Google’s CEO Sundar Pichai announces the dawn of a new era in artificial intelligence – the Gemini era. Gemini is Google’s latest extensive language model, representing a significant leap forward in AI technology.
Gemini is not a singular AI model; it comprises distinct versions tailored for different purposes. Gemini Nano, a lightweight version, operates natively on Android devices offline. Gemini Pro, a more robust iteration, powers Google AI services and forms the foundation of Bard. The most potent model, Gemini Ultra, is designed for data centers and enterprise applications and is set to debut next year.
Google is introducing Gemini through various channels: Bard now runs on Gemini Pro, and Pixel 8 Pro users will benefit from Gemini Nano features. Developers and enterprises can access Gemini Pro through Google Generative AI Studio or Vertex AI in Google Cloud starting December 13th. Initially available in English, plans for incorporating other languages are in the pipeline, with the ultimate goal of integrating Gemini into Google’s search engine, ad products, Chrome browser, and more globally.
In a head-to-head comparison with OpenAI’s GPT-4, Google emphasizes its dominance, having conducted a comprehensive analysis of 32 benchmarks. According to Demis Hassabis, Google’s DeepMind CEO, Gemini outperforms GPT-4 on 30 of 32 benchmarks, showcasing superiority, particularly in understanding and interacting with video and audio, aligning with Gemini’s multimodal strategy.
Gemini’s capabilities extend beyond basic text processing, with advanced models like Gemini Ultra handling images, video, and audio. Google envisions expanding Gemini’s sensory awareness, making it more adept and accurate over time. Despite acknowledging existing challenges like biases, Google is optimistic that continuous improvement will enhance Gemini’s understanding of the world.
Efficiency is a key highlight, with Gemini being faster and more cost-effective than its predecessors. Trained on Google’s Tensor Processing Units, Gemini is also accompanied by the TPU v5p, a new version of the Tensor Processing Unit system designed for large-scale model training and deployment in data centers.
Pichai and Hassabis consider the launch of Gemini a pivotal moment and a significant step toward achieving artificial general intelligence (AGI). Google remains committed to a “bold and responsible” approach, ensuring safety and reliability through thorough testing and red-teaming. While cautious, Google sees Gemini as a groundbreaking development that could rival the transformative impact of the World Wide Web.