Google Company presented AlphaEvolve is an innovative agent based on large language models, capable of creating and improving algorithms for a wide range of mathematical and applied tasks.
This development combines the generative capabilities of Gemini models with automated estimation mechanisms, using an evolutionary approach to optimize the code. As noted by Google, thanks to AlphaEvolve, it was possible to find new mathematical solutions that scientists could not discover for decades.
According to the company, the agent has already brought noticeable results: it has contributed to increasing the efficiency of data centers, optimizing the development of chips and improving the training processes of AI models, in particular those used in AlphaEvolve itself.
In the field of mathematics, the agent was able to improve methods for multiplying complex 4×4 matrices, surpassing the famous Strassen algorithm of 1969, and also set a new lower bound for the kissing number problem in 11-dimensional space.
Google plans to open the AlphaEvolve early access program to scientists. The company believes that this agent has the potential to become a useful tool not only in mathematics and computer science, but also in fields such as materials science, medicine, ecology and others.
