Researchers Deepmind reported that their new version of artificial intelligence alphageometry2 showed the results at the level of gold medalists of the International Mathematics Olympiad ( IMO). In a series of tests, including tasks of the last 25 years, AI successfully solved 42 out of 50 proposed problems.
AlphaGeometry2 was developed on the basis of the first version of AlphaGeometry, released in early 2023. The new version combines several key elements, including the Google Gemini language model and the ALPHA Proof mathematical system, designed to perform evidence.
The peculiarity of solving problems on IMO is the need to introduce additional mathematical structures that AI should generate independently. AlphaGeometry2 analyzes possible moves, checks their logical validity and selects the best steps for evidence.
During the tests, the system solved problems with a geometric bias, which require not only a high level of reasoning, but also the ability to choose optimal ways to solve. In July 2023, AlphaGeometry2 successfully solved 4 out of 6 problems on IMO, and in wider testing demonstrated the level above the average among the gold medalists of the Olympics.
This success opens up new prospects for the development of artificial intelligence systems that can solve complex mathematical problems, which can lead to the creation of more advanced mathematical applications and tools.