On Wednesday, Google introduced a powerful new Gemini model, calling it their “most capable and general” creation yet. Sundar Pichai, Google’s CEO, mentioned that this marks a significant effort in science and engineering for the company’s AI advancements.
Pichai explained, “With Gemini, our most advanced model, we’re moving forward as an AI-first company, achieving top-notch performance across various benchmarks.” The initial version, Gemini 1.0, comes in different sizes: Ultra, Pro, and Nano, representing the first models of the Gemini era and fulfilling the vision set when Google DeepMind was established earlier this year.
Now, Google’s AI platform, Bard, will use a refined version of Gemini Pro, available in English across 170 countries, including India.
Google classifies Gemini models across three categories: Ultra as the largest and most capable for complex tasks, Pro as the best for a wide range of functions, and Nano as the most efficient for on-device tasks.
Demis Hassabis, CEO of Google DeepMind, highlighted Gemini due to collaborative efforts across Google, including Google Research. He emphasized Gemini’s flexibility, running efficiently from data centers to mobile devices, enhancing how developers and businesses work with AI.
Hassabis shared, “Bard will now use a fine-tuned version of Gemini Pro for advanced reasoning, planning, understanding, and more. This major upgrade will be available in English across more than 170 countries, with plans for expansion to different modes and support for new languages soon.”
Google is integrating Gemini into Pixel, making the Pixel 8 Pro the first phone engineered to run Gemini Nano. This powers new features like Summarize in the Recorder app and Smart Reply in Gboard, starting with WhatsApp and expanding to more messaging apps next year.
In the coming months, Google plans to include Gemini in more products like Search, Ads, Chrome, and Duet AI.
Experiments with Gemini in Search have already begun, resulting in a 40% reduction in latency for the Search Generative Experience (SGE) in English in the U.S., with improved quality.
Starting December 13, developers and enterprise customers can access Gemini Pro through the Gemini API in Google AI Studio or Google Cloud Vertex AI.
Gemini Ultra is undergoing extensive trust and safety checks before broad availability. Select customers, developers, partners, and safety and responsibility experts will have early access to experimentation and feedback.
Google announced Bard Advanced, a new AI experience launching early next year, offering users access to Google’s top models, starting with Gemini Ultra.
Gemini 1.0 was trained on Google’s AI-optimized infrastructure using Tensor Processing Units (TPUs) v4 and v5e. Gemini shows significantly faster performance on TPUs compared to earlier models. Google also unveiled Cloud TPU v5p, its most potent and efficient TPU system. It was designed to accelerate Gemini’s development, enabling faster training of large-scale generative AI models for quicker product delivery.