The Facebook owner has struck separate partnerships with smartphone chip makers Qualcomm and Mediatek to make the upgrade a reality. Although they may not be household names, their processors are found inside most Android phones, where they act as the brains of the handsets.
Mediatek revealed on Wednesday that it will release a next-generation chip that harnesses Meta’s generative AI system, known as Llama 2. Qualcomm announced a similar partnership with the social media giant in July.
The upgrade is aimed at overcoming one of the biggest hurdles raised by the new tech trend. Currently, data-hungry generative AI tools like ChatGPT are extremely costly to run as they rely on remote cloud servers.
Operating them natively on mobiles would lower the economic barrier for developers, which in turn could unleash a raft of new apps and services for users. At least that’s the plan.
For its part, Meta is trying to encourage developers to use its AI large language model by giving away the code for free. Some experts have warned that this laissez faire approach could lead to bad actors using the tech for nefarious purposes.
For the general public, the end result could deliver tailor-made personal assistants that offer advice and tips based on your interests, internet activity, fitness data, location and even the way you talk. These so-called AI agents will come in a variety of different personalities for you to interact with. And, because these new services will run on your phone, your data will theoretically be kept more private and secure than it would be on the cloud.
Again, that’s the hype that Qualcomm and Mediatek are peddling. They claim that on-device chatbots will also respond faster as they won’t be reliant on congested internet networks or clogged servers. You’ll even be able to talk to the bots while offline, which could be useful for crafting emails on the tube or prepping for a business meeting on a flight.
Finally, the broader benefit will be reduced energy consumption. It’s no secret that the data centres that keep chatbots ticking require vast amounts of water for cooling. In fact, researchers recently found that training ChatGPT consumed at least 700,000 litres of water, while the average conversation with the bot is equivalent to spilling a 500ml bottle.
Our phones on the other hand can run generative AI models at a fraction of the energy, according to Qualcomm. However, it did not provide a concrete figure for power consumption.
Meta isn’t the only company working on bringing generative AI to smartphones. Google recently declared that it managed to run a lightweight version of PaLM 2, its latest large language model, on mobile devices.