Zap / from 2
Google’s AlphaGeometry2 system already has a resolution rate of 84% of geometry problems of the International Mathematics Olympics, beating the average gold medal, which is 81.8%.
Google Deepmind revealed the AlphaGeometry2 (AG2), an advanced system of artificial intelligence (IA) that surpassed some of the best human competitors in the world in international geometry competitions.
A he Successed 84% of the problems successfully From the prestigious International Mathematics Olympics (IMO), surpassing the average performance of gold medals, which usually solve 81.8% of problems. This discovery, detailed in an available at the Arxiv Prepress Database, points out a significant advance in AI capacity to deal with high-level mathematical reasoning.
AG2 is a improved version of alphageometrywhich was first launched in January 2024. The last iteration demonstrated a 30% increase in performance, with improved capacities to solve complex geometry problems. Unlike algebra or calculation, geometry requires a mixture of visual reasoning and logical deduction, which makes it one of the areas of mathematics more difficult to master by AI. AG2 gets it through a hybrid approach that combines neural language models with symbolic engines.
The AG2 works by translating human -readable geometry problems into mathematical constructions that its symbolic engine can test. If an initial attempt fails, the Itera system through new possibilitiescontinuously refining your approach until you reach a correct solution. The latest version of the AG2 benefits from a larger and more diverse training set, a faster symbolic engine and an improved algorithm to generate geometric evidence, explains the.
Despite its impressive results, the AG2 It also has remarkable limitations: It has difficulties with 3D geometry and nonlinear equations, and cannot solve problems involving variable or infinite points. Moreover, it does not have the ability to explain its reasoning in a way that humans can easily understand.
Experts also warn that AG2 is not considered as a step towards General Artificial Intelligence (AGI), which would require an AI to think and reasoning as a human being in several disciplines.
Looking to the future, Google Deepmind intends to improve the AG2 by increasing its problem solving speedexpanding its ability to deal with more complex geometric concepts and improving its overall reliability.
Although AG2 focuses mainly on geometry, its underlying AI techniques have wider applications in domains such as engineering conception, automated systems verification, robotics, pharmaceutical research and genomic. The long term objective is Create an AI totally automated Able to solve geometry problems with total precision, surpassing the boundaries of what AI can achieve in mathematics and beyond.
