Google Debuts Its "Largest and Most Capable" AI Model, Gemini
The large language model is the first to outperform human experts on massive multitask language understanding (MMLU).
On Wednesday, Google debuted its “largest and most capable” AI model to-date, titled “Gemini.”
The massive language system comes with a suite of different-sized categories: Gemini Ultra, the most-adept model specializing in highly-complex tasks; Gemini Pro, the best for scaling across a wide range of tasks, and Gemini Nano, the most efficient model for tasks on mobile devices. With sophisticated reasoning and advanced coding, Gemini looks to be multimodal, meaning it can “generalize and seamlessly understand, operate across and combine different types of information including text, code, audio, image and video,” per Demis Hassabis, CEO and co-founder of Google DeepMind.
Google has tested its Gemini models with a vast assortment of tasks across image, audio, video and mathematical understanding. Gemini Ultra’s performance results surpassed state-of-the-art results on 30 of the 32 commonly-used academic benchmarks in the field. Scoring 90.0%, the large language model is the first to outperform human experts on massive multitask language understanding (MMLU), which covers 57 subjects including math, physics, history, law, medicine and ethics.
The first-generation Gemini models are able to extract informations from hundreds to thousands of documents with sophisticated reasoning, which the company says will ultimately “help deliver new breakthroughs at digital speeds in many fields from science to finance.” Gemini can also understand, explain and create advanced code in programming languages such as Python, Java, C++, and Go.
Google states that Gemini adheres to its “most comprehensive safety evaluations” of its AI models to date. “We’ve conducted novel research into potential risk areas like cyber-offense, persuasion and autonomy, and have applied Google Research’s best-in-class adversarial testing techniques to help identify critical safety issues in advance of Gemini’s deployment,” the company’s release reads.
Beginning Wednesday, December 6, Bard will operate with a “fine-tuned” version of Gemini Pro in more than 170 countries and territories. On December 13, Gemini Pro will become available via the Gemini API in Google AI Studio or Google Cloud Vertex AI. Meanwhile, Gemini Nano will launch on the Pixel 8 Pro, and Gemini will debut in Google Search, Ads, Chrome and Duet AI in the coming months. Gemini Ultra will roll out to developers and enterprise customers early next year.
