mlsanityData to 14 Oct 2026

What Prompts Don't Say: Understanding and Managing Underspecification in LLM Prompts

arXiv 2505.13360 ↗First posted 19 May 2025cs.CL cs.SEcited by 32

Not on the map. Why?

Abstract (arXiv)

Prompt underspecification is a common challenge when interacting with LLMs. In this paper, we present an in-depth analysis of this problem, showing that while LLMs can often infer unspecified requirements by default (41.1%), such behavior is fragile: Under-specified prompts are 2x as likely to regress across model or prompt changes, sometimes with accuracy drops exceeding 20%. This instability makes it difficult to reliably build LLM applications. Moreover, simply specifying all requirements does not consistently help, as models have limited instruction-following ability and requirements can conflict. Standard prompt optimizers likewise provide little benefit. To address these issues, we propose requirements-aware prompt optimization mechanisms that improve performance by 4.8% on average over baselines. We further advocate for a systematic process of proactive requirements discovery, evaluation, and monitoring to better manage prompt underspecification in practice.

Key facts. Cited by 32 papers in the map (data through 14 Oct 2026). Builds on 37 of its 67 references that are in the map. First posted on arXiv 19 May 2025.

Builds on

37 of 67 references.

References · 37

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  3. 2021Multitask Prompted Training Enables Zero-Shot Task Generalizationcited by 1,470
  4. 2023Instruction-Following Evaluation for Large Language Modelscited by 1,228
  5. 2023Enhancing Chat Language Models by Scaling High-quality Instructional Conversationscited by 839
  6. 2023DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelinescited by 832
  7. 2023Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formattingcited by 817
  8. 2021Cross-Task Generalization via Natural Language Crowdsourcing Instructionscited by 699
  9. 2020Underspecification Presents Challenges for Credibility in Modern Machine Learningcited by 496
  10. 2025LLMs Get Lost In Multi-Turn Conversationcited by 429
  11. 2023How is ChatGPT's behavior changing over time?cited by 270
  12. 2018Stress Test Evaluation for Natural Language Inferencecited by 213
25 more
  1. 2023OctoPack: Instruction Tuning Code Large Language Modelscited by 189
  2. 2024Who Validates the Validators? Aligning LLM-Assisted Evaluation of LLM Outputs with Human Preferencescited by 144
  3. 2024InFoBench: Evaluating Instruction Following Ability in Large Language Modelscited by 135
  4. 2025Multi-Agent Design: Optimizing Agents with Better Prompts and Topologiescited by 113
  5. 2023Clarify When Necessary: Resolving Ambiguity Through Interaction with LMscited by 98
  6. 2025Which Economic Tasks are Performed with AI? Evidence from Millions of Claude Conversationscited by 96
  7. 2022CLAM: Selective Clarification for Ambiguous Questions with Generative Language Modelscited by 69
  8. 2022Shopping Queries Dataset: A Large-Scale ESCI Benchmark for Improving Product Searchcited by 65
  9. 2024Aligning Language Models to Explicitly Handle Ambiguitycited by 56
  10. 2024CLAMBER: A Benchmark of Identifying and Clarifying Ambiguous Information Needs in Large Language Modelscited by 54
  11. 2024On the Worst Prompt Performance of Large Language Modelscited by 38
  12. 2025Ambig-SWE: Interactive Agents to Overcome Underspecificity in Software Engineeringcited by 38
  13. 2025Idiosyncrasies in Large Language Modelscited by 33
  14. 2024SELF-GUIDE: Better Task-Specific Instruction Following via Self-Synthetic Finetuningcited by 32
  15. 2025EvalTree: Profiling Language Model Weaknesses via Hierarchical Capability Treescited by 27
  16. 2024What Should We Engineer in Prompts? Training Humans in Requirement-Driven LLM Usecited by 24
  17. 2023(Why) Is My Prompt Getting Worse? Rethinking Regression Testing for Evolving LLM APIscited by 23
  18. 2024Prompts Are Programs Too! Understanding How Developers Build Software Containing Promptscited by 19
  19. 2024Prompting in the Wild: An Empirical Study of Prompt Evolution in Software Repositoriescited by 18
  20. 2024VibeCheck: Discover and Quantify Qualitative Differences in Large Language Modelscited by 17
  21. 2023Beyond Testers' Biases: Guiding Model Testing with Knowledge Bases using LLMscited by 9
  22. 2024Designing Staged Evaluation Workflows for LLMs: Integrating Domain Experts, Lay Users, and Model-Generated Evaluation Criteriacited by 8
  23. 2024Learning from Contrastive Prompts: Automated Optimization and Adaptationcited by 7
  24. 2025From Hazard Identification to Controller Design: Proactive and LLM-Supported Safety Engineering for ML-Powered Systemscited by 7
  25. 2021Feature Interactions on Steroids: On the Composition of ML Modelscited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 7; 2026: 25.

32 papers.

Sort:

By area:

  1. 2025Flipping the Dialogue: Training and Evaluating User Language Modelscited by 58
  2. 2026AgentNoiseBench: Benchmarking Robustness of Tool-Using LLM Agents Under Noisy Conditioncited by 23
  3. 2026Model Context Protocol (MCP) Tool Descriptions Are Smelly! Towards Improving AI Agent Efficiency with Augmented MCP Tool Descriptionscited by 17
  4. 2026When Benign Inputs Lead to Severe Harms: Eliciting Unsafe Unintended Behaviors of Computer-Use Agentscited by 14
  5. 2026Quantifying the Utility of User Simulators for Building Collaborative LLM Assistantscited by 10
  6. 2026Reframing LLM Agent Security as an Agent-Human Interaction Problemcited by 9
  7. 2025CoBRA: Programming Cognitive Bias in Social Agents Using Classic Social Science Experimentscited by 8
  8. 2025A Taxonomy of Prompt Defects in LLM Systemscited by 8
  9. 2025Beyond the Prompt: An Empirical Study of Cursor Rulescited by 7
  10. 2026Better Harnesses, Smaller Models: Building 90% Cheaper Agents via Automated Harness Adaptationcited by 7
  11. 2026Learning to Act under Noise: Enhancing Agent Robustness via Noisy Environmentscited by 6
  12. 2026Safeguarding LLM Agents from Misalignment through Provenance Analysiscited by 6
20 more
  1. 2026From Anatomy to Smells: An Empirical Study of SKILL.md in Agent Skillscited by 6
  2. 2026The Side Effects of Being Smart: Safety Risks in MLLMs' Multi-Image Reasoningcited by 3
  3. 2026Ambiguity Collapse by LLMs: A Taxonomy of Epistemic Riskscited by 3
  4. 2026Useless but Safe? Benchmarking Utility Recovery with User Intent Clarification in Multi-Turn Conversationscited by 3
  5. 2026Consistency as a Testable Property: Statistical Methods to Evaluate AI Agent Reliabilitycited by 3
  6. 2025From Articles to Code: On-Demand Generation of Core Algorithms from Scientific Publicationscited by 2
  7. 2026The Anatomy of Uncertainty in LLMscited by 2
  8. 2026Don't Make Models Guess Security and Safety: Symbolic Guardrails for Domain-Specific AI Agentscited by 2
  9. 2026Cheap Code, Costly Judgment: A Case Study on Governable Agentic Software Engineeringcited by 2
  10. 2026Power and Limitations of Aggregation in Compound AI Systemscited by 1
  11. 2026AnnoBench: A Benchmark for Visualization Annotation Generationcited by 1
  12. 2025Position: Intelligent Coding Systems Should Write Programs with Justificationscited by 0
  13. 2025Reasoning With a Star: A Heliophysics Dataset and Benchmark for Agentic Scientific Reasoningcited by 0
  14. 2026Exploring Human-in-the-Loop Themes in AI Application Development: An Empirical Thematic Analysiscited by 0
  15. 2026StoryCoder: Narrative Reformulation for Structured Reasoning in LLM Code Generationcited by 0
  16. 2026Instruction Complexity Induces Positional Collapse in Adversarial LLM Evaluationcited by 0
  17. 2026Localizing Prompt Ambiguity in Large Language Models with Probe-Targeted Attributioncited by 0
  18. 2026NL-PAC: Specification Ambiguity and Certified Minimax Risk Floors in LLM-Mediated Supervisioncited by 0
  19. 2026Automatically Evolving Prompt Guidelines for Task-Specific Optimizationcited by 0
  20. 2026Prospective Interpretation Risk: Principled Communication Control Between LLMscited by 0