Thermodynamical AI method improves evolving code and descriptions
T-GADE: Thermodynamical Generative-AI-Driven Evolution of LLM Artifacts
Artificial IntelligenceNeural and Evolutionary Computing
Summary
Combining evolutionary computing with large language models (LLMs) can create better sets of structured outputs like paired descriptions and code. The authors propose a method called T-GADE that uses a physics-inspired selection process to evolve these paired artifacts while keeping diversity in the population. They tested this approach on a bin-packing problem and showed it reduces waste compared to previous methods. Their results confirm that this thermodynamics-based selection helps find better solutions and reuse good results.
What this means in practice
- •For software optimization teams: Improve automated code evolution methods by integrating thermodynamical selection for more efficient solution search.
- •For operations research professionals: Enhance bin-packing and other logistics optimization by applying evolved LLM-generated heuristics with controlled diversity.
- •For ai service providers: Develop advanced AI-driven software tools that evolve code and documentation automatically with better diversity control.$Commercial implications: Enables commercial AI platforms to offer improved automated coding and optimization services with thermodynamics-enhanced evolution.
Authors
Kyoko Ogawa, Naoki Mori
Abstract
Integrating evolutionary computation and large language models (LLMs) requires control of population diversity as well as generative capability. Among LLM outputs, those with explicit structure, such as a description paired with code, are structured artifacts; we use artifact for short. We propose T-GADE, which evolves these artifacts by extending thermodynamical genetic algorithms through LLM-based genetic operators and artifact-level diversity evaluation. A common free-energy objective supports generational and steady-state updates, with Fermi-type occupancy excluding repeated genotypes and Bose-type occupancy permitting them. We establish exact one-member removal and conditions for recovering the zero-temperature survival rule of Evolution of Heuristics (EoH). On the online bin-packing task studied in the EoH paper, excess measures relative bin-count overhead above a volume lower bound. Training excess uses search instances; transfer excess uses instances with another bin capacity. Generational Bose-type T-GADE at $T=0.003$ reduced median training excess by approximately 29%, from 1.152% to 0.815%, over 20 runs per configuration (two-sided Mann-Whitney $p=0.042$, Cliff's $δ=0.378$). Validation selection among its two highest-ranked final candidates reached the same median transfer excess as EoH, 0.496%. These results demonstrate the utility of thermodynamical selection and validation-based use of retained artifacts.