Combinatorial Creativity: A New Frontier in Generalization Abilities

ICLR 2026 Conference Submission22253 Authors

Keywords: creativity, transformers, llms, cognitive science

TL;DR: We study the combinatorial creativity abilities of LLMs

Abstract: Artificial intelligence (AI) systems, and large language models (LLMs) in particular, are increasingly employed for creative tasks like scientific idea generation, constituting a form of generalization from training data unaddressed by existing conceptual frameworks. Though in many ways similar to forms of compositional generalization (CG), combinatorial creativity (CC) is an \emph{open-ended} ability. Instead of evaluating for accuracy or correctness against fixed targets, which would contradict the open-ended nature of CC, we propose a theoretical framework and algorithmic task for evaluating outputs by their degrees of \textit{novelty} and \textit{utility}. From here, we make several important empirical contributions: (1) We obtain the first insights into the scaling behavior of creativity for LLMs. (2) We discover that, for fixed compute budgets, there exist optimal depths and widths for creative ability. (3) We find that the \emph{ideation-execution gap}, whereby LLMs excel at generating novel scientific ideas but struggle to ensure their practical feasibility, may be explained by a more fundamental \emph{novelty-utility tradeoff} characteristic of creativity algorithms in general. Importantly, this tradeoff remains persistent even at scale, casting doubt on the long-term creative potential of LLMs in their current form. Together, our conceptual framework and empirical findings provide a foundation for understanding and improving creativity in modern AI models, marking \emph{a new frontier in generalization abilities.}

Primary Area: foundation or frontier models, including LLMs

Submission Number: 22253