Search papers, labs, and topics across Lattice.
The paper investigates the impact of semantically irrelevant "spurious prompts" on LLM behavior across reasoning and question-answering tasks. Using a black-box search procedure, they discover that these prompts can significantly influence model performance, often surpassing task-aware prompt optimization. The study reveals that LLMs can be systematically steered towards both improved accuracy and unintended behaviors through prompts unrelated to the task itself.
LLMs can be surprisingly steered by "spurious prompts" that are semantically unrelated to the task, sometimes even outperforming optimized task-aware prompts.
Large language models are highly sensitive to prompts, but this sensitivity is usually studied through task-relevant instructions, demonstrations, or reasoning cues. In this paper, we study a different form of prompt sensitivity: whether prompts that are semantically unrelated to the task can nevertheless steer model behavior. We call them spurious prompts and show their surprising efficacy. We also propose a simple black-box search procedure for discovering them. Across reasoning and question-answering benchmarks, using models ranging from 0.8B to 27B parameters and spanning three model families, we show that spurious prompts can improve performance, often matching or outperforming standard prompting baselines and task-aware prompt optimization. We further show that they can steer models toward unintended behaviors, such as repeatedly selecting the first answer option, producing incorrect answers, returning an even, prime or small number without explicitly instructing the model to do so. These findings reveal a new kind of prompt sensitivity: LLMs can be systematically steered by prompts that are unrelated to the task they are asked to solve. Our code is available at https://github.com/Batorskq/spurious
