[whalee]

I am deeply disturbed they decided to go off-device for these services to work. This is a terrible precedent, seemingly inconsistent with their previous philosophies and likely a pressured decision. I don't care if they put the word "private" in there or have an endless amount of "expert" audits. What a shame.

[tr3ntg]

They didn't have a choice. Doing everything on-device would result in a horrible user experience. They might as well not participate in this generative AI rush at all if they hoped to keep it on-device. Which would have looked even worse for them.

Their brand is equally about creativity as it is about privacy. They wouldn't chop off one arm to keep the other, but that's what you're suggesting they should have done.

And yes, I know generative AI could be seen specifically as anti-creativity, but I personally don't think it is. It can help one be creative.

[roncesvalles]

I don't think it would've looked bad for their brand to have not participated. Apple successfully avoided other memes like touchscreens laptops and folding phones.

[adpirz]

Siri is bad and is bad for their brand. This is making up for that ground.

[Terretta]

> Doing everything on-device would result in a horrible user experience. They might as well not participate in this generative AI rush at all if they hoped to keep it on-device.

On the contrary, I'm shocked over the last few months how "on device" on a Macbook Pro or Mac Studio competes plausibly with last year's early GPT-4, leveraging Llama 3 70b or Qwen2 72b.

There are surprisingly few things you "need" 128GB of so-called "unified RAM" for, but with M-series processors and the memory bandwidth, this is a use case that shines.

From this thread covering performance of llama.cpp on Apple Silicon M-series …

https://github.com/ggerganov/llama.cpp/discussions/4167

… "Buy as much memory as you can afford would be my bottom line!"

[philjohn]

Yes - but people don't want to pay $4k for a phone with 128GB of unified memory, do they?

And whilst the LLM's running locally are cool, they're still pretty damn slow compared to Chat-GPT, or Meta's LLM.

[theshrike79]

Depending on what you want to do though.

If I want some help coding or ideas about playlists, Gemini and ChatGPT are fine.

But when I'm writing a novel about an assassin with an AI assistant and the public model keeps admonishing me that killing people is bad and he should seek help for his tendencies, it's a LOT faster to just use an uncensored local LLM.

Or when I want to create some people faces for my RPG campaign and the online generator keeps telling me my 200+ word prompt is VERBOTEN. And finally I figure out that "nude lipstick" is somehow bad.

Again, it's faster to feed all this to a local model and just get it done overnight than fight against puritanised AIs.

[chuckadams]

To say nothing of battery life.

[wilg]

You are deeply disturbed by the idea that some services can be better implemented server-side? Who do you think pressured them, the illuminati?

[grishka]

Here's a shocking suggestion: maybe wait some time before these services could be implemented on-device, and implement them on-device, instead of shipping this half-baked something? Apple seems to be the perfect company to make it happen, they produce both the hardware and the software, tightly integrated with each other. No one else is this good at it.

[wilg]

They implemented way more on the device than anyone else is doing, and I don't see how it makes it "half-baked" that it sometimes needs to use an online service. Your suggestion is essentially just not shipping the product until some unspecified future time. That offers no utility to anyone.

[grishka]

It is, however, very much Apple's philosophy to wait it out and let others mature a technology before making use of it.

[TeMPOraL]

At the current rate of advancement, we might get a runaway AGI before the technology "matures".

[grishka]

Or we might not. LLMs are remarkably dumb and incapable of reasoning or abstract thinking. No amount of iterative improvement on that would lead to an AGI. If we are to ever get an actual AGI, it would need to have a vastly different architecture, at minimum allowing the parameters/weights to be updated at runtime by the model itself.

[dleink]

It offers utility to user privacy.

[qeternity]

Here’s a shocking suggestion: if you’re not comfortable using it, don’t use it.

[electriclove]

I like their approach. Do everything possible on device and if it can only be done off-device, provide that choice.

[NewJazz]

You misunderstand.

They will go off device without asking you, they just ask if you want to use ChatGPT.

[rahkiin]

No: they do on device, ask to do off device in their private cloud. Chatgpt is then a separate integration / intent you ask can for

[NewJazz]

I don't see anything to that effect in tfa, and a few people in the comments have claimed otherwise.

[pertymcpert]

Yeah that's how it works.

[unshavedyak]

Are they giving us a choice? I thought the choice was primarily about using ChatGPT? It sounded like everything in apples "Private Cloud" was being considered fully private.

[cedws]

Circa 2013 Snowden says the intelligence agencies are wiretapping everything and monitoring everyone.

In 2024 they don't have to wiretap anything. It's all being sent directly to the cloud. Their job has been done for them.

[buildbuildbuild]

I hear you but caution against such oversimplification. Advanced Data Protection for iCloud is a thing. Our culture of cloud reliance is truly dangerous, but some vendors are at least trying to E2E data where possible.

There are big risks to having a cloud digital footprint, yet clouds can be used “somewhat securely” with encryption depending on your personal threat model.

Also, it’s not fair to compare clouds to wiretapping. Unless you are implying that Apple’s infrastructure is backdoored without their knowledge? One does not simply walk into an Apple datacenter and retrieve user data without questions asked. Legal process is required, and Apple’s legal team has one the stronger track records of standing up against broad requests.

[hu3]

iCloud end-to-end encryption is disabled by default.

So by default, user data is not protected.

https://support.apple.com/en-us/102651

[theshrike79]

Yes, because the UX is better that way.

With ADP if your mom loses her encryption keys, it's all gone. Forever. Permanently.

And of course it's Apple's fault somehow. That's why it's not the default.

[ggamecrazy]

Broadly, in the US, the Federal Wiretap Act of 1968 still applies. You're going to have to convince a judge otherwise.

Yes, perhaps broad dragnet type of might be scoffed down by some judges (outside of Patriot act FISA judges ofc)

I would warn you about the general E2E encryption and encrypted at rest claims. They are in-fact correct, but perhaps misleading? At some point, for most, the data does get decrypted server-side - cue the famous ":-)"

[lancesells]

It's been going to the cloud since at 2013 as well.

[drexlspivey]

That’s a necessary temporary step until these powerful LLMs are able to run locally. I’m sure Apple would be delighted to offload everything on device if possible and not spend their own money on compute.

[giancarlostoro]

They prompt you before you go off-service, which makes the most sense.

[Me1000]

They prompt you before they send your data to OpenAI, but it's clear that they prompt you before they send it to Apple's servers (maybe they do and I missed it?). And their promise that their servers are secure because it's all written in Swift is laughable.

Edit:

This line from the keynote is also suspect: "And just like your iPhone , independent experts can inspect the code that runs on the servers to verify this privacy promise.".

First off, do "independent experts" actually have access to closed source iOS code? If so we already have evidence that this is sufficient (https://www.macrumors.com/2024/05/15/ios-17-5-bug-deleted-ph...).

The actual standard for privacy and security is open source software, anything short of that is just marketing buzz. Every company has an incentive to not leak data, but data leaks still happen.

[ethbr1]

They're promising to go farther than that.

>> Independent experts can inspect the code that runs on Apple silicon servers to verify privacy, and Private Cloud Compute cryptographically ensures that iPhone, iPad, and Mac do not talk to a server unless its software has been publicly logged for inspection. Apple Intelligence with Private Cloud Compute sets a new standard for privacy in AI, unlocking intelligence users can trust.

[pdpi]

That promise made my ears perk up. If it actually stands up to scrutiny, it's pretty damn cool.

[ethbr1]

I look at things like that from a revenue/strategic perspective.

If Apple says it, do they have any disincentives to deliver? Not really. Their ad business is still relatively small, and already architected around privacy.

If someone who derives most of their revenue from targeted ads says it? Yes. Implementing it directly negatively impacts their primary revenue stream.

IMHO, the strategic genius of Apple's "privacy" positioning has been that it doesn't matter to them. It might make things more inconvenient technically, but it doesn't impact their revenue model, in stark contrast to their competitors.

[Hizonner]

Their disincentive to delivering it is that it's not actually possible.

[curious_cat_163]

I agree with you about this being a bad precedent.

However, to me, the off-device bit they showed today (user consent on every request) represents a strategic hedge as a $3T company.

They are likely buying time and trying to prevent people from switching to other ecosystems while their teams catch up with the tech and find a way to do this all in the “Apple Way”.

[evrenesat]

I hope at some point they start selling a beefy Mac mini variant that looks like a HomePod to work as an actual private AI server for the whole family.

[pram]

This is called a Mac Studio

[evrenesat]

It would be great if they let us install the private cloud server on our Macs, but I’m not holding my breath. Then again, in the name of more privacy, maybe they want to sell a dedicated local AI hub as another hardware product. They could even offer it for an affordable upfront cost that can be amortized into a multi-year iCloud subscription.

[29athrowaway]

Siri and other assistants already do this no?

[re]

Yes. Siri debuted with the iPhone 4s (running iOS 5) in 2011. It wasn't until iOS 15 in 2021 that Siri gained the ability to do some things without an internet connection, on devices with the A12 Bionic chip (the 2018 iPhone XR/XS or later).

[hrdwdmrbl]

Yes

[steve1977]

You can’t charge for a service so easily if it runs on-device.