Reducing LLM Hallucinations using Epistemic Neural Networks

TL;DR

This paper proposes a novel approach using epistemic neural networks to improve uncertainty estimation in large language models, aiming to reduce hallucinations and enhance output reliability.

Contribution

First to train an epistemic neural network for next token prediction on a large language model to specifically address hallucination reduction.

Findings

Reduced hallucinations on the TruthfulQA dataset.

Improved uncertainty estimates for large language models.

Abstract

Reducing and detecting hallucinations in large language models is an open research problem. In this project, we attempt to leverage recent advances in the field of uncertainty estimation to reduce hallucinations in frozen large language models. Epistemic neural networks have recently been proposed to improve output joint distributions for large pre-trained models. ENNs are small networks attached to large, frozen models to improve the model's joint distributions and uncertainty estimates. In this work, we train an epistemic neural network on top of the Llama-2 7B model combined with a contrastive decoding feature enhancement technique. We are the first to train an ENN for the next token prediction task and explore the efficacy of this method in reducing hallucinations on the TruthfulQA dataset. In essence, we provide a method that leverages a pre-trained model's latent embeddings to reduce hallucinations.