Google has introduced the latest iteration of its AI model, Gemini 1.5. This updated version is touted to provide a significant advancement in comprehending long-context data.
Google has introduced the next-generation Gemini 1.5 AI model, with the initial version called Gemini 1.5 Pro being released for testing purposes. According to Google, Gemini 1.5 Pro boasts comparable capabilities to the previous Gemini Ultra 1.0 model but with the added benefit of requiring less computational resources.
Just a week following the initial rollout of Gemini Ultra 1.0, Google has swiftly advanced its technology with the release of the next-generation Gemini 1.5 AI model. The first iteration of this new release, called Gemini 1.5 Pro, is now available for testing. Google CEO Sundar Pichai highlights the significant progress made, noting dramatic improvements across multiple aspects. Pichai emphasizes that Gemini 1.5 Pro achieves quality comparable to Gemini Ultra 1.0 while utilizing fewer computational resources.
Pichai reveals that Google has enhanced Gemini 1.5’s intelligence, enabling it to process a significantly larger amount of information simultaneously. Gemini 1.5 can now handle up to 1 million pieces of information at once, setting a new record for big models. This capability allows Gemini 1.5 Pro to process extensive data sets, including 1 hour of video, 11 hours of audio, codebases with over 30,000 lines of code, or over 700,000 words. Google’s research has even successfully tested up to 10 million tokens with Gemini 1.5.
Gemini 1.5 utilizes a novel Mixture-of-Experts (MoE) architecture, a concept endorsed by Demis Hassabis, CEO of Google DeepMind, for its intelligence. This architecture enables Gemini 1.5 to intelligently activate only the relevant parts of its neural network for a given task, enhancing both efficiency and speed. By leveraging this approach, Gemini 1.5 can perform tasks more effectively and rapidly, showcasing its advanced capabilities in handling various types of data and tasks.
“Our latest advancements in model architecture empower Gemini 1.5 to rapidly learn complex tasks while upholding quality and efficiency in training and deployment. These enhancements enable our teams to iterate, train, and deliver more advanced versions of Gemini at an accelerated pace. We continue to explore further optimizations to enhance its performance,” Hassabis shared in a Gemini 1.5 blog post.
During testing, Google provided the Gemini 1.5 Pro model with a 44-minute silent movie. Impressively, the model accurately analyzed various plot points and events, demonstrating its ability to reason about intricate details that might otherwise be overlooked.
As of now, Gemini 1.5 is exclusively accessible to developers and enterprise users. Google has not disclosed specific details regarding the timeline or method for its public release. However, in its announcement blog, Google has affirmed its dedication to responsibly introducing each new generation of Gemini models to billions of people, developers, and enterprises worldwide.
Google plans to release Gemini 1.5 Pro to a wider audience with a standard context window of 128,000 tokens. Subsequently, the company will introduce various pricing tiers, beginning with the standard window size and gradually increasing to accommodate up to 1 million tokens as the model improves. During the testing phase, early users will have the opportunity to experiment with the 1 million token window at no cost. However, Google advises that users should anticipate longer response times from the model due to this new feature.
Read more Tech News
