30s Summary

Google’s AI lab, DeepMind, has developed a new model, Gemini 2.0, laying a foundation for more advanced artificial intelligence (AI) agents. These agents can understand complex directions, interact with websites and provide gaming strategies. They also facilitate multi-step research plans, ordinary task assistance, and coding help for developers. Future plans include integration with Chrome browsers and gaming assistance in real-time. DeepMind is also exploring applications within physical robotics.

Full Article

Google’s AI research lab, DeepMind, has developed a new artificial intelligence model called Gemini 2.0. This new model will serve as a foundation for building more advanced AI agents.

A recent AI agent launched on December 11, which uses Gemini 2.0, is pretty impressive. It can understand complicated directions, make plans, think, interact on websites, and even give a hand with video game strategy. That’s according to DeepMind’s CEO Demis Hassabis and CTO Koray Kavukcuoglu, who filled us in via a blog post.

“The use of AI agents is a research area full of opportunities,” Hassabis and Kavukcuoglu mentioned. They went on to say, “We’re diving deep into this new world with a series of designs that can help people get things done and achieve objectives.”

Hassabis and Kavukcuoglu specified that DeepMind is testing out various AI assistant projects powered by Gemini. Each one has different functions.

One project, called Deep Research, assists users with exploring complicated topics by forming multi-step research plans. It goes on the web, finds data, and then produces extensive reports on its discoveries.

Another one, Project Astra, is like a universal AI assistant for ordinary tasks. It gives suggestions and information based on things the user says or asks — it could be anything from “how to do laundry” to “tell me more about this famous monument”.

Project Mariner focuses on making an AI agent that can take over your Chrome browser. It can move the cursor, click buttons, fill out forms, and surf the web. The projects are still in the works but eventually, they want to make them widely used in products.

Another one, Project Jules, is aimed at helping developers. It can be integrated directly into a GitHub workflow and help with tasks such as planning and coding.

Moreover, they’ve made agents using Gemini 2.0 that can assist gamers. It helps players decide what to do next in real-time conversation and can find a bunch of gaming info online.

“We’re partnering with big game developers like Supercell to see how these agents work, testing their ability to understand rules and challenges across a variety of games, from strategy to farming simulators,” they commented.

Hassabis and Kavukcuoglu also mentioned they were testing out AI agents to help out in the physical world through robotics. For now, Google’s AI agents are only being shared with testers and developers.