Apple Picks Gemini

"Apple needed this all in-house for years"

What does "this" refer to in this sentence? LLMs? I think Apple lucked out not being part of that insane money pit.

So it is bad to make money from customer’s data but it is OK to make money using a technology from a company that used customer’s data?

Also wasn’t it suggested in the past that the Titan project was not a complete waste of billions of dollars because Apple would use the deep-learning/AI stuff they worked on during this failed project?

"Sort of weird that they would announce such a big deal this way rather than official releases/interviews/etc"

It sure is a weird coincidence that this would come out now, in such an informal, let's-just-get-this-out-there-now fashion, just as all these calls for Apple and Google to deal with X/Grok according to their policies are piling up. Huh!

Does this not all seem reminiscent of Google Maps for iPhone until Apple Maps? And then Google Maps again and essentially still because even to this day Apple Maps is a poorly launched, very slow to improve also ran? That basically only exists because "Apple needs this in house" even though it is incredibly expensive to properly create and maintain and in the end its only real monetary benefit is ads?

I realize that was a lot of questions with no real point. My point is this: Apple has gotten pretty far away from their core competencies chasing a lot of nonsense the last many years. An unnecessary streaming service, the car thing, arguably Vision Pro, Liquid Glass, etc etc.

I see this as a neutral to good thing because if they are going to do this and clearly not be good at it at least for once just outsource it to someone who is and do a half decent job of integrating with a third party service like they know how to do when it benefits them.

Apple needs to focus on making great hardware, and the best operating systems in the world to run on that hardware, all in service to the user of that hardware as much as possible.

If this is what it takes to finally fix Siri and move on to fixing the other things they've neglected the last decade, then great.
If that means AI is what people want and Google does it then just let them do it and focus on what Apple can do.

Apple can't execute.

Very slowly, then very quickly. That's how a failure occurs. Apple is now in the cascading failure stage. The low hanging fruit gains of Apple Silicon is exhausted, the wheels are going to fall off that soon enough as talent leaves to do things Apple won't, or can't do because of the institutional drag the company's culture (corporate, and national) creates. Failures are only going to become more frequent from here.

Sovereignty, and user-ownership of data is the new universal trend for technology, that's the next 10-15 years, and Apple is the most proprietary of American companies.

Apple could fix a lot of these problems if they just brought back the @Mac.com domain.

(I kid!)

Firstly, LOL.

Now that that's out of the way, I agree @Someone, time for Apple to (re-)enter the SOHO router and server markets! Imagine an app marketplace for Linux containers to run on your home server? They have the technology already …

So they build slowly but steadily hardware ML processing capabilities for years. From the private AMX mini core to full ML cores in A mobile processors and above. All of that paired with unobtrusive software solutions helping to predict words, recognize images and more.

Could have branded that Apple Learning Models years ahead of the competition. And iterate releasing updates year after year. Waited instead letting Microsoft and others dominate the AI narrative. And now handing Apple customers data over to Google? Because this is what really is about.

Can we have Scott Forstall back now.

Sounds like this is basically licensing a trained neural net -- probably a small one for on-device, and a larger one for in the cloud -- which is… fine, I guess. As long as customer data isn't being used to train future versions.

Outsourcing to another company also means Apple has someone to blame when it does incorrect things.

I imagine similar licensing would need to happen for other countries, as well? (China, specifically)

LLMs are probably, at some point, going to be somewhat commodity, so waiting to release your own might not be so bad, after all. Saves energy, in the meantime, also. Recall how DeepSeek leapfrogged. Apple has a disadvantage, though, in that they were only using licensed data to train their public models, unlike the competition. I wonder if that restriction will hold (I hope so).

But what would be more interesting is MCP-style stuff on iOS devices (where AI can control apps)... how will that work? That might be where Apple can perhaps do some privacy/risk-managed innovation, because that whole realm is pretty sketchy and dangerous right now, but that's what users dream about.

Right now, LLMs are basically conversational-style, wildly-inefficient search engines. Is that really what we want?