tech
Phone makers increasingly run AI features like live translation and photo editing directly on the handset instead of sending your data to a distant server.
On-device, or "edge," AI runs directly on a phone's own chip rather than sending requests to a distant data center. This means features like live translation or photo editing keep working without an internet connection, and since your data, such as your voice or photos, never leaves the device, it also offers a meaningful privacy advantage over sending that information to a remote cloud server.
On-device AI does not make models far larger than cloud AI; the opposite is true, since phone hardware has tighter limits on processing power, memory, and battery life, forcing on-device models to be smaller and more efficient. It also doesn't require a constant 5G connection; the whole point is that it can work with no network at all.
As phone chips have grown more powerful for AI tasks, manufacturers have increasingly built dedicated "neural processing units" into their devices, hardware designed to run these compact models efficiently without draining the battery.
tech
How does a 'Mixture of Experts' model achieve this?What is this AI-generated training material called?What is the second approach—training on examples—called?How does this text 'watermarking' generally work?What is this training technique called?What is this compression technique called?What are AI models that handle several input types at once called?Which company operates this fully driverless commercial robotaxi service?On the widely used SAE scale, what is the highest level number, representing full automation?What is the name of the humanoid robot unveiled by carmaker Tesla?Which company pioneered this medical drone-delivery service in Africa?Amazon's warehouse robots grew from its 2012 purchase of which robotics maker?Quration — Quration Play