Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Seriously, I think some people want to specifically limit the information they take in just to maintain a perspective. it says right there that google has no access to data:
stateless computation, enforceable guarantees, no privileged runtime access, non-targetability, and verifiable transparency
...and if you don't think ANY of these tech companies can be held to their word — at least from a legal perspective — why in the world do you own one of their devices, much less put your information on it? This deal isn't about data, it's about money. Google is getting payyyyyyyyyed in this — projected to be between $1-5 B annually, and it's almost guaranteed to improve y-o-y. Very much like the search deal they have together.

I really think this is as simple as Apple needed the Gemini LLM short term, but has a Hardware requirement from a privacy perspective — and on the flip side, Google loves the idea that it's LLM was on every modern smart device. But there is no way in hell Google will let their LLM run on any hardware they don't own. Luckily google's hardware meets the established requirements, so a deal that doesn't compromise Apple's pre-set PCC bar for requirement can be made. So Google can monetize Gemini without risking giving Apple the opportunity to study their model running on their hardware.
 
I really think this is as simple as Apple needed the Gemini LLM short term, but has a Hardware requirement from a privacy perspective — and on the flip side, Google loves the idea that it's LLM was on every modern smart device. But there is no way in hell Google will let their LLM run on any hardware they don't own.

Eh???? The overwhelming vast majority of Android devices are not Google hardware. (i.e. hardware they don't own). And Yet Gemini Nano runs just fine on them.

This who notion that Gemini can only run on specific hardware relatively baseless. Gemini runs on Google TPUs (hardware they own) , on Nvidia hardware (the configuration that Apple is using), and on a wide variety of hardware.

Apple licenese the Gemini models ( pragmatically software) so they can run it where they want to. This isn't the 'raw' Gemini models with a slick Apple facade slapped on top as a shiny wrapper.
 
No link required, if Google is processing it, they have to have access to unencrypted data. Therefor they can steal the data just like they steal other data.

Google (the company) isn't processing it. The INtel chip and the Nvidia chip are processing it. Those are being run at a Google own server farm and Google bought the hardware, but 'Google' isn't directly doing the work.

From the Intel side the mechanism is TDX.

".. Intel® Trust Domain Extensions (Intel® TDX) is Intel's newest confidential computing technology. This hardware-based trusted execution environment (TEE) facilitates the deployment of trust domains (TD), which are hardware-isolated virtual machines (VM) designed to protect sensitive data and applications from unauthorized access. ..."

This is virtual machine that is being run encrypted. So even if a Google Admin logs into the underlying Hypervisor.... they cannot see what is going on inside the virtual machine. The memory images are decypred 'just in time' to do work inside the processor.

Similar mechanism for the Nvidia GPU. Any hypervisor admin peeking in from the 'outside' is not going to see anything but encrypted memory.

Part of the long validation/debug process is that can't really see what is going on inside once it is running.

Data is sent into the VM encrypted. The keys to decrypt the user data sent in is also encrypted.

If there is a defect in the TDX implementation then can perhaps snoop, but if it is working correctly then generally no operator at Google needs to have access to this data for it to be processed.
 
Is there an option to disable external AI features or is this an all or nothing kind of feature?

Back in post 68 https://forums.macrumors.com/thread...-run-on-googles-servers.2483566/post-34623871 there is a link to times of India article.

Only a specific subset of the new Models models ( AFM Cloud Pro ) will go to Google hosted hardware. So if you don't use the "ultra advanced features) then it won't go there. There are optional 'big models' that can be set, but not clear if can say 'nobody's big model'. There is a subduction of Apple Intelligence that decides where to dispatch to.

However, Apple has also set up new APIs so that App developers can also pick external models to dispatch to. So unlikely there will be one master toggle switch that stops everything from going off device.
 
Google recently signed to a $30 billion contract to use XAI servers. I thought it was just for the SpaceX IPO, but this really seems now like Google needs to instantly ramp up data centres for a huge new user base expansion?

It is mainly to 'cook the books' to make the IPO look better. SpaceX has to cover up the huge 12 month bridge loan that is being used to misdirect from the financial crater that is 'Twitter'. Twitter has some huge junk debt that now belongs to SpaceX that is being hand=waved away.

Apple already was one of Google Clouds largest customers. Mush of the current iOS/IpadOS/MacOS user base was already using Google servers as part of Apple cloud services. It isn't so much the user base is growing as it is the computer that user base is using is growing. ( not more users, more compute). More likely though this is a ballon squeeze for other customers to this temporary xAI stuff rather than Apple. ( Apple may be 'pushing' some other current customers off and Google needs some interium place to put them. )

Apple mentioned that the this 'big model' stuff would be rate limited. ( a fixed number of request and will run out of quota for the month. iCloud+ having a longer leash because paying.). But pretty doubtful Apple is about to give away $30B of compute for free ( no charge to end users). A big chunk of that $30B is likely other folks who are paying to actually make some money from that compute to pay that bill. Apple is probably going to give away a couple Billion in 'free' compute; just not 30B.
 
Last edited:
  • Like
Reactions: hans1972 and iZac
Except via their Distributed Cloud air-gapped solution. They're happy to sell you the necessary hardware.

There are some serious misunderstandings around here about how Google operates, what their AI related offerings are, and how the world of really big computing works.
So many misunderstandings, presumably based on a lot of ignorance about how the technology works. It’s all just dog-whistle commentary. Google is bad. I get it. I don’t use those Google services either. People getting all fired up about this would do well to take the time to understand the difference between using GMail and posing a question complex enough that Siri AI needs to route part of it to the subsystem running securely on external servers that’s designed to handle that request. Servers that just happen to be operated by Google. There is just no comparison. Siri AI looks like it will be by far the most private and secure large scale AI service currently available.
 
When data is cryptographically secured so only my/Apple's keys unlock it, Google et al. cannot snoop. That is a guarantee as far as this thread’s mistrust goes.

Yes yes backdoors and quantum and secret radiodetectors and other theoretical things, nothing is ever perfect, an enemy could listen to your monitor’s EM emissions to spy on you if you’re paranoid enough, but in all the practical ways we interact with our world these hyperscaled cloud providers are secure.

And Apple has their security openly vetted by security researchers: https://security.apple.com/documentation/private-cloud-compute/verifiabletransparency.
 
Last edited:
I don’t get why everyone’s still hung up on Google’s privacy issues. Does Apple really have a better alternative, and should it partner with Grok, LLaMA, or ChatGPT? 😀 Ultimately, any company that builds AI infrastructure can monetize it and target users with ads, so the core trade‑off is control versus dependence on third‑party models.

Apple has to suck it at this time. This is the result of missing the AI wave by focusing on the wrong priorities like the worthless goggle; now Apple has to pivot and secure deals with real AI model providers to stay competitive. Yes, Apple avoided huge upfront investment, but that saved capital comes with increased dependence on third‑party models and vendors.

Missing AI is reminiscent of Microsoft under Steve Ballmer, which failed to capitalize on the mobile market - when a CEO is not tech‑oriented and is overly profit‑driven, similar strategic blind spots can occur.

(I am not bashing Tim Cook, so don't ban me.)
Well, given that Google is now running some AI loads on Grok's servers (since Grok has done so poorly they don't need them for their original purpose, and apparently renting is more lucrative than selling), Apple is practically partnering with Musk already. 😬

Also as far as Apple missing the wave, yes they are. As with Maps, they'll live to regret this short-term cost savings path and then be relegated to always playing catch-up. One last fateful decision by Tim Cook.
 
I don’t get why everyone’s still hung up on Google’s privacy issues. Does Apple really have a better alternative, and should it partner with Grok, LLaMA, or ChatGPT? 😀 Ultimately, any company that builds AI infrastructure can monetize it and target users with ads, so the core trade‑off is control versus dependence on third‑party models.

Apple has to suck it at this time. This is the result of missing the AI wave by focusing on the wrong priorities like the worthless goggle; now Apple has to pivot and secure deals with real AI model providers to stay competitive. Yes, Apple avoided huge upfront investment, but that saved capital comes with increased dependence on third‑party models and vendors.

Missing AI is reminiscent of Microsoft under Steve Ballmer, which failed to capitalize on the mobile market - when a CEO is not tech‑oriented and is overly profit‑driven, similar strategic blind spots can occur.

(I am not bashing Tim Cook, so don't ban me.)

I read that as "worthless google", and tried to figure out what you meant. Then I saw "goGGLE", and got it.

The Apple Vision Pro is FAR from worthless. There's a distinction to make between end products (sold in the marketplace), and the technology that goes into them. While the AVP may not be a huge seller (that was actually expected by Apple), the tech going into it will massively change our world over the next decade. That's a fact.
 
My initial reaction is “boo!”. But I’m reasonable enough, I think, that I can wait to judge until we know more technical implementation detail, which they are still working out. After all, it isn’t like Apple itself is perfect and that it has always done things perfectly in these areas.

I assume they just don’t have the necessary chips and capacity, and don’t want to pay the capital costs for acquiring and managing such at the top of this AI bubble. And this path means they can do this with other partners latter.

It is the adult path. But I still have a voice in my head saying “boo!”.
This is also no different to how iCloud used to run (still runs?), they used Google and MS infrastructure for the storage, but put their own privacy layers on top, to ensure that the data was encrypted end to end and at rest.

At a guess, the Apple AI will use custom instances, under Apple control, not the open Gemini models. It won't be your query going through Apple to Google's public Gemini instances, running the query (and possibly extracting information from it), but going from Apple to an Apple instance in the Google data center, running a special version of Gemini for Apple, meeting their privacy requirements, which include not using the query to learn from or storing the information provided in the query.
 
Trying hard to reserve judgment until the white paper comes out explaining the technical reasons my data is safe. Google and data privacy. What an oxymoron.
 
  • Like
Reactions: nt5672
This is also no different to how iCloud used to run (still runs?), they used Google and MS infrastructure for the storage, but put their own privacy layers on top, to ensure that the data was encrypted end to end and at rest.

At a guess, the Apple AI will use custom instances, under Apple control, not the open Gemini models. It won't be your query going through Apple to Google's public Gemini instances, running the query (and possibly extracting information from it), but going from Apple to an Apple instance in the Google data center, running a special version of Gemini for Apple, meeting their privacy requirements, which include not using the query to learn from or storing the information provided in the query.
Keep guessing bud. Great strategy. Why not wait for the technical paper to inform us before guessing?
 
Well no wonder they can't get it approved by the EU, in combination with access to all the data they gather from mobile devices!

What the hell are they thinking?

Does anyone in the US slightly realise how Trump's actual repercussions to political opponents in Europe are undermining trust in American services?
 
  • Like
Reactions: Lioness~
Well no wonder they can't get it approved by the EU, in combination with access to all the data they gather from mobile devices!

What the hell are they thinking?

Does anyone in the US slightly realise how Trump's actual repercussions to political opponents in Europe are undermining trust in American services?

Where do you think iCloud is currently hosted? A mix of Amazon Web Services and Google Cloud.

Read a bit about how any of this works before jumping to conclusions, they are being very open about how all of this works and they even encourage independent auditors to validate it really is how they say it is.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.