Google DeepMind AI beats world’s top students on geometry problems

As the planet’s largest companies race to create artificial general intelligence (AGI), they are looking for new ways to improve their operations. symbolic thinking and beat your competitors when it comes to AI development. As part of this structure, Google Deepmind decided to change the way it operates and introduced an AI called AlphaGeometry to try to defeat the Californian company’s main competitors: OpenAI and Meta.

For this reason they placed this language model in recreate the International Mathematical Olympiad and the results, as we will see below, are amazing.

Google DeepMind takes a step forward

AlphaGeometry has a very clear goal: to revolutionize the way artificial intelligence (AI) approaches geometry problems, as DeepMind published in a paper in the journal Nature. This almost futuristic tool managed to solve 25 of the 30 problems in the International Mathematical Olympiad, held between 2000 and 2022, in which the best high school students from around the world compete, so this is no small milestone. This achievement puts AlphaGeometry ahead of many high school students and in second place behind those who won gold in this competition.

The key to AlphaGeometry’s success lies in its hybrid approach. Although traditional language models are good at detecting patterns, they often lack the ability to reason accurately. DeepMind has overcome this obstacle combining a language model with a symbolic deduction mechanism, specializing in algebraic and geometric reasoning. This dual approach allows AlphaGeometry to not only predict useful designs, but also explain and test its solutions.

This is also a problem for this AI as it was actually conceived and designed. solely to be able to pass the exam and be good at geometry, so we’re not talking about a more general chatbot like GPT-4, Claude, or Google Bard itself.

The developers were Trieu H. Trinh, a researcher at New York University and former Olympic medalist, Evan Chen and assure that the data he offers is impressive. The decisions are not only correct, but also They follow clear and human reasoning..

AlphaGeometry is part of a larger DeepMind project aimed at improving the capabilities language model of reasoning. By generating 100 million synthetic geometry tests, AlphaGeometry is trained to solve complex problems creatively and efficiently. This hybrid approach could be a significant step to broader progress in artificial intelligencepushing the boundaries of what language models can learn and decide, so it wouldn’t be strange if we see this within general purpose AI in the early future.

To sum it up:

• Google’s new artificial intelligence, Alpha Geometry, correctly solved 25 out of 30 problems in International Mathematical Olympiad tests from 2000 to 2022.
• This AI is specifically designed to pass such tests.
• To do this, they added a symbolic inference mechanism.
• It turns out that this is very important.

To stay updated with the latest technology, subscribe to our official and verified Andro4all WhatsApp channel.