Google has unveiled the next-generation Gemini 1.5 AI model, which the company claims represents a breakthrough in long-term context understanding.
Just a week after launching Gemini Ultra 1.0, Google has made some progress and introduced the next-generation Gemini 1.5, with the Gemini 1.5 Pro being the first model released for testing. “It shows dramatic improvements across a number of dimensions, and 1.5 Pro achieves comparable quality to 1.0 Ultra, while using less compute,” says Google CEO Sundar Pichai in a blog post.
According to Pichai, Google has improved Gemini 1.5 by allowing it to absorb a large amount of information at once. It can now handle up to 1 million pieces of data at once, the highest capacity of any large model. “This means 1.5 Pro can process massive amounts of data in one go, including 1 hour of video, 11 hours of audio, codebases with over 30,000 lines of code, and over 700,000 words.” “In our research, we’ve successfully tested up to 10 million tokens,” claims Google.
Gemini 1.5
The Gemini 1.5 employs a novel Mixture-of-Experts (MoE) architecture, which Demis Hassabis, CEO of Google DeepMind, describes as intelligent since it only uses the parts of the model’s brains required for the task. This improves its performance and speed.
“Our most recent model architecture advances enable Gemini 1.5 to learn complicated jobs faster and with higher quality, while training and serving more efficiently. These efficiencies enable our teams to iterate, train, and produce more advanced versions of Gemini more quickly than ever before, and we’re working on additional optimizations,” Hassabis stated in a Gemini 1.5 blog post.
Interestingly, during their testing, Google offered the Gemini 1.5 Pro model a 44-minute silent movie, and it was supposedly able to “accurately analyse various plot points and events, and even reason about small details in the movie that could easily be missed”.
Gemini 1.5 work ?
Gemini 1.5 is currently exclusively accessible to developers and business users. Google has not yet disclosed the date or method of its public release. Google has reaffirmed that it is “committed to bringing each new generation of Gemini models to billions of people, developers, and enterprises around the world responsibly” in its announcement blog.
Furthermore, Google states that Gemini 1.5 Pro will have a typical context window with 128,000 tokens when it is ready to be made available to a wider audience. In the future, as the model improves, the company will offer varying pricing tiers, starting with the basic window size and increasing to 1 million tokens. Early adopters are able to test the 1 million token window at no cost during the testing phase. Google cautions testers, though, that with this new function, the model may take longer to reply.
Stay tuned to WebArena360 for more updates.