Google has unveiled Gemini, a family of large language models (LLMs) that represent the company's most ambitious and advanced AI project to date. This groundbreaking multimodal model is the most capable and general model they've built and promises to revolutionize the way people interact with technology and unlock new possibilities across various fields.
The first version, Gemini 1.0, has been optimized at three sizes for different applications:
- Gemini Ultra: The biggest and most powerful variant for complex reasoning tasks. It has surpassed human-level performance on several benchmarks.
- Gemini Pro: A model tuned to balance scale and speed for commercial applications.
- Gemini Nano: A leaner model designed to run efficiently on smartphones and other devices.
Google says that Gemini Ultra has exceeded state-of-the-art results on 30 out of 32 key benchmarks used in AI research. This includes outscoring humans for the first time on tests of broad knowledge and problem-solving abilities.
Unlike its earlier models which only handled text, Gemini was created from the ground up as a multimodal model which understands text, images, audio, video and other formats simultaneously. It can seamlessly reason about cross-connected concepts in a way no other AI system currently can.
Google plans to introduce Gemini capabilities across many of its products in the coming months. This includes upgrades to search, Pixel phones, and the Bard conversational model. Meanwhile developers and enterprise customers can access Gemini Pro through Google AI Studio and Cloud Vertex AI starting December 13th. Early next year, Google will launch Bard Advanced, a new service offering access to its best models,starting with Gemini Ultra.
Safety and oversight are stated priorities for Gemini’s ongoing development, with Google detailing extensive evaluation processes from internal and external experts. However, rapid advances and commercial pressures in AI present growing societal concerns over potential downsides which Google acknowledges it will continue grappling with.
With Google now holding the title of having the world's most capable general-purpose foundation model, the launch of Gemini is a defining moment in their AI journey. However, with OpenAI already having impressive breakthroughs with it's upcoming GPT-5 model, 2024 is shaping up to be another exciting year for AI. Game on.