Note from Google and Alphabet CEO Sundar Pichai:
Information is the basis of human progress. That’s why for over 26 years we have focused on our mission of organizing the world’s information and making it accessible and useful. And that’s why we continue to push the boundaries of artificial intelligence to organize this information at any input and make it available via any output so that it can be truly useful to you.
This was our vision when last December we launched Gemini 1.0. Gemini 1.0 and 1.5, the first model built with native multimodality in mind, made great strides with multimodality and long context, making it possible to understand and process much more of the information in text, video, images, audio, and code.
Now millions of developers build with Gemini. This helps us rethink all of our products – including all 7 with 2 billion users – and create novel ones. NotebookLM it’s a great example of what multimodality and long context can enable people to do, and why it’s so loved by so many.
Over the past year, we’ve invested in developing more agent-based models, meaning they can better understand the world around you, think many steps ahead and take action on your behalf, under your supervision.
Today we’re excited to launch a novel era of models built for a novel era of agency: introducing Gemini 2.0, our most effective model ever. With novel advances in multimodality – such as native video and audio output – and native apply of tools, this will enable us to build novel AI agents that will bring us closer to our vision of the universal assistant.
Today we are handing over version 2.0 to developers and trusted testers. We’re working quickly to incorporate them into our products, leading the way with Gemini and search. Starting today, our experimental Gemini 2.0 Flash model will be available to all Gemini users. We are also launching a novel feature called Deep researchthat uses advanced reasoning and long-context capabilities to act as a research assistant, investigating intricate topics and producing reports on your behalf. It is now available in Gemini Advanced.
No product has been more transformed by AI than search. Our AI reviews now reach 1 billion people, enabling them to ask completely novel types of questions – quickly becoming one of our most popular search features ever. The next step will be to introduce Gemini 2.0’s advanced inference capabilities to AI reviews to be able to solve more intricate topics and multi-step questions, including advanced mathematical equations, multimodal queries and coding. We started confined testing this week and will roll it out more widely early next year. We’ll continue to make AI overviews available in more countries and languages over the next year.
Progress 2.0 builds on a decade of investment in our differentiated, end-to-end approach to AI innovation. It’s built on custom hardware like Trillium, our sixth-generation TPU. TPU processors powered 100% of Gemini 2.0 training and inference, and today Trillium is generally available customers so they can build with it too.
If Gemini 1.0 was about organizing and understanding information, Gemini 2.0 was about making it much more useful. I can’t wait to see what the next era brings.
-Sundara