Artificial intelligence (AI) technology is advancing by leaps and bounds, and Google is at the forefront with its new AI model, Gemini. Sundar Pichai, CEO of Google and Alphabet, highlights that Gemini represents the most profound technological change of our era, surpassing even the transition to mobile devices or the web. This breakthrough involves not only improvements in processing power, but also in how AI can help people around the world, from simplifying everyday tasks to creating new opportunities for innovation and economic progress.
Innovation and Multimodality with Gemini
Gemini, developed by the Google DeepMind team led by Demis Hassabis, is distinguished as a highly capable, general AI model designed to understand and operate in a multimodal way. This means that it can process and combine information of various types, such as text, code, audio, image and video, in a more comprehensive way than previous models. The creation of Gemini responds to the ambition to build an AI that functions as an expert, intuitive and useful assistant, capable of understanding and interacting with the world in a human-like manner.
Gemini’s Cutting-Edge Performance
The Gemini model has demonstrated exceptional performance on a variety of tasks, outperforming human experts and other AI models in standard academic tests. For example, Gemini Ultra, a version of the model, has achieved a score of 90.0% on massive multitasking language understanding (MMLU), outperforming even human experts in areas such as mathematics, physics, history, law, medicine, and ethics. These achievements underscore Gemini’s advanced reasoning capabilities and ability to process information more deliberately and efficiently.
Next Generation Capabilities and Sophisticated Reasoning
Unlike previous models that required the marriage of separate components for each modality, Gemini has been designed to be natively multimodal from the outset. This allows it to understand and reason about inputs of various natures more smoothly and effectively. Its advanced reasoning capabilities make it particularly adept at discovering insights that may be difficult to discern in large amounts of data, which can lead to significant breakthroughs in fields such as science and finance.
Reliability, Scalability and Efficiency
Gemini has been trained on Google’s optimized AI infrastructure, using internally designed Tensor Processing Units (TPUs). This makes it Google’s most reliable, scalable and efficient model to date, capable of operating significantly faster than previous, less capable models. In addition, the introduction of the Cloud v5p TPU system, the most powerful to date, promises to further accelerate Gemini’s development and make it easier for developers and enterprise customers to train large-scale generative AI models.
Accountability and Security at the Heart of Gemini
Google has implemented a responsible and safe development approach for Gemini, incorporating comprehensive safety assessments, including bias and toxicity testing. In addition, research has been conducted to identify and mitigate potential risks in areas such as cyber offense and persuasion. This commitment to security and accountability is a fundamental pillar in the development and deployment of its AI models.
Gemini at the World’s Reach
Gemini is already being deployed in a variety of Google products and platforms, such as Bard, which uses a tweaked version of Gemini Pro for more advanced reasoning. In addition, The Pixel 8 Pro becomes the first smartphone designed to run Gemini Nano, a more compact and efficient version of the model. New features include the ability to summarize in the Recorder app and integration with Smart Reply in Gboard, starting with WhatsApp and expanding to more messaging apps next year.
Gemini in More Google Services
In the coming months, Gemini will be integrated into a wider range of Google products and services, including Search, Ads, Chrome and Duet AI. Tests are already underway with Gemini in Search, improving the speed of the Search Generative Experience (SGE) by 40% in the U.S., as well as improvements in the quality of results.
Developer Opportunities with Gemini
Starting December 13, developers and enterprise customers will have access to Gemini Pro through the Gemini API in Google AI Studio or Google Cloud Vertex AI. Google AI Studio provides a web-based development tool to launch applications quickly, while Vertex AI enables a personal