Betul Abali | Anadol | Getty Images
Google is optimistic about the prospect of its most advanced AI models coming to smartphones in the next year.
The internet giant expects that its currently available Gemini Large Language Model (LLM), which competes with the more powerful OpenAI GPT-4 AI model supported by Microsoft, will begin to be incorporated into devices starting next year.
Google already offers the Gemini Nano, which is the company’s most efficient model for “on-device” AI, on its Pixel devices and all other compatible Android devices.
Brian Rakowski, vice president of product management for Google’s Pixel unit, said he expects the company’s more advanced models for large languages, currently accessible only through remote data centers via an Internet connection, to begin arrive on smartphones directly next year.
“There are smaller versions of our Gemini model on the cloud,” Rakowski told CNBC. “There has been a lot of progress in compressing these models to work on the device.”
“Some have already been tested and others are being studied for some applications. It would be great to have all the models on the device. It still has wonderful applications.”
“The Gemini Nano is performing at the level our online models achieved less than a year ago,” Rakowski added. “You can do a lot with these little distilled versions of the models on the device.”
“If you just follow this trajectory, some of the things we thought we were going to have to move to the cloud for the next year will be on the device, which is pretty exciting, which is instant without requiring a connection or a subscription.”
Large language models, or LLMs, are artificial intelligence models that can understand and generate language in a human-like manner. Gemini Ultra is Google’s flagship LLM, with 1.56 trillion parameters. For comparison, OpenAI’s GPT-4 consists of 1.76 trillion parameters.
Dreaming of a “supercycle” for smartphones
Smartphone makers are dreaming of a “supercycle” in their industry, driven by artificial intelligence, after a difficult few years that saw device sales slow aggressively. In 2023, smartphone sales fell to 1.16 billion units, the lowest point for unit shipments in a decade.
Analysts say a supercycle is unlikely to occur in the next few years as there isn’t enough in the market in terms of new features and innovations to convince people who own their old smartphones to upgrade.
“Unfortunately, we don’t expect this boom,” Francisco Jeronimo, vice president of data and analytics at research firm IDC, told CNBC.
“The last supercycle we saw was between 2010 and 2015, when in five years the market grew five-fold, from around 300 million smartphones a year to 1.5 billion.”
However, more and more smartphone makers are making big investments in artificial intelligence in the hope that it will spark more excitement around mobile technology.
Companies like Humane, Rabbit and China’s Meizu are betting on a smartphone future that won’t even look like a traditional smartphone. These are devices that would be smaller and more compact and with which we could interact via voice activation, like an Amazon Echo speaker but on the move.
Google has made huge bets on artificial intelligence in an attempt to gain an edge over its rivals like OpenAI, the Microsoft-backed company behind ChatGPT.
Google recently announced a major rebranding of Bard, its ChatGPT alternative, which includes a new app and subscription options. Bard has been renamed Gemini, the same name as the suite of AI models that power the chatbot.
Android users can download a dedicated Android app for Gemini, while iPhone users can use Gemini in the Google app on iOS.
Alphabet CEO Sundar Pichai highlighted the company’s commitment to artificial intelligence during the company’s Jan. 30 earnings call. Pichai said that eventually he’ll want to offer an AI agent that can complete more and more tasks on a user’s behalf, including within Google Search, though he said there’s “a lot of execution to do.”
Likewise, CEOs of tech giants from Microsoft to Amazon have highlighted their commitment to building AI agents as productivity tools.