Google DeepMind creates super-advanced AI that can invent new algorithms – Ars Technica

🔍 Summary:

Google’s DeepMind has introduced a new AI system called AlphaEvolve, which represents a significant advancement in AI technology with potential applications in solving complex mathematical and scientific problems. AlphaEvolve builds on Google’s existing Gemini large language models by incorporating an evolutionary approach that refines algorithms through automatic evaluation and iterative improvement. This system is designed to generate and optimize solutions across various use cases, enhancing its accuracy and reliability compared to standard AI models.

AlphaEvolve has already demonstrated its utility within Google, notably improving the efficiency of Google’s data centers by optimizing the Borg cluster management system, which resulted in a 0.7 percent global reduction in computing resources—a notable saving for a company of Google’s scale. Additionally, AlphaEvolve has developed a more efficient algorithm for matrix multiplication, surpassing the long-standing Strassen algorithm from 1969, which could revolutionize the efficiency of generative AI technologies.

The AI has also contributed to advancements in Google’s hardware, specifically its next-generation Tensor processing chips, by optimizing the Verilog hardware description language to enhance chip efficiency. Although currently used exclusively within Google due to its complexity, there is potential for broader application or integration into smaller AI tools for research in the future.

DeepMind’s AlphaEvolve exemplifies a shift towards more dynamic, general-purpose AI systems capable of significant contributions across various domains, marking a step forward in the practical application of AI to real-world problems.

📌 Source: https://arstechnica.com/ai/2025/05/google-deepmind-creates-super-advanced-ai-that-can-invent-new-algorithms/