>Knowledge distillation works like this: you take a large model, have it perform tasks with detailed reasoning, then feed those reasoning traces to a smaller model until the student learns to mimic the teacher. The smaller model ends up far more capable than if you’d trained it from scratch on the same data. Apple can now do this with the full Gemini, not just their own in-house models, and the distilled output runs locally. No internet required.
No freaking way. AI companies see this as tantamount to pirating their models. There is no way that Google is not explicitly banning this in their agreement to allow Apple use their models.
> Obviously, the average consumer has no idea what that sentence means, and Apple hasn’t figured out how to make them care.
the article also does nothing to show how or why consumers should care.
> Gartner projects GenAI smartphone spending will reach $393 billion in 2026, up 32% from $298 billion in 2025.
is this because consumers are interested in "GenAI" smartphones, or because every smartphone manufacturer has decided to tack "AI" onto every piece of their branding and there's just no way around it?
Good point there, I totally forgot to think why anyone should care about local models. It's something I've been wishing for while trying to build local first apps and I didn't think why my customers would care in the first place.
Let's obliterate your phone's battery life running billion parameter models locally at snails pace, saving a couple kilobytes of data.
Sounds useful. I'm sure I'll at least think about maybe making use of that once. Even if I won't, thinking that I could will make me feel like such a savvy consumer anyways.
I am looking forward to more capable phones able to run on device AI models. I prefer my phone to be still useful even without internet. Since Apple is using Gemini models to do this, do you think there is a chance Pixel phones will be part of the able to run models on devices, in their next updates?
We're literally less than 10 years from satellite based cell service. Not to mention that the market interest in this is artificially propped up by the industry. They're just trying (and failing) to find more market lines for AI.
I have been craving for a small, extremely accurate voice model for a year. Having that would be an absolute game changer because I think that's the inflection point where people start interacting with their devices primarily through audio.
>Knowledge distillation works like this: you take a large model, have it perform tasks with detailed reasoning, then feed those reasoning traces to a smaller model until the student learns to mimic the teacher. The smaller model ends up far more capable than if you’d trained it from scratch on the same data. Apple can now do this with the full Gemini, not just their own in-house models, and the distilled output runs locally. No internet required.
No freaking way. AI companies see this as tantamount to pirating their models. There is no way that Google is not explicitly banning this in their agreement to allow Apple use their models.
Here is a technical comparison to show how Knowledge Distillation works for on-device AI. It has visual maps for the Teacher vs Student models and explains things like 'Dark Knowledge' - https://vectree.io/compare/apple-intelligence-vs-knowledge-d...
> Obviously, the average consumer has no idea what that sentence means, and Apple hasn’t figured out how to make them care.
the article also does nothing to show how or why consumers should care.
> Gartner projects GenAI smartphone spending will reach $393 billion in 2026, up 32% from $298 billion in 2025.
is this because consumers are interested in "GenAI" smartphones, or because every smartphone manufacturer has decided to tack "AI" onto every piece of their branding and there's just no way around it?
Good point there, I totally forgot to think why anyone should care about local models. It's something I've been wishing for while trying to build local first apps and I didn't think why my customers would care in the first place.
Let's obliterate your phone's battery life running billion parameter models locally at snails pace, saving a couple kilobytes of data.
Sounds useful. I'm sure I'll at least think about maybe making use of that once. Even if I won't, thinking that I could will make me feel like such a savvy consumer anyways.
I am looking forward to more capable phones able to run on device AI models. I prefer my phone to be still useful even without internet. Since Apple is using Gemini models to do this, do you think there is a chance Pixel phones will be part of the able to run models on devices, in their next updates?
We're literally less than 10 years from satellite based cell service. Not to mention that the market interest in this is artificially propped up by the industry. They're just trying (and failing) to find more market lines for AI.
But then I'm even more reliant on some cloud to stay up.
I have been craving for a small, extremely accurate voice model for a year. Having that would be an absolute game changer because I think that's the inflection point where people start interacting with their devices primarily through audio.
Like the one that got a dad investigated by the police when he sent a pic to his doc? That one? Lawyers say he is still permabanned.