Characterizing Truthfulness in Large Language Model Generations with Local Intrinsic Dimension

TL;DR

This paper introduces a novel method using local intrinsic dimension of model activations to characterize and predict the truthfulness of large language model outputs, addressing limitations of existing calibration approaches.

Contribution

It proposes using local intrinsic dimension as a reliable, intractable-free metric for assessing LLM truthfulness, validated across multiple question answering datasets.

Findings

LID effectively predicts LLM truthfulness.

Intrinsic dimensions relate to model layers and training.

LID offers insights into LLM internal representations.

Abstract

We study how to characterize and predict the truthfulness of texts generated from large language models (LLMs), which serves as a crucial step in building trust between humans and LLMs. Although several approaches based on entropy or verbalized uncertainty have been proposed to calibrate model predictions, these methods are often intractable, sensitive to hyperparameters, and less reliable when applied in generative tasks with LLMs. In this paper, we suggest investigating internal activations and quantifying LLM's truthfulness using the local intrinsic dimension (LID) of model activations. Through experiments on four question answering (QA) datasets, we demonstrate the effectiveness ohttps://info.arxiv.org/help/prep#abstractsf our proposed method. Additionally, we study intrinsic dimensions in LLMs and their relations with model layers, autoregressive language modeling, and the training of LLMs, revealing that intrinsic dimensions can be a powerful approach to understanding LLMs.