最阅读新闻
Search engine giant Google on Wednesday launched its Gemini 2.0, an experimental model that can use websites on its own and is basically a model “for everything.” According to Google CEO Sundar Pichai, the latest model will enable users to build new AI agents that “bring us closer to our vision of a universal assistant.” The company has described its model as its most “capable model yet built for the agentic era.” The initiatives come as part of the firm’s efforts in the tech industry’s heightened competition against the likes of Meta, Microsoft, and Meta. Google will roll out the model across products Pichai also revealed the latest model, with multimodal capabilities will be rolled out across its products including native image and audio output. The search engine giant launched the Gemini 1.0 in December last year, which Google said was the first “natively multimodal” model, suggesting the model could process and respond to text, video, image, audio, and code enquiries. The latest version represents Google’s latest efforts in the tech industry’s increasingly competitive AI race. “If Gemini 1.0 was about organizing and understanding information, Gemini 2.0 is about making it much more useful.” Pichai. Gemini 2.0 comes nearly 10 months after Google first launched 1.5. The model is still in experimental preview, according to Google, with only one version of the model, the smaller, lower-end 2.0 Flash released. Google DeepMind CEO and head of the company’s AI efforts Demis Hassabis said this still was a big day for the tech giant. “It’s as good as the current Pro model is. So you can think of it as one whole tier better, for the same cost efficiency and performance efficiency and speed. We’re really happy with that,” Hassabis explained. Currently, only developers and testers are the first to get 2.0 while other Gemini users have access to the 2.0 Flash experimental model, which builds off 1.5 Flash, which was launched as the fastest and most cost-effective model. Google announced an array of new features According to the company, it expects to add the new model’s reasoning capabilities to the AI Overviews feature which is now accessible to one billion users, according to Pichai. He has indicated that it is becoming one of the group’s most popular search features. With this new model, Pichai further highlighted that AI Overviews will now be able to tackle complex multi-step queries, for instance, mathematical equations and multimodal questions. While limited testing for the newest model started this week, the reasoning feature will only be accessible to more users early next year. According to Google, the reasoning model runs off the search engine giant’s 6 th -generation AI chip, Trillium, which also became available to Google Cloud customers on Wednesday. In its blog post , Google revealed that the new chip delivers 4 times better performance and is 67% more energy efficient than its predecessor. We’re kicking off the start of our Gemini 2.0 era with Gemini 2.0 Flash, which outperforms 1.5 Pro on key benchmarks at 2X speed (see chart below). I’m especially excited to see the fast progress on coding, with more to come. Developers can try an experimental version in AI… pic.twitter.com/iEAV8dzkaW — Sundar Pichai (@sundarpichai) December 11, 2024 The Google CEO also announced other features, such as the new Gemini feature known as Deep Research which can act as a research assistant by using advanced reasoning and long context capabilities. This new feature is available in Gemini Advanced and is capable of compiling research reports on behalf of a user. Hassabis said that Gemini 2.0 was laying the foundation for a great 2025. “We really see 2025 as the true start of the agent-based era.” Hassabis. This comes as the company announced the launch of another product, Project Mariner , an experimental new Chrome extension that can literally use a web browser for users. Additionally, Google also announced that Jules – an agent created specifically to help developers find and fix bad code- and a new Gemini 2.0-based agent that can look at users’ screens and help them play video games better. According to Hassabis, the game agent which he calls an “Easter egg” points to it being truly multimodal. From Zero to Web3 Pro: Your 90-Day Career Launch Plan