Google has officially introduced its newest artificial intelligence (AI) model called Gemini, heralding it as a significant leap beyond OpenAI's GPT-4. Sundar Pichai, CEO of Google, along with Demis Hassabis, CEO and co-founder of Google DeepMind, unveiled Gemini in a company blog post on December 6.
Gemini boasts various optimized versions tailored for diverse applications and sizes, such as Ultra, Pro, and Nano, and is designed as a multi-modal system capable of comprehending and integrating various types of information. Google highlights its model's proficiency in mathematics and professional coding, a contrast to OpenAI's GPT-4, which lacks mathematical capabilities.
Google claims that the Ultra version of Gemini showcases top-notch performance across 30 out of 32 academic benchmarks typically used in the development of large language models. Furthermore, it reportedly surpasses human expert performance by achieving up to a 90% score on the Massive Multi-Task Language Understanding (MMLU) test.
Jeff Dean, Google's chief scientist, emphasized that Gemini Ultra marks a significant achievement by being the first model to exceed human expert performance and score over 90% on MMLU across 57 subjects. This system is engineered to seamlessly reason across diverse media types, including text, images, audio, and video, setting it apart from its competitors.
Gemini was purposefully designed to be multimodal from its inception, integrating visual and audio encoders right from the start, unlike models that initially focused on text and later integrated other modes. The AI model also demonstrates advanced programming skills, capable of generating high-quality code using AlphaCode 2 and solving intricate programming challenges while collaborating with developers.
Google also highlighted Gemini Pro's superior performance over GPT-3.5, excelling in six out of eight benchmarks, establishing it as one of the most powerful free chatbots currently available. Google has introduced an updated version of Gemini Pro to its own ChatGPT iteration, called Bard, inviting users to explore the enhanced AI model in English across 170 countries and regions, with plans for expansion to other languages and modes in the near future.
Additionally, Gemini is set to debut on Google's flagship smartphone, the Pixel 8 Pro, powering novel features like Summarize in the Recorder app and launching Smart Replies in Gboard for messaging apps like WhatsApp, with plans for further expansion in the coming year. Google aims to integrate Gemini into more products and services, including Search, Ads, and Chrome, in the following months. Moreover, Google is experimenting with leveraging Gemini to enhance its web-centric search engine, promising a generative and advanced search experience.
