LEVI: Better ADRS Results at a Fraction of the Cost
Summary
LEVI is an LLM-based evolutionary framework that lowers ADRS costs by using smaller models for most mutations and reserving larger models for paradigm shifts. It combines stratified model allocation with improved diversity maintenance to avoid convergence on a single solution, and demonstrates strong benchmark results and significant cost savings. The article argues ADRS should be treated as a bespoke, continuous optimization process akin to CI/CD.