Know Your Limits: A Survey of Abstention in Large Language Models

2024-07-25

AI Summary

This paper systematically investigates active abstention in large language models (LLMs) to mitigate hallucination and enhance safety and controllability. Addressing the lack of a unified analytical perspective in prior work, we propose the first three-dimensional framework—integrating query characteristics, model uncertainty, and human values. We develop a comprehensive methodology comprising confidence calibration, uncertainty modeling, value-aligned constraints, and dynamic refusal policies, supported by a multi-dimensional evaluation benchmark and metrics. Our analysis clarifies the applicability boundaries and limitations of mainstream approaches, establishes the first structured taxonomy of abstention techniques, and demonstrates that abstention functions as a cross-task, generalizable meta-capability. We further identify key research directions—including domain-adaptive optimization—and advocate a paradigm shift from ad hoc practice toward evaluable, generalizable, and value-aligned safety enhancement.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Abstention, the refusal of large language models (LLMs) to provide an answer, is increasingly recognized for its potential to mitigate hallucinations and enhance safety in LLM systems. In this survey, we introduce a framework to examine abstention from three perspectives: the query, the model, and human values. We organize the literature on abstention methods, benchmarks, and evaluation metrics using this framework, and discuss merits and limitations of prior work. We further identify and motivate areas for future research, such as whether abstention can be achieved as a meta-capability that transcends specific tasks or domains, and opportunities to optimize abstention abilities in specific contexts. In doing so, we aim to broaden the scope and impact of abstention methodologies in AI systems.

Problem

• Examine abstention in LLMs
• Mitigate hallucinations and enhance safety
• Identify future research on abstention

Innovation

• Abstention framework for LLMs
• Examines query, model, human values
• Meta-capability across tasks