BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Samuel Schapiro (Knowledge Lab\, University of Chicago\; AI Innova tion Institute\, Stony Brook University) DTSTART:20261116T010000Z DTEND:20261116T020000Z DTSTAMP:20261109T050831Z UID:TropicalmathandML/53 DESCRIPTION:Title: Measuring Combinatorial Creativity in Large Language Models and Humans\nby Samuel Schapiro (Knowledge Lab\, University of Chicago\; A I Innovation Institute\, Stony Brook University) as part of Tropical mathe matics and machine learning\n\n\nAbstract\nCombinatorial creativity\, the process of combining familiar ideas in unfamiliar ways\, is one of the for emost engines of scientific\, technological\, and artistic progress. This work introduces Kombine\, an LLM benchmark for combinatorial creativity ta sking models to discover surprising\, original\, and useful associations\, analogies\, and blends between unrelated entities\, and to use those anal ogies and blends to invent novel concepts by re-combining the parts of exi sting ones. In a large-scale study across 35 LLMs\, we find that: (1) LLMs are 10-16 percentage points better at discovering associations and analog ies than blends\, and (2) exhibit an abstraction bottleneck on blending\, in which even frontier LLMs struggle to find a shared abstraction between inputs and instantiate a coherent blended space. (3) LLMs also exhibit inv entive homogeneity\, in which one in five concepts invented for a given it em reuses properties invented by another model. Moreover\, (4) models are 14× more likely to reuse other invented properties on blends than analogi es\, and (5) models from the same provider are 2.3× more likely to reuse properties than those from different providers. Next\, in a comparative st udy involving 75 humans on the same tasks\, we find that (6) humans score below nearly every LLM on each task and (7) both humans and LLMs find blen ding most difficult\, but association is easier than analogy for LLMs\, wh ile analogy is easier than association for humans. Lastly\, (8) humans and LLMs access distinct regions of semantic space on each task\, suggesting that combinatorial creativity could benefit from systematic human-AI colla boration to enhance conceptual diversity. In summary\, we provide a framew ork to assess progress towards LLMs\, AI systems\, and human-AI collaborat ions more broadly\, that can elicit combinatorial creativity in a setting relevant to scientific and technological innovation.\n LOCATION:https://researchseminars.org/talk/TropicalmathandML/53/ END:VEVENT END:VCALENDAR