persistent drawback: the generation of false and unsubstantiated claims commonly known as "hallucinations". To combat this issue, recent research has introduced approaches that involve editing and attributing the outputs of language models, particularly through prompt-based editing. However, the inference cost and speed of using large language models for editing currently bottleneck prompt-based methods. These bottlenecks motivate the training of compact editors, which is challenging due to the scarcity of training data for this purpose. To overcome these challenges, we exploit the power of large language models to introduce corruptions (i.e., noise) into text and subsequently fine-tune compact editors to denoise the corruptions by incorporating relevant evidence. Our methodology is entirely unsupervised and provides us with faux hallucinations for training in any domain. Our Petite Unsupervised Research and Revision model, PURR, not only improves attribution over existing editing methods based on fine-tuning and prompting, but also achieves faster execution times by orders of magnitude.