Apple Is in Talks to Let Google's Gemini Power iPhone Generative AI Features
bloomberg.com
bloomberg.com
I think Apple by now realizes that Android was a necessary defensive play by Google aimed not at them, but at Microsoft, so maybe this partnership could be revived. Probably going to be more difficult now in the current regulatory environment, however.
I don't think Apple's money or it being in bed with the Majors solves either side of "acknowledging authorship when none is provided/known", "obtaining formal consent for commercial use" or "fairly compensating a creators base" when it is larger than all humans on earth.
What appears to be happening right now is that Google, OpenAI, et.al. are negotiating individual agreements with large publishers if and when threatened with legal action.
I'm not sure this will be sufficient. There will be calls to introduce some sort of AI tax or other forms of collective compensation schemes to funnel money to independent creators, small businesses, museums, etc.
This is all very much up in the air and it will take many years if not decades to settle.
I don't think Tim Cook has the same kind of connections, even if Apple is a much bigger company now.
And it's also why I doubt Apple would go with OAI, wouldn't want to depend on Microsoft's private money sink/flaky legal shield.
I can't say I disagree very much, as the AI hype is quite overblown.
Huh? Could you elaborate on this? Maybe I am missing some historical context, but it feels like Android is very much eating into Apple's market share, not into Microsoft's. How does this reasoning work out?
I knew a bunch of folks who were very sour after how Microsoft handled WP7 -> WP8 and moved to other platforms. And these were the people that were championing Windows Phone.
Android is a direct competitor still.
So yes, very surprising.
Gemini is the best all-around AI tool I've used. It falls a bit on programming tasks, but I don't use it for that.
I suspect a lot of us mediocre coders probably are using these chat AIs primarily for this. ChatGPT's been a pretty solid replacement for what used to be Googling for "How do I do X in Y?" kind of queries. Obviously, that's not really what Siri needs to be.
Where Microsoft's integrated approach presents challenges is Gemini provides for a conversational approach (context). If I ask a question and subsequently need to provide further clarification, I can just provide the clarification. I don't need to re-ask the entire edited question. It's amazing how fast I've become accustomed to this conversational approach.
WRT Apple, Siri has become irksome. She can't reliably handle basic tasks on your phone, much less general-purpose questions. At this point I think it's going to become the case that AI is a commodity, and I don't see what Apple gains by building their own commodity service.
This was a reference to Steve Jobs, Apple, iPhone and AT&T (originally Cingular).
PS. This is the reason why many people call Vercel, the Apple of the dev world. Vercel offers services that have great user experience, but under the hood are mostly wrapped AWS services.
They also partnered with Samsung for the AI features on the S24, which will initially be free, but subscription-based somewhere in 2025.
I can't imagine Apple doesn't release an iphone with special hardware running AI locally in the medium or long-term.
I guess Apple's Marketing is doing its thing once again...
But I don't get you. You complained that droid makes it hard and apple makes it impossible. But it would be better for average user security if they could not do it (aka "did not own the device" in anti-apple propaganda), right?
Apple's mea-culpa is that unlike Android they do not ship an Open Source OS ROM for developers to modify. Google's telemetry can be entirely neutralized by removing Google Play services and using Android without Google software. iPhones don't have that escape hatch, leading to a pretty literal limitation of how you "own" your phone and the software on it. On top of that, iOS has a permissions architecture Apple designed to give the user second-class control over the network. You cannot MITM Apple services - they will go around whatever user-land profile you think you've set. On top of that, there are modem emissions that you're never going to catch with a MDM profile hack and certificate pinning. You have fully drank the kool-aid if you think an empty aircrack-ng screen means "you won" against the multitrillion dollar company and coalition of government regulatory bodies.
I didn't complain about anything, I just stated the facts, with a possible exception regarding the snark about how Apple "owns" the device, although I do think that's a defensible position since they have higher access to it than it's "owner". I do think it's shitty though that they don't provide a way (even with some hoops) for the "owner" of the device to get the highest level of access to it, but that wasn't in the comment.
> But it would be better for average user security if they could not do it (aka "did not own the device" in anti-apple propaganda), right?
Why would that be better? I highly doubt it would make any difference at all to the average user. I doubt it even impacts the majority of power users.
The people who are impacted by these restrictions are the technical users who want to capture and inspect their own device's traffic, usually on their own network. Conveniently, these are also the researchers who might publish blog posts and articles about what kind of data and surveillance the device is sending home about the user, without their knowledge.
But someone who slips in a custom CA cert maybe can. That's the point.
(It's still possible to compare how much a blank device phones home but perhaps we wouldn't know all the details of what it talks about)
This is basically the crux of your argument. I mostly agree with you - neither Apple nor Google do enough to protect user traffic in the big-picture. You can Wireshark a lot of data off both OSes, the throughput is even scarier when you track radio emissions.
That being said, a lot of people have taken notes from Apple's "protect user privacy" shtick. Many logging libraries contain the app-equivalent of screen-recording baked in to the app framework, enabling a pipeline where PII gets ingested as a part of the logging process. Startups that incorporate these processes then brag about their self-imposed security compliance as a result of their own ass-backwards philosophy. And these aren't even the bad guys!
Companies like TikTok and Facebook collect lord knows how much information, and use the same "security" tautology as their scrappy startup peers. They consciously stretch the limits of their API capabilities, and then turn around and make puppy-eyes whenever regulators act concerned. Meanwhile, the actual users of these smartphones aren't empowered to regulate their own device's security. They can't turn off their phone because the modem is still on. They can't firewall Facebook analytics when the app is closed. They can't even stop their notifications from being snooped on without disabling the feature altogether. Where's Apple or Google when that's under scrutiny?
It's a bit tangential, but this is why I think Apple made an enormous mistake attempting to commodity privacy. Privacy is idealistic - there will always be perennial exploits on the iPhone to prove them wrong. Because Apple commits to imperfect, conditional privacy, scummier-and-scummier companies can follow their imperfect lead and make the same claims. And because none of them are as big as Apple, they rarely take flak when their systems fail. Apple's attempts to market security is like watching a leading F1 driver start turning into a tailspin, and taking the rest of the racers with them in a firey crash.
Most marketing hinges on non commodities, take coca-cola for example, it's sugar water with a bit of caffeine, if they marketed that they'd be nothing, all of their marketing is about other stuff, intangibles
Kind of like Netflix's proverbial "chill"
Comparatively I'd say privacy is among better things to market
I can't even tell whether it's sarcasm… All those services are closed-source, exchanging over binary protocols, of which there is no public description/documentation, and no stability guarantee.
I believe on Android MITMing even most third party applications (that make zero-to-no effort to prevent this) requires a rooted phone or an emulator running and older Android (8) without Google Play Services and doing a little bit of RE (for instance using some Frida user scripts to patch the apk to circumvent the certificate pinning). I reckon MITMing the actual traffic Google itself can collect would require a lot more RE and network wizardry than I’m even aware of (feel free to link some reading though). Here’s a recent walkthrough I saw in the wild: https://youtu.be/c4wS9n7yilA?si=xAfwCyWIzdrvOiHc
For Apple devices afaict since rooting was…ahem rooted out, no viable amateur-DIY methods for monitoring your devices traffic exist.
I know everything is open source if you’re good enough at assembly but at some point it’s gone from something a tinkerer can do to something you need significant talent and in-depth knowledge to do.
I’d love to read any write-ups or guides to the contrary though.
Not familiar with the term, what does it mean?
An alternative read is that AI in its current form can produce such unpredictable results they're content for now to use Google as a scapegoat for any oddities.
Perplexity does have good competition in the ‘better web’ product space, but for right now Perplexity just makes searching and exploring more fun. For a few years, I think Apple would do well to have Siri partially wrap Perplexity services as appropriate.
Long term, I hope Apple does everything on device with their own models with firm privacy guarantees.
edit typo
Even Google doesn't have enough GPUs.
The TPU v4 was already competitive with Nvidia, and the new v5p is 2.4 times faster and more power efficient than v4.
Google gets to manufacture that demand at scale.
Also, Google is using all these TPUs for their own training and inference at the moment.
Everything I have read/heard tells me that companies would much, much rather have access to Nvidia chips than Googles TPUs.
When it comes to inference, most of that isn’t really relevant. TPUs or any other specialized hardware will have a higher ROI. But Google aren’t giving them to others either
I know you’re asking for a link, but it’s not really something other than Google can prove. However it is a reasonable extrapolation for anyone familiar with the product space to make.
You can install the Gemini app on other phones, but last time I checked the install counts, it’s only a small percentage of the Android user base.
(I’d also be surprised if a large population would be willing to pay the $20 monthly Google One tier to use Gemini for private use.)
https://www.semianalysis.com/p/google-gemini-eats-the-world-...
"They're failing because they're soooo privacy conscious"
If you believe the memes, Siri is already quite capable of searching Google for you.
How is that a meme? It casually searches Google and summarizes the result when required. Siri runs locally on iOS and needs to use a search engine to lookup information.
https://www.theverge.com/2021/6/7/22522993/apple-siri-on-dev...
My favorite example of this is how Siri used to give an audio confirmation that it is hearing me. This went away several iOS versions ago, and now I have to wait a few seconds then hope that I hear Siri's response; if not, I only then know that Siri did not hear my "Siri, ____" and have to start again. Even worse, maybe one out of ten times Siri does provide an audio confirmation at the start of listening! If there is some hidden iOS setting I can set to fix this, please let me know because I sure can't find it.
But that said, Google has played Gemini politically very well. "The other AI options aren't safe" is a message that regulators love to hear.
>Apple is preparing new capabilities as part of iOS 18 — the next version of the iPhone operating system — based on its own AI models. But those enhancements will be focused on features that operate on its devices, rather than ones delivered via the cloud. So Apple is seeking a partner to do the heavy lifting of generative AI, including functions for creating images and writing essays based on simple prompts.
I can't see how they can delivery 'thoughtful integration' with the Apple services when work is being done on Googles servers, but perhaps the "efficient inference on device" (which could be the biggest product leap for Apple since the iPhone) is taking too long and they need a stop gap to stop the downward pressure on their shareprice.
If they end up renting rather than building that capability, it says to me that they’re betting on being able to move most of that locally to the phone on a relatively short time horizon, which is interesting.
I'm not so sure whether this is the direction of travel. The economics are pretty harsh for local, general purpose AI. And battery will always be a limiting factor.
If you're sending a few hundered requests per day to a very powerful AI then sharing the cost of the inference machinery with others has overwhelming economic advantages. And it will need access to current data anyway, so it can't be completely local.
There will be tasks such as keyboard autocomplete and a lot of other specialised tasks where latency or privacy is more important than quality. So yes I do believe in hybrid. But I think the cloud will always do the heavy lifting when it comes to more general tasks.
I can imagine an alternative scenario where a lot of AI processing happens on Mac and PC and mobile devices benefit from that. But many people don't have powerful desktop devices. So I'm not sure Apple can rely on this.
Even Twitter got in the game and made Grok.
Wow! So unambitious coming from Apple.
Citation needed. I’m fairly certain this is all about scale. OpenAI and Anthropic have obvious capacity issues and Apple obviously doesn’t want to get in bed with Microsoft. It’s a smart bet short term if they get data on what people are actually doing with Gemini.
There's a headline about it right on the landing page of today's NYPost, so by now quite a few folks know about it.
Due to its size and infrastructure Amazon has advantages of choice, price, delivery and service. No digital (or human) assistant is going to be able to find a better deals on most items and then negotiate for it to be delivered tomorrow, free or cheaply, with a fairly good, no quibble, return policy.
It's my belief that the lasting innovation of LLMs will be finally making voice a true first order interface for our computers. "We'll just be talking to them" is something I've heard since the 90s and it seems like we can finally see that reality on the horizon.
This will be the case until we have AI agents that are as intelligent and competent as a top-percentile human personal assistant. At the point where that's possible though, I doubt the future of phone interfaces will be the most interesting change.
Brain to machine signaling interface is the logical step forward if we really want to be faster at input.
The digital assistant will need to be impartial and help in comparison with other similar product. Then again different people have different comparison criteria. Based on context I might be okay with a chinese knockoff while most of the time I might prefer the item that will last the longest.
Any other case: NOPE