PretrainRL: Alleviating Factuality Hallucination of Large Language Models at the Beginning

TL;DR

PretrainRL is a novel reinforcement learning framework integrated into the pretraining phase of large language models, aimed at reducing factual hallucinations by reshaping probability distributions and emphasizing truthful knowledge.

Contribution

It introduces a new pretraining approach that actively debiases falsehoods and enhances factual accuracy in large language models, addressing a core root of hallucinations.

Findings

PretrainRL significantly reduces factual hallucinations in LLMs.

The method outperforms existing approaches on multiple benchmarks.

Efficient negative sampling and novel metrics improve factual knowledge learning.

Abstract

Large language models (LLMs), despite their powerful capabilities, suffer from factual hallucinations where they generate verifiable falsehoods. We identify a root of this issue: the imbalanced data distribution in the pretraining corpus, which leads to a state of "low-probability truth" and "high-probability falsehood". Recent approaches, such as teaching models to say "I don't know" or post-hoc knowledge editing, either evade the problem or face catastrophic forgetting. To address this issue from its root, we propose \textbf{PretrainRL}, a novel framework that integrates reinforcement learning into the pretraining phase to consolidate factual knowledge. The core principle of PretrainRL is "\textbf{debiasing then learning}." It actively reshapes the model's probability distribution by down-weighting high-probability falsehoods, thereby making "room" for low-probability truths to be learned effectively. To enable this, we design an efficient negative sampling strategy to discover these high-probability falsehoods and introduce novel metrics to evaluate the model's probabilistic state concerning factual knowledge. Extensive experiments on three public benchmarks demonstrate that PretrainRL significantly alleviates factual hallucinations and outperforms state-of-the-art methods.