I’ve been hearing a lot about AI in mobile phones lately, and I’m kind of confused about how it’s different from the usual smart features that Android phones already have. Like, I know Android has stuff like Google Assistant, face unlock, and all those smart options, but then there’s this “AI” term being thrown around everywhere. What’s the actual difference? Is it just a fancy name for features we’ve been using, or does it really add something new? I’m not super tech savvy, so if you guys could explain it in simple terms or share your thoughts, that’d be great. Maybe even some examples of AI in phones?
The idea is there’s a chip that can run small models on device instead of using an external server over the internet.
Practically the difference is not much. You can get an LLM summary in the weather app etc, so exciting.
Yeah, I was pondering this earlier. In exchange for giving up quite a lot of the CPU die, you can have some NPU functionality instead; rather than having to offload everything to the cloud, you can preprocess some of it first, and then offload it.
Speech-to-text and text-to-speech plus general battery efficiency are reasonable use cases for a phone, certainly more so than a laptop and much more so than a desktop. But as you note, those models are going to be small, and RAM and backing storage on phones tends to be slow and limited.
Was also considering that an NPU gives no benefit at all unless code has been written for it; it’s a bastard to write code for something so concurrent, and giving up CPU die for an NPU instead of more cores and cache means that your phone will generally be slower unless it’s largely used for NPU tasks. Plus if you need to offload to the cloud for most useful tasks anyway, it’s just a marketing gimmick that makes your phone less good for most uses.