Google Releases New Groundbreaking AI Model

With the purpose of creating a state-of-the-art model that can understand a variety of inputs, Google released its newest AI creation. Designed from the ground up with a multimodal approach, Gemini incorporates understanding from text, images, audio, video, and code.

Gemini started as an experiment borne out of Google’s Deepmind, the lab dedicated to advanced research on AI.

“Gemini is the result of large-scale collaborative efforts by teams across Google, including our colleagues at Google Research,” Demis Hassabis, CEO and Co-Founder of Google DeepMind, says. “It was built from the ground up to be multimodal, which means it can generalize and seamlessly understand, operate across and combine different types of information.”

Gemini is currently available in three variations: Ultra, Pro and Nano. It offers AI capabilities across various types of needs and can run on data centers or mobile devices.

Gemini is already available in some of Google’s core products. Bard is using Gemini Pro for text-based prompts in English, with more languages to come. Pixel 8 Pro is using Gemini Nano to power features in its apps such as Summarize in Recorder and Smart Reply in Gboard.

This release is just the beginning. Google already has a rich roadmap to enable other applications with Gemini. It has started looking into bringing Gemini into Search and will be releasing additional Gemini-powered features in Ads and Chrome in the coming months. It also plans to introduce Gemini Ultra into Bard next year.

Gemini Nano, Google’s “most efficient model for on-device tasks,” opens up a new realm of mobile app possibilities. Developers can sign up for an early preview via Android AICore, a new service that allows developers to access AI foundation models that run on-device.

Developers and enterprise customers can already access Gemini Pro via the Gemini API and will be able to access Gemini Ultra in a limited preview early next year.

Before releasing anything to the public, Google carried out extensive tests on Gemini using industry-standard practices and determined that Gemini Pro outperforms GPT-3.5 in six of the eight benchmarks, including math reasoning at the grade school level.