Google CEO Sundar Pichai announced at Google I/O 2024 that the company is fully in its “Gemini era,” a new phase of AI development. The Gemini model can reason across text, images, video, and code and has been integrated into all of Google’s 2-billion user products. Google also introduced Gemini 1.5 Pro, which can run 1 million tokens in production, more than any other large-scale foundation model. Google’s DeepMind team is working on Gemini 1.5 Flash, a lighter model built for scale. Google also announced Trillium, its 6th generation tensor processing unit, which will be available to Cloud customers in late 2024.
Google’s Gemini Era: A New Generation of AI
Google’s CEO, Sundar Pichai, recently announced the company’s entry into the “Gemini era” at the I/O 2024 conference. The development and application of Gemini, a multimodal AI model capable of reasoning across various forms of data, including text, images, video, and code characterizes this new phase. This marks a significant step in transforming any input into any output, a concept Pichai referred to as an “I/O for a new generation.”
The first Gemini models introduced a year ago, demonstrated state-of-the-art performance on every multimodal benchmark. The subsequent release of Gemini 1.5 Pro marked a significant breakthrough in long context, with the ability to consistently run 1 million tokens in production. This capability surpasses any other large-scale foundation model to date.
Today, over 1.5 million developers use Gemini models across various tools for tasks such as debugging code and building the next generation of AI applications. Google has also integrated Gemini’s capabilities across its products, including Search, Photos, Workspace, and Android.
Gemini’s Impact on Google Products
All of Google’s 2-billion user products now utilize Gemini. The company has also introduced new experiences, including direct interaction with Gemini through a mobile app available on Android and iOS. Gemini Advanced provides access to the most capable models, with over 1 million people signing up to try it in just three months.
One of the most transformative applications of Gemini has been in Google Search. The Search Generative Experience has answered billions of queries in the past year, enabling users to search in entirely new ways and ask more complex questions. This revamped experience, known as AI Overviews, will be launched in the U.S. this week and will soon be available in more countries.
Multimodality and Long Context: Unlocking More Knowledge
The multimodal nature of Gemini allows it to understand each type of input and find connections between them. This capability significantly expands the types of questions that can be asked and the answers that can be provided. The long context feature further enhances this by bringing in even more information, such as hundreds of pages of text, audio hours, or video.
Google has been rolling out Gemini 1.5 Pro with long context in preview over the last few months, making a series of quality improvements across translation, coding, and reasoning. Today, this improved version of Gemini 1.5 Pro is available to all developers globally and can be used across 35 languages.
AI Agents: The Future of AI
AI Agents represent the next step in AI development. These intelligent systems can reason, plan, and remember, enabling them to think multiple steps ahead and work across software and systems to accomplish tasks on behalf of the user. While still in the early stages, Google is working on use cases for AI Agents, such as automating the process of returning a product or helping a user settle into a new city.
Infrastructure for the AI Era: Introducing Trillium
Training state-of-the-art models requires significant computing power. To meet this demand, Google has announced its 6th generation of Tensor Processing Units (TPUs), called Trillium. Trillium is Google’s most performant and efficient TPU to date, delivering a 4.7x improvement in compute performance per chip over the previous generation, TPU v5e. Trillium will be available to Google Cloud customers in late 2024.
Responsible AI: A Key Focus for Google
As Google continues to push the boundaries of AI, it remains committed to doing so responsibly. The company is developing an AI-assisted red teaming technique to improve its models and has expanded SynthID, a watermarking tool that makes AI-generated content easier to identify, to include text and video.
In conclusion, Google’s Gemini era represents a significant step forward in AI development. With its multimodal capabilities, long context, and the potential of AI Agents, Gemini is set to revolutionize the way we interact with technology.
External Link: Click Here For More
