DeepMind's AI Program Matches Best Human Performances in Math Olympiad

AlphaGeometry, trained on 100 million algorithmically generated proofs, successfully solves 25 out of 30 past International Mathematical Olympiad problems.

Google DeepMind's AI program, AlphaGeometry, has successfully solved 25 out of 30 geometry problems from past International Mathematical Olympiads (IMOs), a similar success rate to the best human participants.
AlphaGeometry uses a combination of a neural net language model and a symbolic engine to solve problems, and was trained on 100 million algorithmically generated geometric proofs.
The AI program also found a more general solution to a problem from the IMO in 2004 that had previously escaped the attention of experts.
Despite its success, AlphaGeometry's solutions are described as longer and less 'beautiful' than human proofs, and the system is currently limited to the mathematics taught below undergraduate level.
DeepMind has not yet confirmed whether it plans to enter AlphaGeometry in a live IMO contest or whether it is expanding the system to solve other IMO problems not based on geometry.