Unlocking LLM Creativity in Science through Analogical Reasoning

Researchers identify mode collapse in LLM open-ended scientific solution generation and propose analogical reasoning (AR) as a mitigation strategy, mapping cross-domain structural relationships to boost output diversity. Evaluated on biomedical discovery tasks, AR offers a prompt-level intervention requiring no fine-tuning. Relevant to AI-for-science practitioners targeting hypothesis generation pipelines.