New AI Model Advances the “Kissing Problem” and More

Google DeepMind's new AI model, AlphaEvolve, has advanced a long-standing mathematical challenge known as the kissing number problem. The AI increased the known lower bound for the kissing number in 11 dimensions from 592 to 593 possible spheres. This incremental step is significant as it demonstrates an AI's capacity for original scientific contributions beyond replicating existing knowledge.

Beyond abstract mathematics, AlphaEvolve has also tackled practical challenges at Google, optimizing data center operations, improving tensor processing unit design, and accelerating the training of AI models like Gemini. The AI's versatility stems from its ability to address problems expressible and verifiable through code.

This development highlights the growing potential of AI to make novel discoveries in abstract mathematics and solve complex real-world engineering problems, challenging traditional notions of AI capabilities.