Google has officially launched Gemini 2.0, its latest AI model that boasts significant improvements in speed, cost, and capabilities. This new model is designed to be a game-changer in the AI landscape, enabling multimodal interactions and laying the groundwork for the development of agentic AI agents.

According to Demis Hassabis, CEO of Google DeepMind and head of Google's AI efforts, Gemini 2.0 is a major leap forward, offering performance comparable to the current Pro model at a lower cost and with improved efficiency. The new model can natively generate audio and images, and its multimodal capabilities make it an ideal foundation for the development of AI agents.

One of the most significant implications of Gemini 2.0 is its potential to enable agentic AI, which refers to AI bots that can perform tasks on behalf of users. Google has been demoing Project Astra, a visual system that can identify objects, navigate the world, and even help users find lost items. Gemini 2.0 represents a huge improvement for Astra, and Hassabis believes it will be a key driver of the agentic AI era, which he predicts will begin in earnest in 2025.

In addition to Astra, Google is also launching several other projects that leverage Gemini 2.0's capabilities, including Project Mariner, an experimental Chrome extension that can automate web browsing tasks, and Jules, an agent designed to help developers find and fix bad code. There's even a Gemini 2.0-based agent that can assist gamers by analyzing their screen and providing helpful insights.

Hassabis emphasized that Gemini 2.0's improvements go beyond just performance, with significant gains in efficiency and speed. This is particularly noteworthy given the ongoing discussion about an industry-wide slowdown in model improvements. Hassabis believes that Google's continued innovation in this area will be a key differentiator for the company.

Google's plan for Gemini 2.0 is to integrate it into a wide range of products and services, including Google Search, the Gemini bot and app, and eventually, the AI features in Workspace and other Google offerings. The goal is to create a single, general model that can be used across multiple platforms and applications.

As the agentic AI era begins, Hassabis acknowledges that there are both old and new challenges to be addressed. The old ones revolve around performance, efficiency, and inference cost, while the new ones are still being discovered. One significant concern is the safety risks posed by autonomous agents operating in the world. Google is taking steps to mitigate these risks, including testing in hardened sandboxes, but Hassabis recognizes that more research is needed.

Despite being in an experimental stage, Gemini 2.0 is already available for use in the Gemini web app, with wider availability expected early next year. As the technology continues to evolve, it's clear that Gemini 2.0 has the potential to be a game-changer in the world of AI, enabling new capabilities and applications that will transform the way we interact with technology.