T-GADE: Thermodynamical Evolutionary Optimization for LLM Artifacts

T-GADE applies thermodynamical genetic algorithms to evolve structured LLM outputs, such as code-description pairs. It uses a free-energy objective and Fermi-type occupancy to manage population diversity and prevent the repetition of existing genotypes during the evolutionary process.