- Many mathematicians believe the field is undergoing one of its fastest changes in history as AI rapidly enhances its ability to solve math and prove theorems.
- In 2025, mathematician Daniel Litt bet there was only a 25% chance AI could write a math paper on par with top mathematicians before 2030, but just a year later, he admitted he might lose the bet.
- Only a few years ago, AI could barely solve high school math, but it can now handle problems encountered in actual mathematical research.
- AI systems from companies like OpenAI and Google DeepMind have achieved performance equivalent to a gold medal at the International Mathematical Olympiad.
- In January 2026, mathematicians began using AI to solve long-standing problems proposed by mathematician Paul Erdős.
- A new project called First Proof, initiated by Nikhil Srivastava at UC Berkeley, aims to create a set of real research problems to evaluate AI’s mathematical capabilities.
- The first test set consists of 10 problems from various mathematical fields, ranging from medium to advanced difficulty.
- OpenAI’s AI reportedly solved about 5/10 problems correctly, while Google DeepMind’s system solved 6/10, according to expert evaluations.
- Google developed a mathematical AI tool named Aletheia, combining the Gemini chatbot with verification algorithms to detect errors in solutions and improve results through iterations.
- Beyond solving problems, AI is progressing in converting handwritten proofs into computer-verifiable code, a process called formalization.
- The company Math, Inc. surprised the community when its AI, Gauss, automatically formalized and verified Maryna Viazovska’s Fields Medal-winning proof on the sphere-packing problem.
- The AI-generated proof is approximately 200,000 lines of code, equivalent to about 10% of all formalized mathematics in the world.
- Researchers believe that in the future, AI could automatically check new math papers and detect errors during the peer-review process.
📌 Conclusion: AI is profoundly changing the way mathematics is practiced. AI systems have achieved scores of 5/10 to 6/10 on research problems in the First Proof project and even automatically formalized a Fields Medal-winning proof with 200,000 lines of code. This technology can automate the verification of proofs and support research, but it also raises concerns that humans might lose learning opportunities as machines solve problems too quickly.
