Google AI chatbot solves advanced math and science problems

Google DeepMind’s AlphaEvolve represents a significant advancement in applying large language models to solve complex problems in mathematics and computer science. By combining an LLM’s creative capabilities with rigorous evaluation algorithms, this general-purpose AI system has already tackled longstanding mathematical challenges and delivered practical efficiencies for Google’s computing infrastructure. Unlike previous AI scientific tools that were custom-built for specific tasks, AlphaEvolve’s general-purpose design signals a potential shift toward more versatile AI systems that can generate novel solutions across multiple domains.

The big picture: Google DeepMind has created AlphaEvolve, an AI system that uses chatbot models to solve complex problems in mathematics and computer science.

• The system combines a large language model’s creative abilities with algorithms that evaluate and refine the model’s suggested solutions.
• DeepMind released details about AlphaEvolve in a white paper on May 14, 2025, showcasing its ability to tackle both theoretical problems and practical computing challenges.

Why this matters: AlphaEvolve represents one of the first successful demonstrations of general-purpose language models making genuine scientific discoveries and practical improvements.

• Unlike specialized AI tools such as AlphaFold, which were custom-designed for specific tasks, AlphaEvolve can work across various domains using the same underlying approach.
• The system has already delivered tangible benefits to Google, saving 0.7% of the company’s total computing resources.

How it works: AlphaEvolve functions as an “agent” system where multiple AI models interact to solve problems.

• Users input a question, evaluation criteria, and a suggested solution, after which the language model proposes thousands of potential modifications.
• An evaluator algorithm then assesses these modifications against the defined metrics to identify the most effective solutions.

Key achievements: The system has already made notable contributions to both theoretical mathematics and practical computing challenges.

• AlphaEvolve discovered a matrix multiplication method that outperforms the fastest-known approach developed by mathematician Volker Strassen in 1969.
• The system helped improve the design of Google’s next generation of tensor processing units, specialized chips for AI applications.

Expert reactions: Outside researchers see significant potential in AlphaEvolve while expressing some reservations about its current implementation.

• Mario Krenn of the Max Planck Institute called the paper “quite spectacular,” highlighting AlphaEvolve’s unique position as a general-purpose LLM making new discoveries.
• Simon Frieder from the University of Oxford noted that while impressive, the system will likely apply to only a “narrow slice” of problems that can be addressed through code.
• Huan Sun from Ohio State University cautioned that the system’s capabilities should be viewed skeptically until tested by the broader scientific community.

Looking ahead: Though too resource-intensive to be made freely available, DeepMind hopes its announcement will inspire applications in new scientific domains.

• The company is soliciting suggestions from researchers for additional areas where AlphaEvolve might be applied.
• The success of this approach could accelerate the trend toward general-purpose AI systems that can tackle diverse scientific and engineering challenges.