🪔 AI Agent From Google DeepMind Makes Breakthroughs in Math and Computer Science
Google DeepMind has introduced AlphaEvolve, a new AI agent that autonomously finds solutions to fundamental problems in mathematics and computer science.
It's built on a combination of AI models: the fast Gemini Flash generates a wide variety of ideas, the more powerful Gemini Pro provides critical depth, and separate evaluators assess each solution and help craft the prompt for the next iteration.
How AlphaEvolve Is Changing Science:
1️⃣ Algorithms developed by AlphaEvolve have already helped Google optimize data-center scheduling, improve chip design, and even speed up training AI models behind AlphaEvolve itself.
2️⃣ Tasked with over 50 open mathematical challenges, AlphaEvolve rediscovered the best known solutions in 75% of cases and surpassed current human solutions in 20%. It established a new lower bound for the kissing-number problem in 11D and improved one of the most efficient matrix-multiplication algorithms.
3️⃣ In principle, AlphaEvolve can be applied to any problem expressible as an algorithm. Google DeepMind hopes the agent will drive breakthroughs in materials discovery, drug development, and other fields.
Google DeepMind is now building a friendly UI for the model. Selected academic researchers will have early access to AlphaEvolve.
More on the topic:
➡️ The Thinking Game: The film about the CEO of DeepMind, Demis Hassabis
