Google DeepMind AlphaEvolve is an AI tool that generates better algorithms to solve real-world problems using the Gemini 2.0 large language model.
Unlike typical coding AIs, AlphaEvolve uses a unique scoring method to refine its solutions by generating multiple code options, evaluating them for performance, and continuing to refine until the most efficient result is produced.
AlphaEvolve has shown real-world impact by improving server job allocation, saving computing resources, cutting power usage on TPU chips, and speeding up training processes, among others.
The tool works by prompting Gemini 2.0 Flash to generate code based on a problem description, scoring each version, and evolving through an iterative process until the best solution emerges. AlphaEvolve has surpassed previous records in matrix multiplication and solved over 50 math challenges.