Alphabet Inc.’s Google is racing to catch up to OpenAI in the artificial intelligence (AI) arena by opening access to its highly anticipated AI model, Gemini. Google hopes that Gemini will help boost the popularity of its AI products, which have lagged behind competitors in recent years.
Gemini is touted as the largest and most capable AI model ever built by Google. Unlike other competitors, Gemini offers different sizes, including a version that can run directly on smartphones, making it more flexible and accessible. The AI model comes in three versions: Gemini Ultra, Gemini Pro, and Gemini Nano. Eli Collins, vice-president of product at Google DeepMind, highlighted the versatility of Gemini, stating that it can run on everything from mobile devices to large-scale data centers.
The goal of Gemini is to create an AI model that feels more like a helpful collaborator and less like a smart piece of software, according to Collins. To ensure its capabilities, Google ran Gemini through a set of industry benchmarks, where it outperformed OpenAI’s GPT-3.5 model in six out of eight tests. Gemini also surpassed OpenAI’s GPT-4 in seven out of eight benchmarks, showcasing its superior performance in general language understanding, reasoning, math, and coding.
Starting today, Android developers can sign up for the nano version of Gemini, enabling them to build Gemini-powered apps for smartphones and tablets. Google also announced immediate availability of Gemini on its flagship phone, the Pixel 8 Pro, where it will power new generative AI features like summarizing points from a recorded phone conversation. Next week, Google plans to make Gemini Pro accessible for cloud customers through its Vertex AI and AI Studio platforms.
Gemini Ultra, the largest version of Google’s AI model, will initially be available through an early access program for developers and enterprise companies, with broader availability to the public in early 2023.
Gemini will also integrate with Google’s suite of apps and services through Bard, the company’s conversational chatbot. Bard, previously powered by Google’s PaLM 2 model, will now be enhanced by Gemini’s capabilities, enabling advanced reasoning, planning, understanding, and more. Google’s search generative experience, an experimental version of its search engine, will also incorporate Gemini’s powers next year.
While Google is positioning Gemini as a significant leap forward in AI technology, the company acknowledges that the model is not flawless. Google representatives caution that Gemini is still prone to producing false or made-up information, referred to as hallucinations, which remains an unsolved research problem.
Google has prioritized safety evaluations for Gemini, subjecting the AI model to adversarial testing and exposure to real toxicity prompts to ensure it can handle potential misuse or manipulation. The Gemini models benefit from Google’s new underlying supercomputer architecture with updated processing chips, providing faster performance compared to earlier models.
Gemini’s integration with Bard and the wider suite of Google apps signifies the company’s efforts to reinvent its core search business and respond to the rising demand for AI-generated content. By leveraging Gemini’s capabilities, Google aims to regain its leadership position in the AI field and meet the evolving needs of its users.
In conclusion, Google’s release of the Gemini AI model aims to catch up to OpenAI’s success by offering a versatile, powerful, and accessible AI solution. With the ability to run on various devices and outperform competitors in multiple benchmarks, Gemini promises to enhance user experiences and drive innovation across various sectors.
