$\textit{Dial BeInfo for Faithfulness}$: Improving Factuality of Information-Seeking Dialogue via Behavioural Fine-Tuning

TL;DR

This paper introduces BeInfo, a behavioural fine-tuning method that significantly enhances the factual accuracy of large language models in information-seeking dialogues, reducing hallucinations and improving real-world applicability.

Contribution

The paper presents BeInfo, a novel behavioural tuning approach that improves the faithfulness of large language models in dialogue systems across multiple datasets and domains.

Findings

Models with BeInfo are more faithful to source knowledge.

BeInfo improves performance on unseen domains in zero-shot settings.

Fine-tuned models outperform GPT-4 on limited real-world dialogues.

Abstract

Factuality is a crucial requirement in information seeking dialogue: the system should respond to the user's queries so that the responses are meaningful and aligned with the knowledge provided to the system. However, most modern large language models suffer from hallucinations, that is, they generate responses not supported by or contradicting the knowledge source. To mitigate the issue and increase faithfulness of information-seeking dialogue systems, we introduce BeInfo, a simple yet effective method that applies behavioural tuning to aid information-seeking dialogue. Relying on three standard datasets, we show that models tuned with BeInfo} become considerably more faithful to the knowledge source both for datasets and domains seen during BeInfo-tuning, as well as on unseen domains, when applied in a zero-shot manner. In addition, we show that the models with 3B parameters (e.g., Flan-T5) tuned with BeInfo demonstrate strong performance on data from real `production' conversations and outperform GPT4 when tuned on a limited amount of such realistic in-domain dialogues.