Apple Buys DarwinAI Ahead of Major Generative AI Updates Coming in iOS 18
macrumors.com
macrumors.com
I’d love to be surprised at WWDC this year, but it intuitively seems exceedingly unlikely.
Apparently there's a senior exec in Apple that is even more of a puritan than the "AI safety" nuts and forced Siri to be dumbed down to prevent even the slightest risk of any impropriety.
With people like that in charge, innovation can't occur.
At this point I feel like I rarely have any Siri issues and use it all the time so I am confused when people mention how bad it is.
So maybe the varied experience is largely just whether you're American and if so have the right regional accent? (For what it's worth, my go-to impression is I think what you'd call 'valley girl', or whatever the male equivalent is. Since I feel liiike that's basically what all city-dwelling Americans sound like to me anywaaay. Do they sound that way to you allso orrr no? ;) )
For a personal example, I had a six-month period where Siri stopped understanding the word 'half' (as in, 'Hey Siri, volume to half'), and either ignored or misinterpreted any command that included it.
I sooo wish i could delete that *** album because every time it gets me wrong it starts playing it. And I don't even like U2. If I didn't have that it would have nothing to play because my music library would be empty.
There seems to be no way to do it...
I really really would love to see some LLM love in these services rather than all this scripted 2010s stuff.
Whether you want more privacy or more functionality, the solution is the same. Use the device that you can control.
For most people, the choice is between giving that data to Apple or giving it to Google. Personally I am more worried about an advertising company leaking/misusing my data than a hardware company that has made privacy a differentiator in their branding. Not saying I trust that but you have to weigh up your risk factors
Google is also a hardware company
You think Apple cares about your privacy? It's part of their PR, sure.
Suggesting Google has a bad reputation with sharing data to private companies isn't going to be enough. I'd like to see some proof to back that up...
On Android, even Google Android, you're free to use fully on-device maps that don't send your location to anyone, or you could use competing apps that don't send your location to Google. Similarly, you can install apps from an APK from anywhere without telling Google or anybody else that you've installed them, even on Google Android.
User control means privacy, not just better features.
As I said, this isn't an option for most people.
GPS doesn't require Google. It just reads radio signals. Only Apple requires sending the result of analyzing those GPS signals back to Apple, even if you just want to use it to find your location on a map stored locally. Apple doesn't give you any options for privacy.
I have a friend just like you, who looks at something perfectly innocent that Apple does, screams “spyiiiiiing!” and runs into the arms of Google — a company famous for turning people into data products. That’s literally all they do. They don’t even know how to do anything else to make money.
Apple is a hardware company first and foremost that treats their users as customers, which is distasteful to you for some reason.
PS: Apple phones can acquire a GPS lock without cellular connectivity. I’ve relied on this feature many times in the Australian outback.
Oh, and you can turn Find My feature off if you’re that paranoid.
The difference between every other manufacturer and Apple is that every other manufacturer lets you turn it off if you care about privacy.
> Without help from an external service, it would take up to half an hour to acquire a fix using GPS radio signals alone
Up to. If you're under a rock and haven't previously gotten a location.
> Apple is a hardware company first and foremost
Apple is a marketing company first and foremost. Why give users control, the only way to actual privacy, when not giving them control lets you make more money off of them and marketing privacy will fool people into believing you offer actual privacy?
> PS: Apple phones can acquire a GPS lock without cellular connectivity. I’ve relied on this feature many times in the Australian outback.
And yet despite this, your iPhone will send your location back to Apple, including where you were while you were offline, as soon as it gets network connectivity.
> Oh, and you can turn Find My feature off if you’re that paranoid.
That doesn't stop Apple from getting your location.
You seem to believe I'm a Google apologist. You're wrong. I've criticized Google plenty on this forum. I'm a Firefox-using user control apologist.
When it was finally integrated into iOS, it could just barely set timers. What a waste of an amazing team.
People I know just create appointments, do hands free calling and simple stuff that Siri is pretty good at.
I hate when I make these posts when I know something, and have to eat down votes for it. Believe what you will, but it's not true, not even close to true, or in the same ballpark as truth, and it's not some contentious half truth.
It's a bizarre wholly new rumor that should strike everyone reasonably acquainted with AI as trivially wrong, given timelines.
14 years here, I at least had come to expect common courtesy on what is asserted without evidence, can be dismissed without evidence.
Siri could answer questions it no longer can.
When I ask it to play a specific song in the car, it plays a song at random. This has been a known issue for 3+ years: https://discussions.apple.com/thread/253216090?sortBy=best
Literally the only two things I can successfully use it for is setting timers and reminders. Nothing else works robustly.
You say you've worked in the Siri team for 14 years?
Print this out, roll it up, and bop your managers on the nose with it:
YOU'RE MISSING THE BOAT.
You haven't bought a ticket, and you're not even at the pier!
While you guys are still "thinking" about generative AI integration, millions of your users like me have switched to using the Open AI app because it actually understands what I'm saying and can answer usefully.
Just the voice recognition part of Siri alone is being absolutely blown out of the water by demos thrown together by individuals, not entire teams of people working at what is soon to be the former biggest company in the world: https://www.youtube.com/watch?v=WfvLIEHwiyo
If I get a message from my Mum that's not in English, Siri used to say some hilarious verbatim English pronunciation of the words. Total gibberish.
The big improvement now is that she'll just say: "A message in a foreign language" or something to that effect. Simply refusing to do anything useful. The Open AI app can read the same text out loud to me, no problems.
How is that "improving over time"!? What are you guys doing over there? All I see is regressions.
PS: It seems that there aren't even that many people in the Siri team: https://9to5mac.com/2019/01/03/apple-siri-amazon-alexa-hirin...
PPS: Zuckerberg bought 350,000 NVIDIA H100 accelerators. https://www.pcmag.com/news/zuckerbergs-meta-is-spending-bill...
Does the Siri team even have access to a single H100 card? A few dozen maybe? I haven't heard a peep about Apple buying anything from NVIDIA in the last year...
I think it was actually demonstrating a different problem: Siri's inconsistency of answers. Sometimes you will ask Siri a thing, and it will get it right. Other times, the exact same query will give something totally different (and totally wrong).
Anecdotally, I've encountered this issue less and less over the year. IMO the public has a brain worm regarding Siri's badness that doesn't really match reality. I'm guilty of the same thing—and yet I constantly see people using Siri for things I would have bet money it couldn't do, successfully. Just recently I asked it a followup question, and it worked perfectly. I didn't think Siri could do that. I was stuck in ChatGPT mentality, or I wouldn't have even tried.
I am not claiming Siri is perfect. I am not claiming it's as good as anyone else (largely because I haven't tried the other voice assistants in a long time). I'm not even saying it's particularly good, let alone "good enough". Apple has dropped the ball, no question.
However, I am saying it's better than online sentiment would have you believe.
Apple doesn't have the right kind of "vision" for what they could achieve with Siri, and they don't have the stomach for the type of risk-taking that it would take to get there even if they did.
If they did, we'd be hearing about Apple buying hundreds of thousands of GPUs, not Facebook.
"Apparently there's a senior exec in Apple that is even more of a puritan than the "AI safety" nuts and forced Siri to be dumbed down to prevent even the slightest risk of any impropriety.
With people like that in charge, innovation can't occur."
And it seemed to be a facetious comments. It's how "apparently" is used in colloquial speech nowadays.
Some of it's rather baffling. Like being able to convert certain common units but not others.
Specific example of Siri seemingly recognizing exactly what I said, but playing something totally different. (In this case it is another song from my library, but I'm not usually in a position to take a screenshot or record the screen when I'm driving.) https://i.imgur.com/t9dnoTx.png
What would be an example of unit conversion you need to get done while en route?
Here’s an actual example I did last week: “how much does a gallon of gas weigh?” Response: open your phone to search the web. I just tried it again tonight and it gave the answer directly.
I get a time about half the time, and then if I immediately ask how much left on the timer, Siri informs me there are no timers. Which then promptly goes off.
(Folks irl do this too... damn you english language...)
So they called nine-eleven, like any piggy would
The sent out Rambo, just as fast, as they could
I figure I am just holding it wrong.
Millions of users use the iPhone. So many more than any hardware that Google might have. But here's the other caveat, Google, Microsoft and Amazon all have their own huge cloud infrastructure while Apple doesn't. This means if Apple were to make a push for generative AI like the rest of the companies are doing it would cost them a lot. Everyone with an iPhone will want to try it, it would be a huge launch.
So... what's the answer to this? Well, it's pretty obvious. Ship small and efficient models that might have much more limited capabilities but still might be good enough particularly if they're well integrated with the rest of the software. Basically make Siri much smarter and capable.
It's kind of what Google has started to do as well with their 8 pro. There's a couple features that run entirely on-device. It's pretty exciting if Apple manages to get something that feels good running on device; that's the kind of machine learning i'm excited about.
I think they’ll want local because it fits into their profit margins and pushes higher end hardware. Imagine really having a reason to buy an iPhone18Pro or whatever.
And I think llms will be able to scale digitally if they are compact enough to run locally (ie, Apple pays for development once, deploys it and then all operational costs are on the client). This is really different from openAI, and other cloud LLMs, where each user has a marginal cost to operate so requires a monthly recurring cost.
i don’t expect Apple to focus on a clone of ChatGPT or Gemini. They are more likely to use it for photo operations, control of Siri, and to make each of their apps and features work better together.
And now…they snagged DarwinAI.
Are you referring to the Apple Beta Software Program? Would you rather them _not_ allow people to test their beta software so more bugs make it into the public release?
It's worth noting that beta tests are there to catch these bugs, it's literally the point.
https://uwaterloo.ca/systems-design-engineering/news/alumni-...
[1] https://plex.tv/
Just as they were in laptops, MP3 players, smartphones, and wearables. That Apple is "behind" means nothing in terms of future success.
> …and they refuse to make key acquisitions.
In fact, Apple bought 30+ AI startups last year, more than any other tech giant. This HN post is about yet another AI aquisition.
The difference is the things that you mentioned are hardware.
AI is software and data dependent. First movers have a big advantage in that space.
That aside, Apple has been doing a lot of AI for quite a while now, and executing most of it very well.
Not saying that’s cutting edge but there is more to the AI division than meets the eye.
https://www.apple.com/ios/ios-17/pdf/iOS_All_New_Features.pd...
Apple is going to force people to use its crappy AI models for some BS reason. Some bullshit about privacy or whatever. It's so dumb, it will spell the end of their dominance of the device category in the US. The models have an extremely standardized, fungible interface, and Apple's not going to let OpenAI or whoever plug into their first party Siri or Messages or whatever text, vision and audio interfaces. It will be the most anti-consumer position to take, because it will prevent users from choosing the best models, not just the "most private" or "most X" or whatever aesthetic or subjective concerns users may have, and it will only take a while to impact consumers because Google is really shitty too.
When was the last time it made sense to pay attention to an iOS platform technology? Developers have chosen, for the most part, to make the specific platform concerns someone else's problem.
Maybe not since the Retina display has anything they have done really mattered to me specifically. I have "just" ignored Swift completely. MPS is cool and Apple Silicon is cool but it's not my problem, it's "just" yet another backend to the thing I actually want to use, like Torch or whatever. Same thing with Siri. Mobile Safari is perpetually deeply behind. Mobile Safari on iOS doesn't have WebXR, and they're aspiring to be an XR company. I ignored the Watch, and I'm pretty much ignoring the Apple Vision Pro, except for supporting it as a Unity target. In fact, even on the AVP and iPhone, all the innovation seemingly is being done in Unity, and maybe only because Epic is beefing with them.
I get that people write applications in Swift or whatever. But I don't think it necessarily makes any sense to do that. Just because something happens, doesn't mean it makes sense.
Apple doesn't give you access to any of the first party APIs, which is what developers actually want. The AI thing will be the straw that breaks the camel's back.
In fact, what piece of software of theirs do you like to use that you aren't basically forced to? Like if the private APIs Photos or Messages used were open to any third party app, well of course we'd all be using third party apps to save our camera photos and send messages with compatibility with people too low tech to use an App Store. Or maybe they would use alternatives.
This is just factually wrong, it has WebXR, I tested it myself on AVP recently. You just gotta enable a toggle in Safari settings once, and then you are able to consume any WebXR content just fine.
> You just gotta enable a toggle in Safari settings once
That is part of the problem. Like why excuse their nonsense? Why is any of that behind a flag?
But the situation with WebXR as a pure software development component is a bit simpler. Obviously Apple has the resources to fully develop and release WebXR if it was a priority. My guess is, they are playing their cards close to their chest on how open to make Vision Pro. They won't release WebXR until they've determined it's not a threat to their ambitions. That could be because they scale down their ambitions so it doesn't matter much any more, or alternatively, they successfully establish a fully closed / locked down ecosystem and aren't worried that WebXR could pose a threat.
I would argue that AI-as-a-Product is a fairly recent development and probably why our expectations are skewed. I read most comments like yours as "if it's not AGI then its not even AI". Which is silly of course.
- Apple is making a multi-modal LLM.
- There are a lot of pre-existing multi-modal LLMs that perform well.
- Their interface is essentially standardized. You could call it an API to AI models.
- Apple will utilize a standardized interface when developing and deploying their multi-modal LLM.
- The end user will not be able to "choose" "which" multi-modal LLM their end-user software on iOS devices will interact with.
- This is really bad, because Apple is not likely to make the best multi-modal LLM, or to have the best multi-modal LLM at all times, so it harms the end-user.
(the ai features first rolled out like a decade ago bud, but even if they hadn’t that’s some peak wish-casting right there. It’s incredibly unlikely to ever actually happen, but because you’re so wrapped up in parasocial attachments towards/against these brands, it seems like a very real and possible idea! Another place it shows up a lot is “nvidia should/will stop releasing consumer gpus!” lol - people have been predicting that for years even before AI and it never made any sense even now, but certainly not back then. Just a Rorschach blot reflecting people’s desires back at them.)
> It’s incredibly unlikely to ever actually happen
Opening up the first party APIs? We might have an alternative app store by the end of the year, in the EU, which I didn't think was likely either. Anything seems to be possible, suddenly.
There's a low possibility that Apple specifically will "solve the summer school scheduling problem," and a high possibility someone else will. If an Android phone lets me do that, "bud," I'm going to stop using an iPhone.
It is delusional to think a stable company would follow the research hype of whatever model people are talking this week.
Ah yes, I too dislike privacy or whatever.
/s
> In fact, what piece of software of theirs do you like to use that you aren't basically forced to?
Let's see, I use Mail, Calendar, Notes, and Books because they're good enough that I don't want to switch to something else. Final Cut Pro is excellent, I use it on both my Mac and my iPad.
My favorite piece of Apple software has to be Freeform I think. If I'm on a call with someone and need to sketch something out, I can just quickly open Freeform and do it from there. Even neater, I can open the same drawing on my iPad and use the Apple Pencil, which will sync in real time with the app on my Mac while on the call. And yeah Excalidraw exists, but Freeform has all of the nice Apple Pencil shortcuts that you can't get in Excalidraw.