Google launched a pared-down version of what it calls its “largest and most capable” AI model, Gemini, in its AI chatbot Bard on Wednesday, while the most advanced version of Gemini is expected in 2024—adding fuel to the race among tech giants to lead AI development.
A pared-down version of Gemini is available in Google’s AI chatbot Bard. (Photo by Smith ... [+]
Gemini is multimodal, meaning it can recognize and understand different types of information such as text, code, audio, image and video at the same time, Google said.
The first version of Gemini comes in three sizes: Gemini Ultra, for “highly complex tasks,” Gemini Pro, for “a wide range of tasks,” and Gemini Nano, which can be used for specific tasks and on mobile devices.
Google said Gemini Ultra is the first AI model to “outperform human experts” on massive multitask language understanding, or MMLU, which tests world knowledge and problem-solving abilities across 57 subjects including math, history, medicine and ethics.
Gemini can also understand and generate code in programming languages like Python, Java and C++.
Google said it's adding protections for Gemini because of its multimodal capabilities, and is testing for potential risks during development, including for bias, toxicity, violent content and negative stereotypes.
Last month, 18 countries including the U.S. released an international agreement calling on companies developing and deploying AI to make them “secure by design” against bad actors.
Gemini outperformed all other AI models including GPT-4, OpenAI’s most advanced AI model, on 30 of 32 industry benchmarks.
In March, Google released its AI chatbot Bard in response to the hype around Microsoft-backed OpenAI’s ChatGPT chatbot that attracted millions of users when it launched last November. However, the Bard rollout didn’t go smoothly, with a Bard ad showing a wrong answer to a query about telescopes, leading to a price drop in Alphabet shares. Google employees also criticized company CEO Sundar Pichai over Bard’s launch, calling it “rushed,” “botched” and “un-Googley,” CNBC reported. Leading up to Gemini’s launch, Pichai reportedly decided to cancel a series of events around the model’s debut because Google found it couldn’t reliably process some non-English queries, according to The Information.
The Gemini Pro upgrade to Bard is available in English in over 170 countries and territories, and will be expanded to support new languages and locations in the future, Google said. Developers and Google enterprise customers will get access to Gemini Pro through the Gemini API in Google AI Studio or Google Cloud Vertex AI starting on Dec. 13.
U.S. Joins 17 Other Countries In Agreement To Keep AI Safe (Forbes)
White House Unveils ‘Sweeping’ AI Strategy As Biden Pushes For Transparency And Safety (Forbes)