Personally, I don't really care if they ever ship a working AI feature as long as they don't neglect and shittify the rest of iOS/macOS in the process.
IMO there has been something rotten in the Siri house for 7+ years. It was bad even back then compared to competitors like alexa or google on a pure voice assistant basis. Open-ish models like Whisper translated speech to text far better for years, with local models, and they just sit on their worse software anyway. The hardware for their microphones is phenomenal too, which makes it an extra worse shame.
To me it's been wild to be using whisper for stunningly good transcriptions and whenever I ask siri to set a reminder for "vacuuming", I end up getting a reminder to "fuck you". which /is/ funny, to be fair.
The macOS desktop is fully neglected. They haven't touched the activity monitor in 20 years. Keyboard input for Java swing never gets fixed. Alt tab on mac doesn't show window previews.
If I had to guess, I'd say they're coming up against the limits of LLMs where LLMs are ... stupid. There's a quality issue that is difficult to get around when you try to make an LLM do something out of its comfort zone and it can suggest you to do dangerous things very easily.
I wish they would forget the really advanced features and just focus on little ones. Like particularly with Alexa back when I used to use that, one of the major frustrations was figuring out the magical word order that would get it to do what I wanted, it would only do a thing if I said it in a certain way otherwise it would get condfused or do the wrong thing. An LLM would be great at making that not happen.
But instead of just that little bit, they're trying to do Star Trek levels of computer voice control, missing out on incremental improvements.
It's a demonstration of what an entity can do, and will do. The only innovation from Apple since Job's died were unfinished efforts that completed under Cook. There will be no future innovation from Apple. There maybe acquisitions and optimizations around current and existing positions, nothing more. Apple is IBM now and has been for some time. Apple doesn't know how to develop software or deliver it to market. It has no leadership, only management, and managers are not known for relinquishing their position so I don't see any reason this will ever change. This pattern is endemic in Silicon Valley as old products have run their course. It's like the automobile from the 1900s just playing itself out. Phones and computers are a commodity I don't expect any differentiation within anymore. As to Siri and AI it's an emerging space that Apple can't even deliver on a strategic partnership for within 2 complete phone and iOS release cycles. I'm not sure how an institution like Apple can or will ever overcome its own misaligned culture. Which brings me back to it being IBM.
This is a great example of when a company needs to make a new product and not simply add capability to an old one. They decided to go with Apple Intelligence, which was partway there, but leveraging Siri branding is a mistake. It's also an organizational nightmare. Ok Siri people, go continue to support all the zillions of users you currently have and all their bizarre use cases, and we'll just one day make Siri smarter? Bad idea.
They need to leave Siri behind and make a whole new thing, roll it out to their most loyal and spendy customers, and then let it make its way down the customer chain as it gains more capability.
> They need to leave Siri behind and make a whole new thing
Keep Siri! Siri can become the dumb, reliable agent. When I want my garage door opened, I ask Siri. When I want something more thoughtful, I ask this other thing.
i have found nothing in my usage to suggest that current Siri can reliably open a garage door. Or do anything else - it’s actively worse than 5 years ago.
To the extent that it's worse, it's worse because they try to layer more stuff on top of it that ends up conflicting with the basic set a timer, trigger HomeKit, send a text message functionality.
This is exactly what Google did, probably for this exact reason: When you want to use Gemini instead of Google Assistant, you need to actively go and "change" the assistant that's on your phone. You can switch back whenever, but you choose if you want the old reliable behaviour or the fancy new AI.
They need to keep existing Siri capabilities, including third-party app integrations (SiriKit, Siri Shortcuts). That’s difficult to make consistent with a “whole new thing”.
One thing that I find very interesting is that in the past year Apple marketed new iPhones mostly through its supposed AI features while consumers in two major markets (EU and China) can’t use these features. It appears like Apple hastily tried to jump on the AI bandwagon without having a real strategy.
Apple has the best consumer local hardware for AI, and then ... Siri. How did the gap between hardware and software that could fully utilise it grow so large?
I think pretty much because the neural engine hardware stuff was to accelerate various existing ML features (face detection, voice recognition, etc) and not any kind of foresight about how ML would become relevant for LLM-type AI features (which can mostly use the cloud anyway so local isn't that important).
It's not "accelerate", that isn't important on a battery powered device and you shouldn't primarily think about it as a developer. Everything is about perf/power ratio.
The LLM does run on the neural engine though. LLMs are simple architecturally.
What a colossal waste of money that could have went towards tackling bugs, improving the Siri we have now, etc. I know many consumers want new shiny, but so many of us just want them to improve testing/bug hunting. For a while I almost thought they wouldn't get baited by the hype, or at least have their product actually working before announcing it to the world with the dumbest name possible.
Instead we have things like Image Playground, which are not only their own app but apparently so important that they get a badge in Settings until you acknowledge its existence.
> Walker said the decision to delay the features was made because of quality issues and that the company has found the technology only works properly up to two-thirds to 80% of the time. He said the group “can make more progress to get those percentages up, so that users get something they can really count on.”
Sounds about right for generative AI running on smaller local models. For consumer tech having it fail in weird ways 1/3 to 1/5 of the time is unacceptable.
Take the L man, you are sitting on 3 countries worth of GDP in cash. Buy Mistral, move on. That's what Amazon (basically) did and they're about to ship [1].
Why not let users have default open-source model and “free” limited token access to a model where they can pay extra to use further or swap. Like Perplexity does with their surface. Over time they can build their Sonar equivalent.
Didn't MS already enter into some sort of agreement with Mistral?
Also - I don't think Amazon (or more specifically Alexa) is exactly worth emulating. I am willing to bet there are a substantial number of issues with their new Alexa features that don't happen with Siri specifically because of its reduced scope (can't browse the internet and shop for you, etc.)
That's what Amazon (basically) did and they're about to ship
Ship what?
Will what they ship be anything better than just using ChatGPT? Right now companies don't need to buy anything, because everyone has what is essentially the exact same product. Maybe OpenAI's is a bit better than everyone else's in everyday use, but a run of the mill AI offering with maybe? better usability is still a run of the mill AI offering.
That's the problem. There really is no differentiation out there. You know you're all using pretty much the same thing when fanboy-ism rears its ugly head. "My ChatGPT is better than your Claude!" Or vice-versa ad-nauseum.
The only way to win in that environment is to leverage your other businesses to force use of your models onto the market.
But I guess, kudos to Apple for at least trying to improve their product. They really don't need to do that. But how long will that attitude last? Sooner or later even Apple will start foisting their models onto the markets by leveraging their other business lines. Like Microsoft has figured out recently, "hey, we don't need to be good at AI to win. We just need to force everyone to use our models!" Expect the same from Amazon, FB, Google et al.
Alexa has integrations with a bunch of third party services - Uber, smart home, etc. - not to mention Amazon shopping, and Alexa+ taps into that preexisting wide ecosystem. So as an agent it's already a lot more effective than ChatGPT for "getting things done". I think they've beaten Apple at their own game by at least a couple years.
The possibility to do things does not mean it's able to actually finish the task, at least not until you and I have an opportunity to try it out ourselves and verify that things actually work, and work at least as well as you do it yourself on a first party app. Why are people so eager to promote these fairytales?
They know they have a run of the mill model. Maybe even not as good as Gemini or ChatGPT. (or even DeepSeek? Who knows?) So they leverage other business to force use of the models onto the market.
Everyone else will do the exact same thing. Why? Because I don't think their models will be better in any material way than the slop Amazon is slinging.
In Amazon's case, Alexa. The voice assistant market seems like a great place to leverage this tech but two of the three major services in this space (Apple, Google, and Amazon) have nothing to show and Google is only recently getting their act together.
How hard can it be? Models like Deepseek R1 are state-of-the-art and open source. Apple could just take one of these, tune it, and put a nicer user interface on top of it.
Sure, maybe it is embarrassing for Apple that they haven't made their own model, but the price of LLM AI is dropping like a rock, and eventually, Apple's large user base, distribution, UI design, and convenience will win the day.
Yeah, you need to say “change” instead of “set”. And then there’s the problem that if the alarm was turned off, it remains turned off with the new time, and Siri doesn’t bother to call attention to that.
Oof didn't even notice the latter part. Also when saying "change" it shows you every alarm you have (even if they're disabled, I have like 10) and forces you to interact with it on screen. There's no voice-only interaction that lets a user update tomorrow's wake-up alarm. It's so weird, it's like they don't use their own product.
Siri did basically not change since it was released, it can barely create timers and dictate really really basic text, but still fails 50% of the time.
IMO the last decade has mostly been a stagnation in software quality at Apple, both in design and code quality. Apple used to have to focus on users and quality because they were both an underdog and premium priced. Now it feels like they're slowly turning into Disney in the sense that most decisions are how to extract as much money as possible from you.
* WebGL is half-assed in safari because of "security" risks, but reeks that they don't want to risk mobile gaming bypassing the app store.
* They've let mobile gaming be ruined by abusive micro-transaction games.
* iTunes Genius was SO GOOD at recommending new music, but now that they can just grab a monthly fee music discovery on Apple music is very meh.
It doesn't surprise me in the least that that their top-down product driven development cycle can't handle a sudden, new technology (that is arguably overhyped).
The tragedy is that the only mobile alternative is an operating system run by an adtech company and windows is...regressing again after some promise.
> WebGL is half-assed in safari because of "security" risks, ...
Not commenting on anything in the latter clause of your sentence, but I can tell you the security risks are very, very real. Having worked on GPU architecture as well as a browser engine -- so, so many features/guardrails/guarantees that we take for granted on desktop PCs are basically in their infancy when you look over at the GPU world. Address-space isolation, virtualization of hardware resources, separation of virtual machines, privilege isolation -- when you go to the GPU, all of that is kinda-there if you're on newer hardware, but even then without anywhere near the level of coverage or battle-testing that modern CPUs & kernels have. For example, Pascal (i.e. that old 1060 your friend is still using in their kid's desktop) doesn't even have per-process memory isolation -- a buffer overflow in your browser engine could attack any GPU process on your machine! Even the vendors themselves won't claim to support "proper" secure multi-tenancy until the Hopper era, and again -- even that's much less well-tested than the CPU-level isolation that browsers normally rely on. Consider how much of a security impact PPO/site-isolation has had: so many attacks that used to be possible (e.g. your password being stolen off of your banking site by a malicious ad in the sidebar) are now essentially impossible. That only works because of the deep strength and reliability of address-space separation in modern hardware platforms.
So to take code written by random people on the internet, in a dynamic language, JIT compile it, and then give it access to your GPU -- well, I hope you can understand why that might put the security people on edge. Again, I don't mean to dismiss the rest of your comment, but I want to clarify that when it comes to punching holes in the browser sandbox, the security concerns are really very real.
> They've let mobile gaming be ruined by abusive micro-transaction games.
As much as I'd love to blame Apple here, I'm not sure what they could have done differently. Once the race to the bottom in terms of up-front pricing started ad-supported and microtransaction-supported games were inevitable. Meanwhile, games with cash shops were already popular in South Korea[1] and "free" games allowed the same model to gain a foothold here.
I would love if Apple labelled all gacha games as 18+ because they're gambling simulators, but it's really up to consumers to avoid non-gambling MTX if they don't like this business model.
John Gruber also ripped Apple a new one: https://daringfireball.net/2025/03/something_is_rotten_in_th...
Personally, I don't really care if they ever ship a working AI feature as long as they don't neglect and shittify the rest of iOS/macOS in the process.
Yeah, John seemed pretty upset. I do think they’ve already compromised iOS/macOS enough, though: https://taoofmac.com/space/blog/2025/03/14/1830
IMO there has been something rotten in the Siri house for 7+ years. It was bad even back then compared to competitors like alexa or google on a pure voice assistant basis. Open-ish models like Whisper translated speech to text far better for years, with local models, and they just sit on their worse software anyway. The hardware for their microphones is phenomenal too, which makes it an extra worse shame.
To me it's been wild to be using whisper for stunningly good transcriptions and whenever I ask siri to set a reminder for "vacuuming", I end up getting a reminder to "fuck you". which /is/ funny, to be fair.
The macOS desktop is fully neglected. They haven't touched the activity monitor in 20 years. Keyboard input for Java swing never gets fixed. Alt tab on mac doesn't show window previews.
If I had to guess, I'd say they're coming up against the limits of LLMs where LLMs are ... stupid. There's a quality issue that is difficult to get around when you try to make an LLM do something out of its comfort zone and it can suggest you to do dangerous things very easily.
I wish they would forget the really advanced features and just focus on little ones. Like particularly with Alexa back when I used to use that, one of the major frustrations was figuring out the magical word order that would get it to do what I wanted, it would only do a thing if I said it in a certain way otherwise it would get condfused or do the wrong thing. An LLM would be great at making that not happen.
But instead of just that little bit, they're trying to do Star Trek levels of computer voice control, missing out on incremental improvements.
That ship has unfortunately sailed.
It's a demonstration of what an entity can do, and will do. The only innovation from Apple since Job's died were unfinished efforts that completed under Cook. There will be no future innovation from Apple. There maybe acquisitions and optimizations around current and existing positions, nothing more. Apple is IBM now and has been for some time. Apple doesn't know how to develop software or deliver it to market. It has no leadership, only management, and managers are not known for relinquishing their position so I don't see any reason this will ever change. This pattern is endemic in Silicon Valley as old products have run their course. It's like the automobile from the 1900s just playing itself out. Phones and computers are a commodity I don't expect any differentiation within anymore. As to Siri and AI it's an emerging space that Apple can't even deliver on a strategic partnership for within 2 complete phone and iOS release cycles. I'm not sure how an institution like Apple can or will ever overcome its own misaligned culture. Which brings me back to it being IBM.
This is a great example of when a company needs to make a new product and not simply add capability to an old one. They decided to go with Apple Intelligence, which was partway there, but leveraging Siri branding is a mistake. It's also an organizational nightmare. Ok Siri people, go continue to support all the zillions of users you currently have and all their bizarre use cases, and we'll just one day make Siri smarter? Bad idea.
They need to leave Siri behind and make a whole new thing, roll it out to their most loyal and spendy customers, and then let it make its way down the customer chain as it gains more capability.
> They need to leave Siri behind and make a whole new thing
Keep Siri! Siri can become the dumb, reliable agent. When I want my garage door opened, I ask Siri. When I want something more thoughtful, I ask this other thing.
Right I mean leave it behind in its current garage door opener form.
i have found nothing in my usage to suggest that current Siri can reliably open a garage door. Or do anything else - it’s actively worse than 5 years ago.
To the extent that it's worse, it's worse because they try to layer more stuff on top of it that ends up conflicting with the basic set a timer, trigger HomeKit, send a text message functionality.
This is exactly what Google did, probably for this exact reason: When you want to use Gemini instead of Google Assistant, you need to actively go and "change" the assistant that's on your phone. You can switch back whenever, but you choose if you want the old reliable behaviour or the fancy new AI.
I don't work there, so this is just a guess. It's likely a "whole new thing" with Siri being the name kept for marketing.
They need to keep existing Siri capabilities, including third-party app integrations (SiriKit, Siri Shortcuts). That’s difficult to make consistent with a “whole new thing”.
One thing that I find very interesting is that in the past year Apple marketed new iPhones mostly through its supposed AI features while consumers in two major markets (EU and China) can’t use these features. It appears like Apple hastily tried to jump on the AI bandwagon without having a real strategy.
Apple has the best consumer local hardware for AI, and then ... Siri. How did the gap between hardware and software that could fully utilise it grow so large?
Siri has always been of dubious quality, the gap has been growing for a decade.
I think pretty much because the neural engine hardware stuff was to accelerate various existing ML features (face detection, voice recognition, etc) and not any kind of foresight about how ML would become relevant for LLM-type AI features (which can mostly use the cloud anyway so local isn't that important).
It's not "accelerate", that isn't important on a battery powered device and you shouldn't primarily think about it as a developer. Everything is about perf/power ratio.
The LLM does run on the neural engine though. LLMs are simple architecturally.
What a colossal waste of money that could have went towards tackling bugs, improving the Siri we have now, etc. I know many consumers want new shiny, but so many of us just want them to improve testing/bug hunting. For a while I almost thought they wouldn't get baited by the hype, or at least have their product actually working before announcing it to the world with the dumbest name possible.
Instead we have things like Image Playground, which are not only their own app but apparently so important that they get a badge in Settings until you acknowledge its existence.
Here's the bit that stood out to me:
> Walker said the decision to delay the features was made because of quality issues and that the company has found the technology only works properly up to two-thirds to 80% of the time. He said the group “can make more progress to get those percentages up, so that users get something they can really count on.”
Sounds about right for generative AI running on smaller local models. For consumer tech having it fail in weird ways 1/3 to 1/5 of the time is unacceptable.
https://archive.ph/jnIV5
Take the L man, you are sitting on 3 countries worth of GDP in cash. Buy Mistral, move on. That's what Amazon (basically) did and they're about to ship [1].
[1]: https://www.aboutamazon.com/news/devices/new-alexa-generativ...
I wouldn't use Amazon as an example until we see Alexa's real world performance.
Mistral is French; that's their whole value. A US company can't buy them.
What? Of course a US monopoly could buy them, anything else would be communism they always tell us.
Why not let users have default open-source model and “free” limited token access to a model where they can pay extra to use further or swap. Like Perplexity does with their surface. Over time they can build their Sonar equivalent.
Or, heck, just buy Perplexity. You have the cash.
Didn't MS already enter into some sort of agreement with Mistral?
Also - I don't think Amazon (or more specifically Alexa) is exactly worth emulating. I am willing to bet there are a substantial number of issues with their new Alexa features that don't happen with Siri specifically because of its reduced scope (can't browse the internet and shop for you, etc.)
That's what Amazon (basically) did and they're about to ship
Ship what?
Will what they ship be anything better than just using ChatGPT? Right now companies don't need to buy anything, because everyone has what is essentially the exact same product. Maybe OpenAI's is a bit better than everyone else's in everyday use, but a run of the mill AI offering with maybe? better usability is still a run of the mill AI offering.
That's the problem. There really is no differentiation out there. You know you're all using pretty much the same thing when fanboy-ism rears its ugly head. "My ChatGPT is better than your Claude!" Or vice-versa ad-nauseum.
The only way to win in that environment is to leverage your other businesses to force use of your models onto the market.
But I guess, kudos to Apple for at least trying to improve their product. They really don't need to do that. But how long will that attitude last? Sooner or later even Apple will start foisting their models onto the markets by leveraging their other business lines. Like Microsoft has figured out recently, "hey, we don't need to be good at AI to win. We just need to force everyone to use our models!" Expect the same from Amazon, FB, Google et al.
Alexa has integrations with a bunch of third party services - Uber, smart home, etc. - not to mention Amazon shopping, and Alexa+ taps into that preexisting wide ecosystem. So as an agent it's already a lot more effective than ChatGPT for "getting things done". I think they've beaten Apple at their own game by at least a couple years.
The possibility to do things does not mean it's able to actually finish the task, at least not until you and I have an opportunity to try it out ourselves and verify that things actually work, and work at least as well as you do it yourself on a first party app. Why are people so eager to promote these fairytales?
You've just reiterated what I was saying.
They know they have a run of the mill model. Maybe even not as good as Gemini or ChatGPT. (or even DeepSeek? Who knows?) So they leverage other business to force use of the models onto the market.
Everyone else will do the exact same thing. Why? Because I don't think their models will be better in any material way than the slop Amazon is slinging.
Ship what?
In Amazon's case, Alexa. The voice assistant market seems like a great place to leverage this tech but two of the three major services in this space (Apple, Google, and Amazon) have nothing to show and Google is only recently getting their act together.
I meant Alexa the home assistant (100M MAU, currently mostly setting timers), not Alexa the web-based chat tool that no one uses.
Once Alexa (the home assistant) gets Claude capabilities natively, I can see it becoming infinitely more useful.
How hard can it be? Models like Deepseek R1 are state-of-the-art and open source. Apple could just take one of these, tune it, and put a nicer user interface on top of it.
Sure, maybe it is embarrassing for Apple that they haven't made their own model, but the price of LLM AI is dropping like a rock, and eventually, Apple's large user base, distribution, UI design, and convenience will win the day.
This is just a director level person. Either
1. This is not important enough that a senior person (VP or above) is assigned to take care of Generative AI (the biggest shift since Mobile/Cloud) or
2. Robby Walker is cannon fodder and is polishing his resume.
The reporting is based on leaked details of an internal meeting.
"During the all-hands gathering [...] people familiar with the matter have said." is journalist speak for "I heard this from inside sources".
Maybe next time marketing and product will ask engineering if something’s ready for release before commissioning and airing multimillion dollar ads.
My Siri experience last night as I was going to bed, not wanting to look at my device before I closed my eyes: 'set my alarm to 10am'
Rather than realize it was late at night and I was referring to my normal wakeup alarm, Siri added a second alarm.
Yeah, you need to say “change” instead of “set”. And then there’s the problem that if the alarm was turned off, it remains turned off with the new time, and Siri doesn’t bother to call attention to that.
Oof didn't even notice the latter part. Also when saying "change" it shows you every alarm you have (even if they're disabled, I have like 10) and forces you to interact with it on screen. There's no voice-only interaction that lets a user update tomorrow's wake-up alarm. It's so weird, it's like they don't use their own product.
When will the big AI companies sue Apple because of bundling?
Siri did basically not change since it was released, it can barely create timers and dictate really really basic text, but still fails 50% of the time.
Their fake AI promises are just embarrassing.
Sounds like someone's doing a Steve impression!
IMO the last decade has mostly been a stagnation in software quality at Apple, both in design and code quality. Apple used to have to focus on users and quality because they were both an underdog and premium priced. Now it feels like they're slowly turning into Disney in the sense that most decisions are how to extract as much money as possible from you.
* WebGL is half-assed in safari because of "security" risks, but reeks that they don't want to risk mobile gaming bypassing the app store.
* They've let mobile gaming be ruined by abusive micro-transaction games.
* Apple ignores user bugs unless they get a lot of press ( my current frustration: https://discussions.apple.com/thread/255473542 )
* iTunes Genius was SO GOOD at recommending new music, but now that they can just grab a monthly fee music discovery on Apple music is very meh.
It doesn't surprise me in the least that that their top-down product driven development cycle can't handle a sudden, new technology (that is arguably overhyped).
The tragedy is that the only mobile alternative is an operating system run by an adtech company and windows is...regressing again after some promise.
At least there's always linux? :-/
> WebGL is half-assed in safari because of "security" risks, ...
Not commenting on anything in the latter clause of your sentence, but I can tell you the security risks are very, very real. Having worked on GPU architecture as well as a browser engine -- so, so many features/guardrails/guarantees that we take for granted on desktop PCs are basically in their infancy when you look over at the GPU world. Address-space isolation, virtualization of hardware resources, separation of virtual machines, privilege isolation -- when you go to the GPU, all of that is kinda-there if you're on newer hardware, but even then without anywhere near the level of coverage or battle-testing that modern CPUs & kernels have. For example, Pascal (i.e. that old 1060 your friend is still using in their kid's desktop) doesn't even have per-process memory isolation -- a buffer overflow in your browser engine could attack any GPU process on your machine! Even the vendors themselves won't claim to support "proper" secure multi-tenancy until the Hopper era, and again -- even that's much less well-tested than the CPU-level isolation that browsers normally rely on. Consider how much of a security impact PPO/site-isolation has had: so many attacks that used to be possible (e.g. your password being stolen off of your banking site by a malicious ad in the sidebar) are now essentially impossible. That only works because of the deep strength and reliability of address-space separation in modern hardware platforms.
So to take code written by random people on the internet, in a dynamic language, JIT compile it, and then give it access to your GPU -- well, I hope you can understand why that might put the security people on edge. Again, I don't mean to dismiss the rest of your comment, but I want to clarify that when it comes to punching holes in the browser sandbox, the security concerns are really very real.
> They've let mobile gaming be ruined by abusive micro-transaction games.
As much as I'd love to blame Apple here, I'm not sure what they could have done differently. Once the race to the bottom in terms of up-front pricing started ad-supported and microtransaction-supported games were inevitable. Meanwhile, games with cash shops were already popular in South Korea[1] and "free" games allowed the same model to gain a foothold here.
I would love if Apple labelled all gacha games as 18+ because they're gambling simulators, but it's really up to consumers to avoid non-gambling MTX if they don't like this business model.
[1] https://en.wikipedia.org/wiki/MapleStory
So many things Apple permits were not, nor do they remain, inevitable.
Agree, but they're still building great hardware.