Google recently unveiled its highly anticipated large language model, Gemini, marking it as the most extensive and advanced model the company has developed to date.
Gemini, available in three sizes—Ultra, Pro, and Nano—demonstrates Google's commitment to versatility, enabling it to operate seamlessly across a spectrum of devices, from mobile devices to data centers.
The initial deployment of Gemini is through the Bard chatbot, powered by Gemini Pro, the first version of this groundbreaking model. Gemini Pro, currently available only in English, is expected to enhance Bard's reasoning and comprehension capabilities significantly.
In early 2024, Google plans to introduce Bard Advanced, leveraging Gemini Ultra's advanced models and capabilities. While details on Bard Advanced are yet to be disclosed, Google promises access to its cutting-edge features starting with Gemini Ultra.
The upcoming Pixel 8 Pro handset will be the pioneer device to run Gemini Nano. Set to launch in the coming year, the Pixel 8 Pro will offer users new features, including a summarization feature in the Recorder app and a developer preview of Smart Reply in Gboard, initially available on WhatsApp.
Gemini's integration extends beyond mobile devices. Google has plans to incorporate Gemini into Google Ads, Chrome, and Duet AI. In the following months, Gemini is set to power features in various Google products and services, promising enhanced user experiences.
Google's experimentation with Gemini in search indicates its potential to accelerate the Search Generative Experience (SGE), making searches more efficient and responsive.
While the blog post doesn't explicitly mention Gemini powering Makersuite features, a leak in October hinted at the potential for Google's web development platform to benefit from Gemini's generative capabilities.
Developers and enterprise customers can access Gemini Pro through the Gemini API in Vertex AI or Google AI Studio starting December 13. Gemini Ultra will be initially available to select groups, with broader availability for enterprise customers expected in early 2024. Android developers keen on building Gemini-powered apps can sign up for an early preview of Nano via Android AICore.
Gemini's journey has not been without its challenges. Originally planned for a private unveiling, CEO Sundar Pichai decided to cancel private events due to concerns over the model's handling of certain non-English queries. Nonetheless, Google aims for Gemini to compete with OpenAI's ChatGPT, with a strong push for an early release of the initial model.
Addressing the computational demands of training large language models, Google has developed new AI accelerators, Cloud TPU v5p, and introduced the AI Hypercomputer. This supercomputer architecture, designed for optimizing AI training and tuning, combines AI-optimized hardware, software, ML frameworks, and consumption models.
To access Cloud TPU v5p or the AI Hypercomputer, enterprises are encouraged to reach out to their dedicated Google Cloud account managers. Google's advancements in both language models and computational infrastructure underscore its commitment to staying at the forefront of AI innovation.