Google is making strides in AI development: the Gemini chatbot has been updated, and two neural networks have been released to tackle complex mathematical problems.
Google’s AI Takes on Complex Math Challenges
DeepMind, the AI division of the corporation, has trained artificial intelligence to solve complex mathematical problems. This was reported in the company’s blog.
Two AI models have been released:
- AlphaProof, which specializes in mathematical reasoning;
- AlphaGeometry 2, an updated version of a previously introduced model focused on geometry.
The programs successfully solved four out of six problems presented at the International Mathematical Olympiad.
“This is a significant advancement in machine learning and AI. Previously, there was no system capable of solving problems with such success,” commented Pushmeet Kohli, DeepMind’s Vice President of AI for Science Research.
Google addressed the issue of “hallucinations” by translating mathematical problems into technical formulations or “formal language.” The AlphaProof code is continuously updated, enabling the AI to solve more complex equations.
AlphaGeometry 2 managed to solve 83% of all geometry problems included in the International Mathematical Olympiad over the past 25 years, the company noted. However, AI is far from replacing human mathematicians, Google researchers said.
DeepMind’s Vice President of Reinforcement Learning, David Silver, noted that AI is more like a calculator: it is a powerful computational tool but lacks imagination.
Gemini Update
On July 25, Google released an update for the Gemini chatbot.
The lightweight multimodal version of the neural network, 1.5 Flash, became available to everyone in the browser and on mobile devices in the free version of Gemini. It provides an improved process for reasoning and understanding images.
The context window of Gemini has been increased fourfold, to 32,000 tokens. This allows for longer conversations without losing context—the AI does not forget previously discussed topics.
Soon, there will be an option to upload files via Google Drive or from the user’s device. Gemini will be able to analyze documents, assist in finding data, and visualize information.
To combat “hallucinations,” Google’s AI now shows sources after a search query.
Earlier, DeepMind’s robotics division integrated Gemini 1.5 Pro into a robot to teach it to navigate an office and respond to commands.
In May, Google integrated generative artificial intelligence into its search engine, and in June, into Gmail and other company services.
Back in May, the corporation’s AI division was developing technology for creating soundtracks for videos.