201–210 of 674 posts
The writing was on the wall the moment Apple stopped trying to buy their way into the server-side training game like what three years ago? Apple has the best edge inference silicon in the world (neural engine), but they have effectively zero presence in a training datacenter. They simply do not have the TPU pods or the H100 clusters to train a frontier model like Gemini 2.5 or 3.0 from scratch without burning 10 year…
There is no intelligence
This seems like a pretty significant anti-trust issue. One of the two mobile OS makers is using a product from the other for its AI assistance. And that means that basically all mobile devices will be using the same AI technology. I don't expect the current US government to do anything about it though.
What antitrust rule do you think would be breached? I admit I don't see the issue here. Companies are free to select their service providers, and free to dominate a market (as long as they don't abuse such dominant position).
Earlier quoted context omitted.
Is the training cost really that high, though? The Allen Institute (a non-profit) just released the Molmo 2 and Olmo 3 models. They trained these from scratch using public datasets, and they are performance-competitive with Gemini in several benchmarks [0] [1]. AMD was also able to successfully train an older version of OLMo on their hardware using the published code, data, and recipe [2]. If a non-profit and a chip…
No, of course the training costs aren't that high. Apple's ten years of future free cash flow is greater than a trillion dollars (they are above $100b per year). Obviously, the training costs are a trivial amount compared to that figure.
I was under the impression that all these GPUs and such were needed to run the AI, not only ingest the data.
Earlier quoted context omitted.
Yea, I think it’s smart, too. There are multiple companies who have spent a fortune on training and are going to be increasingly interested in (desperate to?) see a return from it. Apple can choose the best of the bunch, pay less than they would have to to build it themselves, and swap to a new one if someone produces another breakthrough.
100%. It feels like Apple is perfectly happy letting the AI labs fight a race to the bottom on pricing while they keep the high-margin user relationship. I'm curious if this officially turns the foundation model providers into the new "dumb pipes" of the tech stack?
Earlier quoted context omitted.
Could you elaborate a bit on why you've judged it as privacy theatre? I'm skeptical but uninformed, and I believe Mullvad are taking a similar approach.
Mullvad is nothing like Apple. For apple devices: - need real email and real phone number to even boot the device - cannot disable telemetry - app store apps only, even though many key privacy preserving apps are not available - /etc/hosts are not your own, DNS control in general is extremely weak - VPN apps on idevices have artificial holes - can't change push notification provider - can only use webkit for browsers…
Works great for me with NextDNS.
Orion browser - while also based on WebKit - is also awesome and has great built in Adblock and supposedly privacy respecting ideals.
OpenAI had it , they had the foot in the door with their integration last year with Siri. But they dropped that ball and many other balls.
Earlier quoted context omitted.
I don’t think there’s a clear boundary of “Siri-level” tasks. In particular, properly determining whether a task is “Siri-level” or not is likely to require off-device AI.
I'd hope it could be the other way around. Some stuff should be relatively straightforward -- summarizing notifications, emails, setting timers, things like that should be obviously on-device. But aside from that, I would hope that the on-device AI can make the determination on whether it is necessary to go to a datacenter AI for a better answer. But you may be right, maybe on-device won't be smart enough to decide i…
Earlier quoted context omitted.
No, of course the training costs aren't that high. Apple's ten years of future free cash flow is greater than a trillion dollars (they are above $100b per year). Obviously, the training costs are a trivial amount compared to that figure.
I have no idea what AI involves, but "training" sounds like a one-and-done - but how is the result "stored"? If you have trained up a Gemini, can you "clone" it and if so, what is needed? I was under the impression that all these GPUs and such were needed to run the AI, not only ingest the data.
Earlier quoted context omitted.
> So because Apple chose not to spend money to develop it's own AI, it must be punished for then choosing to use another companies model? And the reason that this is an issue is because both companies are large? No they were already being sued for antitrust violations, it just mirrors what they are accused of doing to exploit their platform. https://storage.courtlistener.com/recap/gov.uscourts.njd.544...
So if it mirrors something they were already accused of (like you're saying), my questioning should be pretty easy to map onto that issue as well? It's the line of thinking that I'm trying to dig into more, not the specifics of this case. Now it feels like you're saying "this is anti-trust because someone accused them of anti-trust before". If that case was prosecuted and Apple was found guilty, I suppose you can poi…
No it's antitrust because they have a failed product, but purely by virtue of shutting out competitors from their platform they have been able to turn three years of flailing around into a win-by-outsourcing. What would Siri's position be like today if they hadn't blocked default voice assistants? Would they be able to recover from their plight to dominate the market just by adopting Google's technology? How would that measure against OpenAI, Anthropic or just using Google directly? This is why it's an antitrust issue.
I don't understand why Apple cannot implement their own LLM at the user phone level for easy pickings? like settings control? or app-specific shortcuts? or local data searching? I understand other things like image recognition, wikipedia information, etc require external data sets, and transferring over local data to that end can be a privacy breach. But the local stuff should be easy, at least in one or two language…
In the original announcement of the Siri revamped a couple of years ago, they specifically talked about having the on-device model handle everything it can, and only using the cloud models for the harder or more open ended questions.