Google DeepMind’s AI system, AlphaGeometry2, appears to outperform the average gold medalist in solving geometry problems from the International Mathematical Olympiad (IMO). In a newly published study, DeepMind researchers report that AlphaGeometry2 successfully solved 84% of the geometry challenges from IMO competitions over the past 25 years, surpassing the typical gold medalist score.
AlphaGeometry2 is an improved iteration of its predecessor, AlphaGeometry, and incorporates a language model from Google’s Gemini family alongside a robust "symbolic engine." The Gemini model assists the symbolic engine in applying mathematical rules and logical reasoning to generate feasible proofs for complex geometry theorems. Additionally, a dedicated search algorithm allows the system to perform multiple solution searches concurrently, storing valuable findings in a shared knowledge base.
The DeepMind team selected 50 geometry problems from past IMO competitions. According to the study, AlphaGeometry2 solved 42 out of these 50 problems, thereby outperforming the average IMO gold medalist. Discovering innovative methods to tackle challenging geometry problems may be pivotal in advancing AI capabilities. These problem-solving skills, if further refined, could become a critical component for future general-purpose AI models and may be for artificial general intelligence (AGI).
Use the SaveMyLeads service to improve the speed and quality of your Facebook lead processing. You do not need to regularly check the advertising account and download the CSV file. Get leads quickly and in a convenient format. Using the SML online connector, you can set up automatic transfer of leads from Facebook to various services: CRM systems, instant messengers, task managers, email services, etc. Automate the data transfer process, save time and improve customer service.