Google has recently unveiled its latest language model, Gemini, which is touted as their most advanced and versatile AI model to date. Gemini has the remarkable ability to comprehend and manipulate various forms of information, such as text, code, audio, images, and video. However, there is more to Gemini than initially meets the eye. It is not just a single model, but rather a collection of AI models available in three different sizes: Gemini Ultra, Gemini Pro, and Gemini Nano.
According to Google, Gemini has made its debut as version 1.0 and serves as their flagship AI model. Each size of Gemini is specifically designed to cater to different requirements and can seamlessly operate across a wide range of platforms, from data centers to mobile devices. Sundar Pichai, the CEO of Google, explains in an official blog post that Gemini was developed to be inherently multimodal, with initial pre-training on various modalities. It was then further refined through fine-tuning using additional multimodal data to enhance its effectiveness.
The Three sizes of Gemini:
Gemini Nano: Google’s Gemini Nano is an AI model designed specifically for mobile devices, with a particular focus on the Pixel 8. This lightweight model allows for seamless on-device tasks, even when offline. Whether it’s providing relevant suggestions in chat applications or summarizing text content, Gemini Nano offers robust AI capabilities without the need for external servers. This ensures a user experience that is both smooth and responsive.
Gemini Pro: Gemini Pro, the enhanced version of Gemini AI, is set to emerge as the primary driving force behind various Google AI services. It will serve as the fundamental backbone for Bard, starting from today. Operating on Google’s data centers, Gemini Pro has been meticulously crafted to empower the latest iteration of our AI chatbot, Bard. This cutting-edge technology ensures swift response times and an exceptional capability to comprehend intricate queries.
Gemini Ultra: Gemini Ultra, developed by Google, is renowned as the most dominant LLM in existence. Its exceptional capabilities make it exceptionally adept at handling intricate tasks, making it the ideal choice for data centers and enterprise applications. While it is not yet widely accessible, Google has described Gemini Ultra as surpassing the “current state-of-the-art results on 30 of the 32 widely-used academic benchmarks used in large language model (LLM) research and development.” This signifies that it outperforms other LLMs on the majority of standardized assessments. With its focus on intricate assignments, the release of Gemini Ultra is anticipated once it completes its ongoing testing phase.
Gemini, starting with Gemini Nano, is now accessible on Pixel 8 Pro, bringing new features like Summarise in the Recorder app and Smart Reply on Gboard, which were initially introduced in WhatsApp. The expansion of Gemini will include various Google products and services such as Search, Ads, Chrome, and Duet AI.
Furthermore, Gemini is already being integrated into Google Search to enhance the Search Generative Experience (SGE), resulting in a 40% reduction in latency in English in the US, along with improvements in quality.
From December 13th, developers and enterprise users can access Gemini Pro through the Gemini API in Google AI Studio or Google Cloud Vertex AI. Android developers can also utilize Gemini Nano via AICore, a new feature available on Pixel 8 Pro devices running Android 14. While Gemini Ultra undergoes trust and safety evaluations, a select group of customers, developers, partners, and safety experts will have early access for testing and feedback. It will be made available to a wider audience of developers and enterprise users in the early part of next year.
Additionally, Bard will receive a specially adapted version of Gemini Pro in English, enhancing its reasoning, planning, and understanding capabilities. In the upcoming year, Google will introduce Bard Advanced, providing users with early access to our most advanced models and capabilities, starting with Gemini Ultra.