From Model Multiplicity to Prompt Multiplicity: Emerging Arbitrariness Concerns in the Age of Generative AI

This extended abstract introduces the concept of prompt multiplicity—the phenomenon where different prompts, despite yielding similar overall performance from generative AI models, result in significantly divergent individual outputs. Building on the well-studied notion of model multiplicity, the work argues that the shift toward prompt-centric design in large language models has introduced new forms of arbitrariness that impact transparency, fairness, and robustness. The abstract highlights how prompt multiplicity can complicate evaluation and auditing processes, underscoring the need for new metrics and methodologies to ensure accountable and secure deployment of generative AI systems.