Beyond Magic Words: Sharpness-Aware Prompt Evolving for Robust Large Language Models with TARE

2025-09-28

AI Summary

Large language model (LLM) prompts exhibit high sensitivity to semantically equivalent paraphrasing, leading to substantial performance fluctuations; existing methods optimize only pointwise accuracy, neglecting robustness in the discrete semantic space. Method: This work formally defines prompt “sharpness” and introduces a semantic neighborhood robustness criterion. We propose TARE, a black-box optimization framework featuring anisotropic weighting and dynamic radius adjustment, integrated with derivative-free adversarial sampling and robust selection. We further extend it to ATARE, enabling learnable neighborhood structure. Contribution/Results: Across diverse tasks, our approach significantly improves prompt stability against synonymous paraphrases while maintaining tractable computational overhead. It consistently outperforms accuracy-only prompt search baselines, establishing a new state of the art in robust prompt engineering.

Technology Category

• Natural Language Processing: Prompt Engineering / Prompting
• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

The performance of Large Language Models (LLMs) hinges on carefully engineered prompts. However, prevailing prompt optimization methods, ranging from heuristic edits and reinforcement learning to evolutionary search, primarily target point-wise accuracy. They seldom enforce paraphrase invariance or searching stability, and therefore cannot remedy this brittleness in practice. Automated prompt search remains brittle: small, semantically preserving paraphrases often cause large performance swings. We identify this brittleness as the textual sharpness of the prompt landscape. In this work, we provide the first formal treatment of textual sharpness in the discrete, semantic space of prompts, together with an operational robustness criterion over a semantic neighborhood; the design is black-box or API-only, requiring no gradients to update the model's parameters. Then we introduce TARE (Textual Sharpness-Aware Evolving), a derivative-free framework that alternates between an inner, sampling-based adversarial search that stresses a prompt with hard paraphrases and an outer, robust selection that prefers candidates whose neighborhoods remain strong. We further propose ATARE, which learns anisotropic weights to shape the semantic neighborhood and adapts its radius over time to balance exploration and fidelity. Diverse tasks evaluate our methods, whose design for minimizing textual sharpness gap leads to prompts that preserve accuracy under paraphrasing, outperforming accuracy-only prompt search while remaining computationally practical.

Problem

• Optimizing prompt robustness against semantic paraphrasing in LLMs
• Reducing performance instability caused by small prompt variations
• Developing gradient-free methods for sharpness-aware prompt evolution

Innovation

• Sharpness-Aware Prompt Evolving framework TARE
• Derivative-free adversarial search with hard paraphrases
• Anisotropic neighborhood shaping for robustness