Google Unveils Gemini AI Model
Company touts performance against industry benchmarks, offers new model in three sizes.
- By Upside Staff
- December 7, 2023
According to Google and Alphabet CEO Sundar Pichai, “the transition we are seeing right now with AI will be the most profound in our lifetimes, far bigger than the shift to mobile or to the web before it.” He’s just announced the company is “taking the next step on our journey with Gemini, our most capable and general model yet, with state-of-the-art performance across many leading benchmarks.”
Gemini 1.0 is the company’s “largest and most capable AI model,” according to a blog post from Pichai and Google DeepMind chief Demis Hassabis. It offers multimodal reasoning capabilities that enable it to make sense of complex written and visual input so it can uncover “knowledge that can be difficult to discern amid vast amounts of data.” Hassabis says Gemini can extract insights from “hundreds of thousands of documents through reading, filtering, and understanding information.”
Characterizing this “new era of models” as one of the “biggest science and engineering efforts we’ve undertaken as a company,” Pichai says he is “genuinely excited for what’s ahead, and for the opportunities Gemini will unlock for people.”
The AI model will be available in two sizes this month -- Pro (for general purposes, most notably powering Google Bard) and Nano (for devices with limited memory). Ultra, for the most complex tasks, won’t be available until next year after the company completes more tests. Ultra is designed to understand text, image, audio, video, and code input.
Having three versions means Gemini is able to run on “everything from data centers to mobile devices.”
The company says Bard now uses a specifically tuned version of Gemini Pro (in English) for “more advanced reasoning, planning, [and] understanding.” The company claims that the Gemini Pro version bested GPT-3.5 in six out of eight industry benchmarks.
Early next year the company will introduce Bard Advanced, which will use the Ultra version of Gemini.
Gemini Nano provides on-device generative AI features for the Pixel 8 Pro. Nano helps “prevent sensitive data from leaving the phone” and can use features without a network connection. More capabilities are promised for early next year.
Coders will be particularly interested in this initial version of Gemini Pro because it can (via an API) “understand, explain, and generate” code in popular programming languages, including Python, Java, and C++. Hassabis says Gemini’s ability “to work across languages and reason about complex information” makes it a leading foundation model for coding.
More information about Gemini 1.0, including performance reports, use cases, and safety issues, is available on Google’s DeepMind website.