New AI agent evolves algorithms for math and practical applications in computing by combining the creativity of large language models with automated evaluators

Google DeepMind has been working with “Alpha” in various forms since early “best at Chess” versions and a later version of which was able to beat the world’s best “Go” player…and they’ve continued to advance with new versions. The current “AlphaEvolve” is edging towards the even more vertical innovation graph for AI, where the AI knows how to improve itself and can do so in very rapid iterations.

Such self improving AI is considered by some to be the required step to get to Artificial General Intelligence and in short order go to achieve Artificial Super Intelligence. AGI >>> ASI

To the extent that AlphaEvolve proves to be a tool to get to AGI sooner than later, we are going to hear a lot more about Singularities, and “are we there yet?”. Which for now is still heavily speculative, but the kicker is the giant impacts of “getting there”.

AlphaEvolve, an evolutionary coding agent powered by large language models for general-purpose algorithm discovery and optimization. AlphaEvolve pairs the creative problem-solving capabilities of our Gemini modelswith automated evaluators that verify answers, and uses an evolutionary framework to improve upon the most promising ideas.