Google DeepMind’s advancements in AI technology have recently reached significant milestones, including the capability to predict three-dimensional structures of nearly all known proteins. Additionally, DeepMind has greatly improved the accuracy of weather forecasting. However, the latest DeepMind innovation from the UK-based laboratory, AlphaEvolve, is a more extraordinary AI coding agent, which utilizes the same models and code it aims to refine.
By employing Gemini Flash and Gemini Pro, AlphaEvolve generates code proposals to solve specific problems. Each code snippet undergoes automated evaluations to assess accuracy, efficiency, and creativity. It retains the best-performing snippets to inform subsequent generations. Over multiple iterations, this method progressively enhances the solutions, establishing AlphaEvolve as a self-evolving AI.
This evolving AI technology has been tested and successfully used to design advanced chips, and to expedite AI training processes. Recent and notable achievements for AlphaEvolve include fine-tuning the “kissing number problem” by establishing a new lower bound in 11-dimensional space, and separately enhancing Google’s data centers’ energy efficiency by 0.7% over the past year.
The ainewsarticles.com article you just read is a brief synopsis; the original article can be found here: Read the Full Article…
