Summary

  • Google’s DeepMind unit has revealed an AI agent called AlphaEvolve, which creates new algorithms and uses them to boost efficiency within the company’s data centres and chip designs.
  • Based on large language models called Gemini, AlphaEvolve evolved from the firm’s FunSearch system that specialised in discovering new mathematical solutions.
  • After more than a year of operation, the algorithm has delivered a 0.7% efficiency boost for Google’s computing resources by identifying stranded resources in data centres, as well as re-writing aspects of chip design to remove unnecessary elements.
  • It has also solved mathematical problems that had never been answered before, such as improving upon a 56-year-old matrix multiplication algorithm.
  • The technology will be made available to select academics and researchers in the near future.

By Michael Nuñez

Original Article