Google is set to release Gemini, its latest AI model, which boasts unparalleled capabilities and versatility. Scheduled for global deployment starting Wednesday, Gemini 1.0 will be seamlessly integrated into Bard and the cutting-edge Pixel 8 Pro smartphones.
Sundar Pichai, Google's CEO, expressed the model's groundbreaking performance across various benchmarks, highlighting its optimization across different sizes: Ultra, Pro, and Nano. Pichai underscored this achievement as a realization of the vision conceived during the establishment of Google DeepMind.
The result of extensive collaboration across Google teams, Gemini has been designed as a multimodal model, proficient in understanding and combining different data types—text, code, audio, images, and videos. One notable demonstration showcased Gemini's ability to perceive, evaluate, and suggest actions in real-time, closely mimicking human cognition.
Gemini's various iterations—Ultra, Pro, and Nano—will cater to different task complexities and device requirements. While Ultra targets highly complex tasks, Pro excels in a wide task spectrum, and Nano operates seamlessly on devices. Gemini Nano, starting with Pixel 8 Pro, will introduce new features like Summarise in the Recorder app and Smart Reply in WhatsApp.
Moreover, Gemini's deployment will extend to various Google products and services like Search, Ads, Chrome, and Duet AI. The Search Generative Experience (SGE) will notably see a 40% reduction in latency in the US, enhancing user experience.
For developers and enterprise users, Gemini Pro will be accessible via the Gemini API in Google AI Studio or Google Cloud Vertex AI from December 13. Additionally, Android developers can leverage Gemini Nano through AICore, debuting in Android 14 on Pixel 8 Pro devices.
Despite undergoing trust and safety checks, Gemini Ultra will be available for select users, developers, and partners for early experimentation and feedback, with plans for a broader rollout in the coming year.
Bard, too, will receive a specially adapted version of Gemini Pro, enhancing its reasoning, planning, and understanding capabilities. Google is set to introduce Bard Advanced, offering access to advanced models like Gemini Ultra in the near future.
In addressing concerns about AI hallucinations, Eli Collins, VP of Product at Google DeepMind, highlighted ongoing efforts to enhance Gemini's factuality and accuracy in responses integrated into products like Bard.
Google touted Gemini Ultra's exceptional performance, surpassing human experts on various academic benchmarks and excelling in multitask language understanding. It claimed Gemini's proficiency in comprehending and generating high-quality code across popular programming languages like Python, Java, C++, and Go.
