FLAME: Factuality-Aware Alignment for Large Language Models

TL;DR

This paper introduces a factuality-aware alignment method for large language models that reduces hallucinations and improves factual accuracy during instruction tuning and reinforcement learning.

Contribution

It identifies factors causing hallucinations in standard alignment and proposes a new alignment approach that enhances factual correctness without sacrificing instruction-following.

Findings

Factuality-aware alignment reduces hallucinations in LLM outputs.

The method maintains instruction-following capabilities.

Experiments demonstrate improved factual accuracy.

Abstract

Alignment is a standard procedure to fine-tune pre-trained large language models (LLMs) to follow natural language instructions and serve as helpful AI assistants. We have observed, however, that the conventional alignment process fails to enhance the factual accuracy of LLMs, and often leads to the generation of more false facts (i.e. hallucination). In this paper, we study how to make the LLM alignment process more factual, by first identifying factors that lead to hallucination in both alignment steps:\ supervised fine-tuning (SFT) and reinforcement learning (RL). In particular, we find that training the LLM on new knowledge or unfamiliar texts can encourage hallucination. This makes SFT less factual as it trains on human labeled data that may be novel to the LLM. Furthermore, reward functions used in standard RL can also encourage hallucination, because it guides the LLM to provide more helpful responses on a diverse set of instructions, often preferring longer and more detailed responses. Based on these observations, we propose factuality-aware alignment, comprised of factuality-aware SFT and factuality-aware RL through direct preference optimization. Experiments show that our proposed factuality-aware alignment guides LLMs to output more factual responses while maintaining instruction-following capability.