algorithmic-discovery

Hill Sampling is a simple test-time scaling method that repeatedly samples edits to the best verified program using frozen LLMs, achieving state-of-the-art results on algorithmic problems like circle packing and Erdős' minimum-overlap problem.

This blog post introduces LEVI, a framework for AI-driven research for systems (ADRS) that reduces the cost of algorithmic discovery by using smaller models for most mutations and reserving large models for paradigm shifts, achieving 3-7x cost reduction. It argues that ADRS should be integrated into CI/CD for continuous, bespoke optimization per deployment.