Adaptation with Self-Evaluation to Improve Selective Prediction in LLMs

TL;DR

This paper introduces a novel adaptation framework with self-evaluation for large language models, enhancing their selective prediction ability by allowing abstention in uncertain cases, demonstrated through improved QA benchmark results.

Contribution

The work presents a parameter-efficient tuning approach that improves LLMs' self-evaluation and selective prediction, outperforming existing methods on QA datasets.

Findings

Improved AUACC from 91.23% to 92.63% on CoQA

Enhanced AUROC from 74.61% to 80.25% on CoQA

Outperforms state-of-the-art selective prediction methods

Abstract

Large language models (LLMs) have recently shown great advances in a variety of tasks, including natural language understanding and generation. However, their use in high-stakes decision-making scenarios is still limited due to the potential for errors. Selective prediction is a technique that can be used to improve the reliability of the LLMs by allowing them to abstain from making predictions when they are unsure of the answer. In this work, we propose a novel framework for adaptation with self-evaluation to improve the selective prediction performance of LLMs. Our framework is based on the idea of using parameter-efficient tuning to adapt the LLM to the specific task at hand while improving its ability to perform self-evaluation. We evaluate our method on a variety of question-answering (QA) datasets and show that it outperforms state-of-the-art selective prediction methods. For example, on the CoQA benchmark, our method improves the AUACC from 91.23% to 92.63% and improves the AUROC from 74.61% to 80.25%.