# Characterizing LLM Abstention Behavior in Science QA with Context Perturbations

arXiv: https://arxiv.org/abs/2404.12452 · HTML: https://mlsanity.com/paper/2404.12452/

## Key facts

- Cited by 26 papers in the map (data through 14 Oct 2026).
- Builds on 28 of its 37 references that are in the map.
- First posted on arXiv 18 Apr 2024.

## Builds on (most cited first, 28 of 28)

- [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://mlsanity.com/paper/1910.10683/) (2019)
- [Self-Consistency Improves Chain of Thought Reasoning in Language Models](https://mlsanity.com/paper/2203.11171/) (2022)
- [PaLM: Scaling Language Modeling with Pathways](https://mlsanity.com/paper/2204.02311/) (2022)
- [Large Language Models are Zero-Shot Reasoners](https://mlsanity.com/paper/2205.11916/) (2022)
- [Finetuned Language Models Are Zero-Shot Learners](https://mlsanity.com/paper/2109.01652/) (2021)
- [Scaling Instruction-Finetuned Language Models](https://mlsanity.com/paper/2210.11416/) (2022)
- [Know What You Don't Know: Unanswerable Questions for SQuAD](https://mlsanity.com/paper/1806.03822/) (2018)
- [Least-to-Most Prompting Enables Complex Reasoning in Large Language Models](https://mlsanity.com/paper/2205.10625/) (2022)
- [PubMedQA: A Dataset for Biomedical Research Question Answering](https://mlsanity.com/paper/1909.06146/) (2019)
- [Large Language Models Can Be Easily Distracted by Irrelevant Context](https://mlsanity.com/paper/2302.00093/) (2023)
- [Galactica: A Large Language Model for Science](https://mlsanity.com/paper/2211.09085/) (2022)
- [Cross-Task Generalization via Natural Language Crowdsourcing Instructions](https://mlsanity.com/paper/2104.08773/) (2021)
- [Adversarial Examples for Evaluating Reading Comprehension Systems](https://mlsanity.com/paper/1707.07328/) (2017)
- [A Dataset of Information-Seeking Questions and Answers Anchored in Research Papers](https://mlsanity.com/paper/2105.03011/) (2021)
- [Can Generalist Foundation Models Outcompete Special-Purpose Tuning? Case Study in Medicine](https://mlsanity.com/paper/2311.16452/) (2023)
- [Do Large Language Models Know What They Don't Know?](https://mlsanity.com/paper/2305.18153/) (2023)
- [Adversarial GLUE: A Multi-Task Benchmark for Robustness Evaluation of Language Models](https://mlsanity.com/paper/2111.02840/) (2021)
- [Out-of-Distribution Detection and Selective Generation for Conditional Language Models](https://mlsanity.com/paper/2209.15558/) (2022)
- [Reframing Instructional Prompts to GPTk's Language](https://mlsanity.com/paper/2109.07830/) (2021)
- [Knowledge of Knowledge: Exploring Known-Unknowns Uncertainty with Large Language Models](https://mlsanity.com/paper/2305.13712/) (2023)
- [The Curious Case of Hallucinatory (Un)answerability: Finding Truths in the Hidden States of Over-Confident Large Language Models](https://mlsanity.com/paper/2310.11877/) (2023)
- [RECALL: A Benchmark for LLMs Robustness against External Counterfactual Knowledge](https://mlsanity.com/paper/2311.08147/) (2023)
- [Challenges in Information-Seeking QA: Unanswerable Questions and Paragraph Retrieval](https://mlsanity.com/paper/2010.11915/) (2020)
- [CONDAQA: A Contrastive Reading Comprehension Dataset for Reasoning about Negation](https://mlsanity.com/paper/2211.00295/) (2022)
- [Visconde: Multi-document QA with GPT-3 and Neural Reranking](https://mlsanity.com/paper/2212.09656/) (2022)
- [Open Domain Multi-document Summarization: A Comprehensive Study of Model Brittleness under Retrieval](https://mlsanity.com/paper/2212.10526/) (2022)
- [Overview of BioASQ 2021: The ninth BioASQ challenge on Large-Scale Biomedical Semantic Indexing and Question Answering](https://mlsanity.com/paper/2106.14885/) (2021)
- [Fine-Tuning Deteriorates General Textual Out-of-Distribution Detection by Distorting Task-Agnostic Features](https://mlsanity.com/paper/2301.12715/) (2023)

## Built on by (most cited first, 26 of 26)

- [Know Your Limits: A Survey of Abstention in Large Language Models](https://mlsanity.com/paper/2407.18418/) (2024)
- [Knowledge Boundary of Large Language Models: A Survey](https://mlsanity.com/paper/2412.12472/) (2024)
- [Do Language Models Mirror Human Confidence? Exploring Psychological Insights to Address Overconfidence in LLMs](https://mlsanity.com/paper/2506.00582/) (2025)
- [Agentic Abstention: Do Agents Know When to Stop Instead of Act?](https://mlsanity.com/paper/2606.28733/) (2026)
- [Over-Searching in Search-Augmented Large Language Models](https://mlsanity.com/paper/2601.05503/) (2026)
- [Causal Evidence that Language Models use Confidence to Drive Behavior](https://mlsanity.com/paper/2603.22161/) (2026)
- [Laboratory-Scale AI: Open-Weight Models are Competitive with ChatGPT Even in Low-Resource Settings](https://mlsanity.com/paper/2405.16820/) (2024)
- [Systematic Characterization of LLM Quantization: A Performance, Energy, and Quality Perspective](https://mlsanity.com/paper/2508.16712/) (2025)
- [RealFin: How Well Do LLMs Reason About Finance When Users Leave Things Unsaid?](https://mlsanity.com/paper/2602.07096/) (2026)
- [Textual Bayes: Quantifying Prompt Uncertainty in LLM-Based Systems](https://mlsanity.com/paper/2506.10060/) (2025)
- [Bridging the Knowledge-Prediction Gap in LLMs on Multiple-Choice Questions](https://mlsanity.com/paper/2509.23782/) (2025)
- [Recognizing Limits: Investigating Infeasibility in Large Language Models](https://mlsanity.com/paper/2408.05873/) (2024)
- [Can LLMs replace Neil deGrasse Tyson? Evaluating the Reliability of LLMs as Science Communicators](https://mlsanity.com/paper/2409.14037/) (2024)
- [Clarify or Answer: Reinforcement Learning for Agentic VQA with Context Under-specification](https://mlsanity.com/paper/2601.16400/) (2026)
- [The Chameleon Nature of LLMs: Quantifying Multi-Turn Stance Instability in Search-Enabled Language Models](https://mlsanity.com/paper/2510.16712/) (2025)
- [UniKnow: A Unified Framework for Reliable Language Model Behavior across Parametric and External Knowledge](https://mlsanity.com/paper/2502.13648/) (2025)
- [From Test-taking to Cognitive Scaffolding: A Pedagogical Diagnostic Benchmark for LLMs on English Standardized Tests](https://mlsanity.com/paper/2505.17056/) (2025)
- [VisionTrap: Unanswerable Questions On Visual Data](https://mlsanity.com/paper/2507.17262/) (2025)
- [ERA: Evidence-based Reliability Alignment for Honest Retrieval-Augmented Generation](https://mlsanity.com/paper/2604.20854/) (2026)
- [Prompt-Based Abstention Fails Under Misleading Context: A Controlled Study of Small Frozen RAG Models](https://mlsanity.com/paper/2608.22228/) (2026)
- [Twin Worlds: Equivariance-Based Abstention for Evidence-Grounded Reasoning](https://mlsanity.com/paper/2608.28018/) (2026)
- [Fragments to Facts: Partial-Information Fragment Inference from LLMs](https://mlsanity.com/paper/2505.13819/) (2025)
- [Can One-sided Arguments Lead to Response Change in Large Language Models?](https://mlsanity.com/paper/2602.06260/) (2026)
- [OCC-RAG: Optimal Cognitive Core for Faithful Question Answering](https://mlsanity.com/paper/2606.00683/) (2026)
- [You Frame It: How Conceptual Representations Shape LLM Detection and Reasoning about Antisemitism](https://mlsanity.com/paper/2607.04945/) (2026)
- [A Knowledge-Injection Framework for Zero-Shot Adaptation of LLMs to Delirium Prediction](https://mlsanity.com/paper/2607.20453/) (2026)

## Abstract (arXiv)

The correct model response in the face of uncertainty is to abstain from answering a question so as not to mislead the user. In this work, we study the ability of LLMs to abstain from answering context-dependent science questions when provided insufficient or incorrect context. We probe model sensitivity in several settings: removing gold context, replacing gold context with irrelevant context, and providing additional context beyond what is given. In experiments on four QA datasets with six LLMs, we show that performance varies greatly across models, across the type of context provided, and also by question type; in particular, many LLMs seem unable to abstain from answering boolean questions using standard QA prompts. Our analysis also highlights the unexpected impact of abstention performance on QA task accuracy. Counter-intuitively, in some settings, replacing gold context with irrelevant context or adding irrelevant context to gold context can improve abstention performance in a way that results in improvements in task performance. Our results imply that changes are needed in QA dataset design and evaluation to more effectively assess the correctness and downstream impacts of model abstention.