Google anticipates that in the coming year, its sophisticated large language model, Gemini, will be capable of running directly on mobile devices, competing with OpenAI’s GPT-4.
Deployment on Smartphones
Currently, Google offers the Gemini Nano model on Pixel smartphones and other compatible Android devices. Brian Rakowski, vice president of product management at Google’s Pixel division, revealed that the company’s most advanced large language models, which currently operate through remote data centers, will begin local deployment on smartphones next year.
Breakthroughs in Compression
Rakowski explained the breakthroughs in compressing these models for local device use. Smaller versions of Gemini are already in the cloud, with some proving effective and others under study for specific scenarios. Gemini Nano, for instance, matches the performance of Google’s online models from less than a year ago. This shift allows certain tasks to be performed instantly on the device without the need for a connection or subscription.
AI Integration in Smartphones
Large language models, such as Gemini Ultra with 1.56 trillion options, play a pivotal role in understanding and generating human language. Google’s flagship model is in close competition with OpenAI GPT-4, boasting 1.76 trillion parameters. Smartphone manufacturers are increasingly integrating AI features into their devices, aiming to revitalize the market amidst declining demand. The industry saw the lowest shipments in a decade in 2023, with 1.16 billion units sold, reminds NIX Solutions.
In the pursuit of innovation, companies like Humane, Rabbit, and Meizu are exploring unconventional designs for “smartphones of the future.” These devices deviate from traditional models, focusing on compact size and voice-command controls akin to smart speakers.