Generation Constraint Scaling Can Mitigate Hallucination

TL;DR

This paper proposes a geometry-inspired, training-free method to reduce hallucinations in large language models by scaling the readout vector, improving generation quality and efficiency.

Contribution

It introduces a novel scaling technique for memory-augmented LLMs that mitigates hallucinations without additional training or fine-tuning.

Findings

Outperforms state-of-the-art LLM editing methods

Reduces hallucinations in Wikipedia biography generation

Improves runtime efficiency

Abstract

Addressing the issue of hallucinations in large language models (LLMs) is a critical challenge. As the cognitive mechanisms of hallucination have been related to memory, here we explore hallucination for LLM that is enabled with explicit memory mechanisms. We empirically demonstrate that by simply scaling the readout vector that constrains generation in a memory-augmented LLM decoder, hallucination mitigation can be achieved in a training-free manner. Our method is geometry-inspired and outperforms a state-of-the-art LLM editing method on the task of generation of Wikipedia-like biography entries both in terms of generation quality and runtime complexity.