Google has announced the release of Gemini, its most advanced artificial intelligence model, amid growing interest in AI's commercial potential. Gemini consists of three variants: Gemini Ultra, the most powerful; Gemini Pro, versatile across many tasks; and Gemini Nano, tailored for specific tasks and mobile use.
Gemini will initially be available through Google Cloud licensing, allowing customers to incorporate it into their applications. Starting December 13, developers and enterprises can access Gemini Pro via Google's AI interfaces. Gemini Nano will be available for Android developers. Additionally, Gemini will enhance Google products like Bard chatbot and Search Generative Experience, an experimental feature providing conversational search responses.
Gemini Ultra excels in MMLU (massive multitask language understanding), surpassing human expertise in areas like math, physics, and ethics. Its nuanced understanding of complex topics is a key feature.
Sundar Pichai, CEO of Alphabet Inc., highlighted Gemini's multimodal capabilities, enabling it to process and combine various types of information. Gemini Pro will soon power Google's Bard chatbot, offering advanced reasoning and understanding. Next year, "Bard Advanced" will debut, utilizing Gemini Ultra.
This development comes as Google continues to innovate in AI, following the earlier introduction of Bard and competing with OpenAI's GPT models. While Gemini Pro reportedly outperforms GPT-3.5, comparisons with GPT-4 remain undisclosed.
Despite no immediate monetization plans for "Bard Advanced," Google emphasizes creating a robust user experience. Eli Collins of Google DeepMind acknowledged Gemini's novel capabilities but noted ongoing evaluations to fully understand its potential.
Gemini's launch was delayed, recalling Google's previous challenges in AI tool deployment. Eli Collins stressed that testing advanced models takes time and assured that Gemini has undergone extensive testing and safety evaluations.
Remarkably, despite its size, Gemini Ultra is more cost-effective, showcasing efficiency in both capability and resource use. Google plans to share more technical details in a white paper, although specific parameters will remain undisclosed.
Google also introduced its latest tensor processing unit, TPU v5p, for AI model training, promising improved performance at a better value than previous versions. This release coincides with advancements in AI hardware by competitors like Amazon and Microsoft.
In the context of AI's profitability, Google's Search Generative Experience (SGE) represents an ongoing experiment in enhancing search experiences with AI, reflecting the evolving landscape of conversational AI. However, the public launch timeline for SGE remains uncertain.
Pichai expressed enthusiasm for the potential of the Gemini models, underscoring their significance in Google's scientific and engineering endeavors.