Adaptive Retrieval Without Self-Knowledge? Bringing Uncertainty Back Home

TL;DR

This paper analyzes 35 adaptive retrieval methods and uncertainty estimation techniques for QA, showing that simpler uncertainty methods often outperform complex retrieval pipelines in efficiency and self-knowledge without sacrificing accuracy.

Contribution

It provides a comprehensive comparison of adaptive retrieval and uncertainty estimation methods, highlighting the effectiveness of uncertainty techniques in QA tasks.

Findings

Uncertainty estimation techniques often outperform complex retrieval pipelines.

Uncertainty methods maintain comparable QA performance with higher efficiency.

The study covers 6 datasets and 10 evaluation metrics.

Abstract

Retrieval Augmented Generation (RAG) improves correctness of Question Answering (QA) and addresses hallucinations in Large Language Models (LLMs), yet greatly increase computational costs. Besides, RAG is not always needed as may introduce irrelevant information. Recent adaptive retrieval methods integrate LLMs' intrinsic knowledge with external information appealing to LLM self-knowledge, but they often neglect efficiency evaluations and comparisons with uncertainty estimation techniques. We bridge this gap by conducting a comprehensive analysis of 35 adaptive retrieval methods, including 8 recent approaches and 27 uncertainty estimation techniques, across 6 datasets using 10 metrics for QA performance, self-knowledge, and efficiency. Our findings show that uncertainty estimation techniques often outperform complex pipelines in terms of efficiency and self-knowledge, while maintaining comparable QA performance.