A framework (SELLM) for extracting breakthrough, cross-disciplinary solutions from a large language model by brute-force enumeration over a MECE (Mutually Exclusive, Collectively Exhaustive) list of knowledge domains, such as International Patent Classification subclasses or the chemical elements. Each list entry becomes a role-play-prompted domain-expert persona that proposes a specialized solution, so that solutions drawing on seemingly unrelated fields are surfaced without omission. Generated solutions are assessed by similarity-based LLM-as-a-Judge scoring against a reference, by keyword counts, and by human experts. Demonstrated on OLED light-extraction and IGZO thin-film-transistor electrode problems.
A machine-learning methods paper (cs.CL/cs.AI/cs.LG) proposing SELLM (Solution Enumeration via comprehensive List and LLM), a framework for eliciting effective, cross-disciplinary solutions to intrin…