lexical-sensitivity

This paper presents a large-scale analysis of prompt lexical sensitivity in large language models, revealing a scaling law for prompt performance stability and introducing an automated Prompt-Refining Agent that reduces performance variance in tasks like code generation.