John Giannandrea to retire from Apple(apple.com) Additional reporting: https://www.macrumors.com/2025/12/01/apple-ai-chief-retiring... |
John Giannandrea to retire from Apple(apple.com) Additional reporting: https://www.macrumors.com/2025/12/01/apple-ai-chief-retiring... |
Been like this many versions, but dunno if it does it with the latest.
Just checked I’m on iOS 18.2.
Was the MS gig so bad?
It could potentially help tremendously but for that they would need to understand the usefulness of LLMs and tool usage.
What did he even do for Apple's AI strategy for 7 years?
Apple is still far behind in doing anything useful with AI.
[Now I remember- he came in with the metaweb acquisition at a time when search was going all-in on knowledge graphs and semantics]
I hope they don’t do anything remotely like that at Apple.
I am completely okay with the Apple approach to date (privacy and late mover cost advantage over progress and burning money/raising prices).
At this point, their investment to ship a better Siri is nearly zero if they take an open source model and run it on the device. Did John really mishandle it, or did he realise this and decide not to burn $BILS of cash and play the long game instead?
I worked pretty closely with him and his team for a bit at Google, and he seemed like a great human being, in addition to being a great engineer. I wouldn't read too much into a few-month stint at Microsoft.
Perhaps you are not getting it rammed down your throat because you’re not a business user? On personal editions one area where AI has been a failure is taking over the search bar, but you’re right, you can disable it.
We'd have working voice assistants by now. We're held up by the incessant need to game "engagement" and seek rent.
In reality users just want a goddamn voice interface to their phone. Set a timer, remind me of x next time I'm at location y. Turn on the lights. Set home air conditioning to 72.
Simple, trivial bullshit that has absolutely no monetizable worth. Because it's not profitable enough it's not worth developing at all. I'm half convinced the only reason siri and google assistant even still exist is solely and exclusively because the "other guy" has it.
People argue innovation is impossible without capitalism. I argue innovation is impossible with capitalism. If your idea isn't profitable enough it's not worth any amount of investment regardless of how beneficial the idea might be.
Then LLMs came and it still wasn't "real enough."
Honestly he’s had one hell of a career. Even if Siri sucked.
I believe that they are so arrogant that they think Meta just suck when in fact it is just a product category still looking for a valuable use case to the general public.
They market it for watching movies but most peoples already hate using headphones for too long and those are pretty comfortable nowadays.
...
...
...
Siri disappears and song continues playing
...
"Hey Siri, stop playback"
Songs stops playing.
Still. Idle hands, he should get back on that horse if he can. Go do more stuff.
A decade and a half is insane timeline in tech industry, and huge majority of users use Siri the same way today as 15 years ago, setting a timer or an alarm clock.
If they had 0 improvements over these 15 years the situation wouldn't be much different than today.
Every few years I would try to use it for a few days, then quit in frustration at how useless it was. Accidentally activating Siri is a major frustration point of using Apple products for me.
All chat bots suffer this flaw.
GUIs solve it.
CLIs could be said to have it, but there is no invitation to guess, and no one pretends you don’t need the manual.
I’m looking at you, Photos sync.
EDIT: just noticed this exact problem is on the front page in its own right (https://eclecticlight.co/2025/11/30/last-week-on-my-mac-losi...)
I've found I appreciate having Siri for a few things, but it's not good enough to make it something I reach for frequently. Once burned, twice shy.
[1] Made up title
This is why LLMs are the first conversational interface to actually have a chance of working, once you give them enough tools.
It’s such trash. Constant conditioning for garbage.
Timers and alarm clocks it is.
I get the sci-fi "wow" appeal, but even the folks who tried to build Minority Report-style 3D interfaces gave up after realizing tired arms make for annoyed users.
You know I have talked to chatGPT for maybe a 100 hours over the past 6 months. It gets my accent, it switches languages, it humors. It understands what I am saying even if it hallucinates once in a while.
If you can have chatGPT level of comprehension, you can do a lot with computers. Maybe not vim level of editing, but every single function in a driving car should be controllable by voice, and so could a lot of phone and computer functions.
> Alexa, turn on the bedroom lights.
> OK lights turn on
In the evening:
> Alexa, turn on the bedroom lights.
> I'm sorry, I don't know a device called "bedroom lights".
How is it even possible to build a computer system that behaves like this?
Just that... nobody is willing to pay much for a thing that will do some basic search, dictate a recipe, or do unit conversion, or add a thing to a list.
Maybe it’s just my imagination, but it seems like text to speech in the ChatGPT prompt window uses the context around it in the paragraph to fix what I’m saying so it is inordinately accurate compared to the rest of the iOS system.
1. Checking the current temp or weather
2. Setting an alarm, timer, or reminder
3. Skipping a music track or stopping the music altogether roughly 3 seconds after hearing the command, or 1 second after you assume it didn't work
<end of list>
It is ridiculously useless for most things though. Like I’ll ask it a question on my Apple Watch and it will do a web search and give me a bunch of useless links.
for example I say: "play comically long album title by artist on Spotify", it thinks about that for five seconds, does the bing noise, then says "playing comically long album title [special remastered edition] by artist on Spotify", and then a few seconds later starts playing the album, and if you don't wait through that whole thing it will just decide that actually you didn't want to hear the album
Tell me my next event when I’m driving.
Also quite good for making shopping lists, with some bonus amusement when you get a weird transcription and have to try to work out that "cats and soup" is "tonkatsu sauce" several days after you added it to the list.
A whole bunch of assistants have gotten way worse in the last decade by chasing features at the expense of utility. I don't care about whatever new feature my speaker has, but if it fails to play a song or check the weather, I'm PISSED.
I was excited when I recently got an iPhone 16 Pro - it comes with Apple Intelligence! Surely this is how Siri leaps into the future and starts doing things like translating for me, or responding with a photo and some basic facts when I ask who Ariana Greenblatt is, or letting me convert from Krore to USD (it gives results for rupees every time it seems?) or...
Anyways, I asked it something basic, and Siri said it would have to use Apple Intelligence. Not like, prompting me if I want to use it, just saying it's needed, then turning off. I'm pretty confused as to what Apple Intelligence is at this point, since I assumed it would be Siri. "Hey Apple Intelligence" doesn't do anything, so I ask ChatGPT. It informs me that AI is, in fact, part of Siri. I... do not know why it gave me that response.
Back to timers and alarms.
Edit - this is your daily reminder that you can NO LONGER SHUT OFF IPHONES BY HOLDING DOWN THE POWER BUTTON.
When they introduced "machine learning" it started working much worse - things that used to work 99% of the time (if you knew the incantation) now randomly fail in inexplicable ways.
And now it never gets my kids' names right. If I'm doing voice-to-text I will actually see it flash the correct spelling, and then replace it with some random phrase.
Classic Alexa, Gemini and Siri are all just intent based pattern matching systems where you brute force all of the phrases you want to match on (utterances), map those to intents and have “slots” for the variable parts. Like where you are coming from and where you are going.
Then you trigger an API. I’ve worked with the underlying technology behind Alexa for years on AWS - Amazon Lex - with call centers (Amazon Connect).
On the other hand, the capabilities and reliability of both Alexa and Google’s voice assistant have regressed once they moved to an LLM based system. It seems to be a hard problem and I don’t understand why.
I’ve find plenty of free text input -> LLM -> standard JSON output -> call API implementations. It seems like it would just be another LLM + brute force issue.
When it works!
I’ve spend days where it goes wonky and says something went wrong for anything I ask. How is it that with modern phones the voice recognition and whatnot isn’t running locally?
I'm not ever going to talk to my phone. Certainly not for the slim list of things that doing so would make more efficient than just looking at it and poking the button I need.
And there's no way I'm going to enable a battery draining 24/7 surveillance microphone on my phone on the off chance I ever do come up with a reason to give it a voice command.
But Apple really wants me using it. So much that my wife's car won't enable CarPlay unless turn Siri on. Like, there's no way to get it to put a Google map on the car's screen unless I turn this unrelated other thing on. They're happy to burn all our goodwill and convince us to buy Android phones next time (which work fine in a car without their surveillance microphone turned on).
Until then, I bought a $5 phone stand for the dashboard.
Maybe you won't, but there's still value in being able to use it hands free. "Hey Siri, call (name) on speaker" is something I regularly ask it to do while I'm driving.
Things that seemed to work reliably for me 10 years ago but now do not:
1. "Call mom". Siri has apparently forgotten who my mother is. I tried "Hey Siri <name> is my mother" and I got an error. I'm sure it's resolvable but come on.
2. "Directions to <destination>" This always used to fail when it couldn't find places, but lately, when I'm driving, Siri will respond "Getting directions to <destination>" and then... nothing. No directions come up. I have to do it 2-3 times to have the directions actually start.
I think this also interacts with countries and states that have (possibly misguided) strict laws forbidding the "touching" of phones "while driving". My experiences suggest that using Siri when driving and the device is locked, it just gives up - I sort of see the start of it working then, bam, it stops. If I retry, I suspect that I've somehow "looked" at the phone in frustration, it saw my attention and unlocked. I now wonder if where I have placed the device is making a difference.
It does seem to work much better (when driving) if the device is already unlocked.
I also see odd things when using Shortcuts for navigation. If I've previously asked for walking directions and then speak the shortcut while driving it won't give directions until I switch to the "car" icon in maps. I think it might be trying to calculate the 15Km walking directions, but it doesn't complete before I tell it, frustrated, to stop.
When Siri doesn't work it is usually the times when I need it to. This is definitely a multiplier in disastisfaction.
3. It won't reliably play music anymore! I have a good set of songs in my iPhone's Apple Music library. When I say "Hey siri, play <song/artists>", it asks me for access to Pandora (which I do have on my phone). I don't want to play it on Pandora. I have the song! I have just spent the last 10 minutes trying to figure out how to change this, and neither the youtube video I found searching nor this reddit thread (https://www.reddit.com/r/ios/comments/y18ioq/changing_the_de...) seems to work.
Amusingly (?) the reddit people have the opposite problem. They want to use a 3rd party music app but can't get their phones to stop preferring Apple-provided apps.
I can sometimes get this to work by saying "Play <song/artist> on Apple Music", but even that is not reliable.
Interesting that you've also had that problem.
In this case the lie was that Siri could understand you just like a real human but it really can't. AI chat bots have some similar problems (figuring out what it can actually do) but at least they are much better at communicating like a human.
[1] https://reddit.com/r/apple/comments/9q7ugf/it_is_truly_absur...
What is it about Siri’s architecture that causes “Set bedroom light to 30%”, a command that worked for years, to randomly stops working on a random Tuesday with no OS update or home change?
I mean, what on earth are they doing on the back end…?
When discussing a Jeopardy answer with my wife, I say "Hey Siri, who was Pol Pot". Siri said, "OK, calling Scott". So it woke up my friend at 1am..
And if I hear another "I found this on the web", I'm going to scream.
Siri is so bad it makes me want to go back to a pixel.
People increasingly seem to forgo the idea of retaining the data for themselves because they find AI products so fascinating / useful that they're just not caring, at least for the moment. I think this might swing back in the favor of Apple at one point, but right now it is kind of fascinating how liberally people throw everything at hosted AI models.
From a technology or engineering perspective, I have no idea how to work with Apple.
And for audio production - I'm just a dabbler, really, but I've been able to do some really impressive things with just GarageBand and a Fender Mustang Micro amp-plug over USB-C. It "just works" unlike my experience on Linux recently, where there are lots of little bits that are genius, but I couldn't manage to figure out how to get a basic midi synth working with a DAW that had a UI that was designed for humans. (Jack is amazing, though - being able to do arbitrary audio filter chains with random pieces of software is seriously cool.)
But mainly I wanted to share that video because Craig Federighi calling the AI/ML team "AI/MLess" is one heck of a burn
Then one time, in a job interview of all things, (I'm a PM and was asked for an example of a product I liked or didn't like) I went into this spiel, and what do you know, Siri texted the photo to my wife. :-)
So somewhere along the way they did add that feature, and apparently I didn't realize I had missed checking for it.
And then half the time it's "a screenshot of a social media post" - c'mon, read the damn text from the screenshot at least!
Also, Apple tends to recruit internally first for VP positions, which they didn't do in this case. I am sure they considered internal candidates though, but I feel they're getting a bit "desperate" in the AI game to show progress.
From TFA
Some of the teams that Giannandrea oversaw will move to Sabih Khan and Eddy Cue, such as AI Infrastructure and Search and Knowledge. Khan is Apple's new Chief Operating Officer who took over for Jeff Williams earlier this year ... Apple CEO Tim Cook thanked Giannandrea ...
Nah, IS&T has always been under the CFO, and apparently some fraction of AIML is headed under them.
1980s - silicon graphics / general magic
1990s - chief technologist, netscape
early 2000s - CTO Tellme (speech recognition)
late 2000s - CTO Metaweb (knowledge graph) -> acquired into Google
2010s - Google head of Machine Intelligence, Search, Gmail Smart Reply, etc, then took over Google Search and ML driven ranking (BERT)
2018 -> SVP ML/AI Apple to merge Siri/Core ML/all AI offerings under one roof
2023-2025 - led Apple Intelligence push
March 2025 - removed as head of Siri
Dec 2025 - retirement
would love to do an exit interview with him on the last 4 decades in building ai assistants!
-- https://x.com/markgurman/status/1995617560373706942?s=20
cv of his successor Amar Subramanya - 16 years at GDM - head of eng for Gemini chatbot/training. joined microsoft in JULY this year.. and now poached to Apple. lmao.
Now I'm weighing more on the Apple side for not making it better.
Now, it'll show a loading indicator for 5-6 seconds and then do nothing at all... or do something entirely unrelated to my request (eg responding to "hey siri, how much is fourteen kilograms in pounds" by playing a song from my music library).
My personal favourite is Siri responding to a request to open the garage door, a request it had successfully fielded hundreds of times before, by placing a call to the Tanzanian embassy. (I've never been to Tanzania. If I have a connection to it, it's unknown to me. The best I can come up with is Zanzibar sort of sounds like garage door.)
Problem is, Siri is already damaging Apple's reputation with how useless it is..
"Who is speaking?"
The same person who has been speaking the last hundred times, dammit!
https://news.ycombinator.com/item?id=43436174
An organization of Apple's size doesn't fail due to the mistakes of a single person, unless that person is the CEO.
And then my exchange business plan telling me copilot is here over and over in a giant popup screen and then saying - it's not available yet
All of that to realize Siri was kind of boring. Funny thing is it’s been over a decade and it’s maybe 20% better than it was at launch. MAYBE.
I don’t want to blame this one guy for all of that, but part of me can’t help but point at the leader and ask “you couldn’t have done better than… that?”
Haven’t found anything else it’s useful for.
Are there new functions I don't know about? I... can't think of anything else they'd add, but I literally do not understand what their engineers and managers working on Siri were doing on a daily basis. They must have been writing some code at some point. Did it just never launch? Am I simply ignorant?
I thought there is a much deeper problem, this is yet another Tim Cook hiring that didn't work. In fact I am not aware of a single SVP grade report to Tim Cook that came from outside of Apple were successful.
- Call "person"
- Call "business" (please don't say "I don't see so and so in your contacts" and on a second try, work)
- Find "place" (while driving) - Define "phrase or word" (please don't say I found this on the web)
- Set a timer or alarm
- Check the messages (in a sane way)
- Set reminders (this one surprisingly works well)
- Use intents correctly (I just want to be able to say "play 99% invisible in Overcast")
It doesn't need to do all the fancy things they show-cased last year. It just needs to do the basics really well and build from there.
While Eddie Cue seems to be Apple's SaaS man, I can't say I'm confident that separating AI development and implementation is a good idea, or that Apple's implementation will not fall outside UI timeframes, given their other availability issues.
Unstated really is how good local models will be as an alternative to SaaS. That's been the gambit, and perhaps the prospective hardware-engineering CEO signals some breakthrough in the pipeline.
A feature I would love is to toggle "answer calls on speakerphone" based on location, so that I can answer a call with my phone on the desk while I'm at home and not have my ear blasted off taking a call when I'm walking down the street.
it used to be able to set a timer or alarm 100% of the time, now sometimes it decides it needs to ask chatGPT for help.
But I'll not preted Google is any better. Last I used it would have call "CTO of the company I worked for" or "Send message to random friend-of-a-friend that I once helped" as suggested actions in the middle of the night. (Maybe it has improved now? I used to be comically bad, as was other large tech companies: https://erik.itland.no/tag:aifails)
"Hey Siri, what city am I in?"
"Calling Ian"
1. Voice to text transcriptions
2. Text to understanding
3. Adding capabilities where it can do something with #2.
The voice to text that Siri uses seems to be worse than when you are dictating using voice to text from the keyboard.
The latter gets close to 100% with my southern native English accent and does okay when I’m trying to speak Spanish. Siri messes up with English a lot more and it’s a lost cause when I try to speak Spanish.
Honestly with these assistants I'd rather just type my query. Voice input is embarrassing and error-prone. The only place that voice input is useful is in the kitchen.
I suspect they had a very carefully hand-crafted model before, and replaced it with an ML model "that will fix itself over time" - and it never did.
"Siri, play some music"
"Sorry you will need to unlock your phone to do that"
"Siri, play some music"
<music starts playing>
I'm not saying that way to solve a problem, but I refuse to believe that it has to be as bad as it is. The worst part is that it's still better than the alternative of leaving the iOS ecosystem.
Apple is just so bizarre in general. I would say "nowadays" but I think they have always been like this. It took them how many decades to add a unit converter to the iPhone? And after all that time, they buried it in a menu in the Calculator app?
Yet Siri will still tell you about a web results on your phone… but sometimes same question asked? Will check ChatGPT and give you an actual answer (15% hit rate?).
ARG!!!!
"here are some pictures of pot pot"
Recent example: Apple used to hide "search in page" in the share menu in mobile safari. Far from obvious, but at some point one discovers it because there is no other place to look for it.
Now they have finally decided to make a standard fly dropping overflow menu and hide the share button there. But interestingly you still need to open the share menu from there to find the search button.
Meanwhile other buttons that weren't as obviously misplaced in "share" like "Add to Bookmarks" are now on the top level together with the share button.
Same goes for the arguments against things like cut and paste in finder: they didn't create it back in the day and now there is a complete mythology about why cut and paste in Finder would actually be stupid and almost evil.
Anything you ask an Android device to do, or an Alexa device goes to their clouds to be 100% processed there.
Apple tried to make a small and focused interface that could do a limited set of things on device without going to the cloud to do it.
This was built around the idea of "Intents" and it only did the standard intents... and app developers were supposed to register and link into them.
https://developer.apple.com/documentation/intents
Some of the things didn't really get fleshed out, some are "oh, that's something in there?" (Restaurant reservations? Ride Booking?) and feels more like the half baked mysql interfaces in php.
However, as part of privacy - you can create a note (and dictate it) without a data connection with Siri. Your "start workout" command doesn't leave your device.
Part of that is privacy. Part of that is that Apple was trying to minimize its cloud spend (on GCP or AWS) by keeping as much of that activity on device. It wasn't entirely on device, but a lot more of it is than what Android is... and Alexa is a speaker and microphone hooked up to AWS.
This was ok, kind of meh, but ok pre-ChatGPT. With ChatGPT the expectations changed and the architecture that Apple had was not something that could pivot to meeting those expectations.
https://en.wikipedia.org/wiki/Apple_Intelligence
> Apple first implemented artificial intelligence features in its products with the release of Siri in the iPhone 4S in 2011.
> ...
> The rapid development of generative artificial intelligence and the release of ChatGPT in late 2022 reportedly blindsided Apple executives and forced the company to refocus its efforts on AI.
ChatGPT was as much a blindside to Apple as the iPhone was to Blackberry.
Are you referring to https://security.apple.com/com/blog/private-cloud-compute/?
The only way that AI will ever be able to replace each of us, is if it gathers our entire audio, text, etc history. PCC seemed like the only viable option for a pro-AI, yet pro-privacy person such as myself. I thought PCC was one of the most thoughtful things I had every seen a FAANG create. Seriously, whoever pushed that should get some kind of medal.
Are you saying that there is no technical solution for privacy and AI to coexist? Not only that, but that was the blocker?
I am genuinely interested if anyone can provide a technical answer.
They should actually make something useful first, and then work backwards to making it private before releasing it.
In terms of actually building a profitable business, no one seems to know how to make that work with AI right now. Leaks suggest OpenAI may turn to ads to monetize ChatGPT… which will raise all sorts of data questions. Privacy concerns might yet be an issue with AI chatbots.
If you've never been to China, you need to look no further than the streets to understand this (cameras everywhere, social credit system, etc.)
Not the first to bring mp3 players to the market, nor phones, nor tablets. Market leader every time.
They could have just stayed in a corner talking about privacy, offer a solid experience while everything else drowns in slop, researched UX for llms and come 5 years later with a killer product.
I don't get why they went for the rush. It's not like AI is killing their hardware sales either.
For one thing, the iPad (market-leading tablet) and the iPhone (market-leading pocket touchscreen device) were not their first attempt at doing that. That would be the Newton, which was an actual launched product and a commercial failure.
For another thing, even Apple can't just become the market leader by doing nothing. They need to enter late with a good product, and having a good product takes R&D, which takes time. With MP3 players, smartphones, and tablets, they didn't exactly wait until the industry was burnt through before they came in with their offering; they were just later (with the successful product) than some other people who did it worse. They were still doing R&D during those years when they were "waiting."
Apple could still "show up late" to AI in a few more years or a decade, using their current toe-dipping to inform something better, and it would still fit into the picture you have of how they "should've done it." Not to mention, Apple's also lost its way before with things like convoluted product lines (a dozen models of everything) and experimental products (the Newton then, Apple Vision now); messing up for a while also isn't exactly bucking history.
They already lost this superpower in the EU and I think Japan, India, Brazil too. Early next year they've got their US antitrust trial, and later in the year are some class actions challenging their control over app distribution, and at least two pieces of draft legislation are circulating that would require allowing competing apps to be defaults.
If they need another two years they might face an entrenched and perhaps even better competitor, while their own app needs to be downloaded from the App Store.
They were not technically first at creating a portable music player that relied on digital compressed format and digital storage, that's true. However before the iPod, you had either very low storage capacity, reliance on expensive memory cards that you would need to swap or hard drive based players that were unpractical to carry around.
Those solutions also had generally terrible user interface that made things like browsing the library or seeking in a track a pain in the ass. It was a moderate improvement on already existing solutions. It wasn't that much better than a MP3 CD player (especially after they figured out buffering to avoid skipping because of movement). I had a mini-disc player in parallel to the MP3s and it was generally a better solution: cheaper for managing a big library and it had compression. Physical media management was still a problem but that was also a requirement for MP3 player of practical size that used memory cards (and those were awfully expensive).
Then Apple came in with a solution using the newer 1,8 inch hard drive format with a starting capacity of 5GB, which would be bumped to 10GB shortly. It changed everything, it was practical in portability/pocketability, storage capacity and user interface. My brother bought an Archos player released around the same time and it was just a joke compared to the iPod (he got it because it was on sale for much cheaper).
So yeah, technically Apple was not the first to commercialize an MP3 player. But they trailed the earliest entrant in the market only by a few years (about 3) and they basically defined the market. Before the iPod, MP3 players were mostly a waste of money and a curiosity at best. The experience they would offer for the price was ridiculous. The iPod was expensive but it was a major improvement in every way, it made the Portable Media Player not only a possibility but something very desirable because it was useful and competent. The iPhone is basically the same story and before that the Mac (for DTP) and even before that the Apple II (for general computing access to individuals).
I feel that when people argue about Apple being late on innovation they are arguing in bad faith to justify current mismanagement (because they make a shit ton of money regardless). But it's basically splitting hairs. Yes Apple was never strictly first on purely technical grounds but they were always the first at creating a consumer product that would actually be able to achieve the purpose it was supposed to in a satisfactory manner.
I see Apple dusting off its OG playbook.
We're in the minicomputing era of AI. If scaling continues to bear fruit, we'll stay there for some time. Potentially indefinitely. If, however, scaling plateaus, miniaturisation retakes precedence. At that point, Apple's hardware (and Google's mindshare) incumbency gains precedence.
In the meantime, Apple builds devices and writes the OS that commands how the richest consumers on Earth store and transmit their data. That gives them a default seat at every AI table, whether they bother to show up or not.
What does seem slightly odd is Apple have probably saved billions by failing to be dragged into the current model war.
Apple also doesn't have actual privacy since their focus was using the word strategically against their competitors, not actually protecting user data.
> Subramanya, who Apple describes as a "renowned AI researcher," spent 16 years at Google, where he was head of engineering for Gemini. He left Google earlier this year for Microsoft. In a press release, Apple said that Subramanya will report to Craig Federighi and will "be leading critical areas, including Apple Foundation Models, ML research, and AI Safety and Evaluation."
I don't see how Google + Copilot mindset even touches on privacy. I wouldn't be surprised if we users will be forced to pay even more personal data in the near future.
big company bad, or do you have examples?
But Android has been 100% accurate for simple commands for over a decade for me. Things like:
- "weather." tells me forecast of where I am.
- "alarm at 8am" and "alarm in 30 minutes" works as expected.
- calendar commands also work
My favourite is "go home" which opens Google map with a route to home.
These things just work. I don't recall last time I had to repeat myself.
Failing to find any way to get the alarm thing back, I turned off the entire assistant thing.
EG I can talk to it like I would chatgpt and it works well. But I can't be like "hey I want to get dinner with my wife on our anniversary, please book the best available option in my city for fine dining"
It's still way better than Siri, which feels like a voice CLI to me (same as Alexa, which is very low quality IME)
Edit: why in gods name are people downvoting me for politely asking about someone’s differing experience?
"Here is what I found about "The Dragonborn comes at 25" on the Internet" opens Safari
Not true for me at all, it fails at the most basic tasks, sometimes even at tasks it has done before. Three examples:
- "Timer 5 minutes" -> Loading spinner is shown. Siri disappears after a few seconds. No error, no confirmation. I then have to manually check if the timer was set or not (it was not).
- "Turn on the lights in the living room" to which it responds "Sorry, I cannot do that". I have Phillips Hue lights that are connected to Apple Home, of course Siri can do that. It did that before.
- "Add tooth paste to my shopping list". The shopping list is a list I have in reminders. It then tries to search for the query on Google. I then tried "Add tooth paste to the list shopping list in reminders" which worked, but if I have to be this wordy, it is no longer any convenient.
There are many more simple cases in which Siri always / sometimes fails. I also have the feeling that it performs far worse if asked in my native language (German) than in English.
Home Assistant can even share devices with Home app, so you can still use "Siri, turn off the lamp" to have it answer "you don't have any alarms set".
Even basic tasks like “play this song” get screwed up and wonder off into the abyss. Absolute pile of garbage from Amazon on this AI stuff.
If you tell Siri to play some obscure artist or title of which there seem to be about 10 possible hits, then sure, we need confirmation. If I tell you to play Riders on the Storm by the The Doors, just play it damn it.
If it doesn't know where you are then you might live in a Faraday cage.
I have had the same issues myself. The "allow Siri when locked" toggle for my phone is set to on, yet it still often refuses to do stuff like turning my home lights on/off, giving the reason that I need to unlock my phone.
And I've also had the location issue they gave an example of, where asking "what is the weather in Berlin" does not need it to know where you are because you've told it the location you're asking about, so the answer is the same regardless of whether you're asking while in Berlin or Tokyo, yet it still sometimes gets stuck believing it needs location access.
Edit: oops the issue of having said a location in the request wasn't in the comment you replied to, but another reply to it. But the overall point of Siri being annoyingly inconsistent still stands.
Casuals are in there--nontechnical folks for whom "brick breaker deluxe wants to access your contacts" might raise an eyebrow. The stalked are in there--malicious apps that track location of the install-ee are unfortunately not uncommon. The one-device-for-multiple-lives folks are in there (if your work email/contacts are on your phone, it's a good thing that your dating app has to ask permission before acquiring your phone's contacts). So are the forgetful--that periodic "hey, this app has had permissions for ages, do you still want it to have that access?" thing not only helps folks clean up their perms, it reminds lots of folks about services they forgot they paid (or worse, forgot they are still paying) for.
are there solutions to the error rates when picking from dozens or even hundreds of tools i'm not aware of?
I worked on systems for evaluating the quality of models over time and for evaluating the quality of new models before release to understand how the new models would perform compared to current models once in the wild. It was difficult to get Siri to use these tools that were outside of their org. While this wouldn't solve the breadth of Siri's functionality issues, it would have helped improve the overall user experience with the existing Siri features to avoid the seemingly reduction of quality over time.
Secondly, and admittedly farther from where I was... Apple could have started the move from ML models to LLMs much sooner. The underlying technology for LLMs started gaining popularity in papers and research quite a few years ago, and there was a real problem of each team developing their own ML models for search, similarity, recommendations, etc that were quite large and that became a problem for mobile device delivery and storage. If leadership had a way to bring the orgs together they may have landed on LLMs much sooner.
Remember they even had an ad campaign “what happens on your iPhone stays on your iPhone” which is technically incorrect and “smarter Siri” can’t exist around such promises atm.
See also:
https://news.ycombinator.com/item?id=42014588
https://news.ycombinator.com/item?id=41712728
https://appleinsider.com/articles/24/04/10/apple-makes-it-re...
https://sneak.berlin/20231005/apple-operating-system-surveil...
My experience is only through android auto and it honestly makes me furious how bad it is. There is absolutely no other tech product in my life that gets even close to how bad voice commands are handled in Android.
In my experience, literally everything sucks:
- single language voice recognition (me speaking in English with an accent)
- multi language voice recognition (english commands that include localised names from the country I'm in)
- action in context (understand what I'm actually asking it to do)
- supported actions (what it can actually do)
Some practical examples from just this week:
- I had to repeat 3 times that "no I don't want to reply" because I made the mistake of getting google to read a whatsapp message while driving, and it got stuck into the "would you like to reply" (it almost always gets stuck - it's my goto example to show people how bad it is)
- I asked it to queue a very specific playlist on Spotify, and it just couldn't get it right (no matter how specific my command was, I couldn't get it to play a playlist from MY. account instead of playing an unrelated public playlist)
- I asked to add a song to a playlist, and it said it couldn't do that (at least it understood what I was asking? maybe)
And in general I gave up trying to use google maps through voice commands, because it's just not capable of understanding an English command if it contains a street/location name pronounced in the local language/accent.
"Turn on living room lights" - "There are no lights in the living room"
"Turn on living room lights" - lights turn on.
Naming some of my lights Greg has helped a bit.
I pretty much only use it when I can’t look at the phone so I’m not sure if it’s still there.
I was not able to edit / update it! However, there was now a new "maps" option for `Open <type> directions from <Start> to <Destination>`
Where type can now be {driving,walking,biking,transit} and <start> is Current Location by default.
After updating, this now seems to correctly set actual driving directions, even if I'd previously set up a walking route!
This could be personalized, 'does this user do this kind of thing?' which checks history of user actions for anything similar. Or it could be generic, 'is this the type of thing a typical user does?'
In both cases, if it's unfamiliar you have a few options: try to interpret it again (maybe with a better model), raise a prompt with the user ('do you want to do x?'), or if it's highly unfamiliar, auto cancel the command and say sorry.
(I do often get frustrated with dictation quirks that don't have anything to do with my accent, like it choosing the obviously wrong option when their are multiple words that sound the same, especially its insistence on assuming I'm saying the name of a contact rather than the common noun that sounds the same.)
Not to mention the iOS keyboard has gotten so bad in the last year that it took me 3x longer to type this comment (I use the swipe keyboard). I had to fix at least a dozen typos.
Every now and then when they screw up, they’ll have a mea culpa with the press. They haven’t done that with Siri or the keyboard yet.
I continue to be impressed by ChromeOS. With the Linux development environment (Debian VM), it is a brilliant work environment.
Add Android apps as well and ChromeOS is an awesome convergence platform. There are Chromebooks that are high enough quality that I don't miss anything about Apple OS or Microsoft Windows.
> And for audio production
For specialty use-cases, driver support will favour Windows and Apple OS.
And gaming is still Windows-first, although Linux is improving.
The MacBook Air is nice but it still too expensive for most people that only want to browser the web, use social media and edit a few documents once in a while. Their advantage in chip power and efficiency is not enough because people buying this won't make use of the power anyway and will rarely use it long enough at once for battery life to matter that much.
Technically a lot of stuff for linux audio is amazing; it just lacks the level of polish to enable someone to actually just use it out of the box.
I don’t know why that is from a technical level.
I literally just experienced this with RCS failing to activate. No failure message, dug into logs, says userinteractionrequired. Front page of HN, nobody knows, apple corp response, 'thats interesting no you cant talk to engineering'.
Read the RCS spec definition document to fall asleep to after the board swap and the call saying they won't work on it since issue resolved, answers exactly what that meant, Apple never implemented handling for it, my followup post: https://wt.gd/working-rcs-messaging
The list didn’t show the god damn GoPro app, which was taking up 20GB of space from downloaded videos. I guessed it was the problem because it showed up in the device storage list, but literally not reported when you look at the list of data to backup.
iMessage is another great example of a failure. I changed my iMessage email and didn’t receive messages in a family group chat until I noticed — I had to text the chat before stuff started coning through. Previously sent messages were never delivered. And they all have my phone number, which has been my primary iMessage for LITERALLY over a decade. iMessage’s identity system is seriously messed up on a fundamental level. (I’ve had numerous other issues with it, but I’ll digress.)
Knowing that a company had competent product designers that made a good product, but then shitcanned the working product for a bunch of amateur output from people that don't understand dry very basics of UI, from the one company that made good UI its primary feature for decades... well it just felt like full on betrayal. The same thing happened with absolutely shitty Apple Music, which I never, ever use, because it's so painful to remember what could have been with iTunes...
I remember advising many photographs friends on using Aperture for photo library management. Now I feel so bad for ever recommending that. I mean Lightroom now has a stupid subscription, but using Apple software was kind of the point: avoiding the risk of software becoming too expensive or bad because the hardware premium funds the development of good software.
Now you get to pay more for the hardware but you have to deal with shitty or expensive software as well. Makes no sense.
I don't know why it bugs me so much, but I'm at the point of moving my library into a self-hosted Navidrome instance so I can stop using Music.
Will it sync? When? Who knows? You’re on WiFi with a full battery and charging? So? Might be a minute, might be an hour. Oh, you restarted Photos? Who cares? Not Photos.
Not to mention the likely need for continuous internet connectivity and service upkeep. Car companies aren't exactly known for good software governance.
I think well-done voice commands are a great addition to a car, especially for rentals. When figuring out how to do something in a new car, I have to choose between safety, interruption (stopping briefly) or not having my desires function change.
Most basic functions can be voice-controlled without Internet connectivity. You should only need that for conversational topics, not for controlling car functions.
I don't own a car but rent them occasionally on vacation in every one I've rented that I can remember since they started having the big touch screens that connect with your phone, the voice button on the steering wheel would just launch Siri (on CarPlay), which seems optimal—just have the phone software deal with it because the car companies are bad at software.
It seems to work fine for changing music when there's no passenger to do that, subject to only the usual limitations with Siri sucking—but I don't expect a car company to do better, and honestly the worst case I've can remember with music is that played the title track of an album rather than the album, which is admittedly ambiguous. Now I just say explicitly "play the album 'foo' by 'bar' on Spotify" and it works. It's definitely a lot safer than fumbling around with the touchscreen (and Spotify's CarPlay app is very limited for browsing anyways, for safety I assume but then my partner can't browse music either, which would be fine) or trying to juggle CDs back in the day.
Never heard of Tellme, but it sounds impressive on a resume.
Metaweb was a good open-source fact database which subsequently got walled off once Google bought it.
Google Search works significantly worse now than it did under Amit, and I say that as both a user and a websearch Xoogler. (JG took over about a year after I left Google).
Siri is the subject of this article.
The various interviews I've watched of his (and some of the leaked news) shows he's still quite deep in the tools.
Most of their current products seem to be decaying in the dead march towards the next yearly release. ux and ui are becoming more and more poorly thought (see their last design language). They half pursue ideas and don’t manage to deliver (vr, apple car, etc).
I see cargo culting and fad chasing like any average leadership, only with a fatter stack of cash supporting the endeavour.
1. Apple is big enough that it needs to take care of edge cases like offline & limited cell reception, which affect millions in any given moment.
2. Launching a major UI feature (Siri) that people will come to rely on requires offline operation for common operations like basic device operations and dictation. Major UI features shouldn't cease to function when they enter bad reception zones.
3. Apple builds devices with great CPUs, which allows them to pursue a strategy of using edge compute to reduce spend.
4. A consequence of building products with good offline support is they are more private.
5. Apple didn't even build a full set of intents for most of their apps, hence 'remind me at this location' doesn't even work. App developers haven't either, because ...
6. Siri (both the local version and remote service) isn't very good, and regularly misunderstands or fails at basic comprehension tasks that do not even require user data to be understood or relayed back to devices to execute.
I don't buy that privacy is somehow an impediment to #5 or #6. It's only an issue when user data is involved, and Apple has been investing in techs like differential privacy to get around these limitations to some extent. But that is further downstream from #5 and #6 though.
I don't care if I have to carefully say "bibbidy bobbity boo, set an alarm for two" - I just need it to be reliable.
If Apple takes the position that the UX has to fit in around the privacy requirements, so what? Privacy is a core pillar of their product identity—a built-in hallucinating compliments machine isn't.
Edit: to be clear, Siri doesn’t. Still no reason it shouldn’t be able to.
I can think of one time recently where no matter how I prompted it to play an album for (decades old but probably triple platinum,) it kept playing some cardi b song with the band’s name in the title instead… but that’s probably like a 1 in 2000 request problem. Maybe its a genre thing?
How did you learn this hack?
I just noticed it randomly many years ago, I don't remember the occasion but I guess I was scrolling trying to find a page in history lazily and noticed it at the bottom.
It's an example that sums up feature discoverability (well, lack of) on iPhones - there are so many things like this, that are really useful to know if you find out about them but the only way to find out is luck or having a friend tell you. Occasionally the official Apple "Tips" app has useful stuff, but not much.
I actually have a thing in my family Signal chat of every few weeks sharing a new random iPhone tip, as I'm by far the nerdiest in the group. Maybe I should collate them all into a "hard to discover Tips" blog and share on HN...
Both for explaining the simpler way to do it and also for the contect.
(Why? Because it feels good to know that I am not the only one who cannot find documentation for this and other things.)
The one caveat is that once the timer you want is two and a half hours or longer, then Siri replies by asking what you would like to convert to.
As for, "What time is it?"... Try activating Siri and only saying, "Time."
And furthermore - aren't there shells that will give you the --help if you try to tab-complete certain commands? Obviously there's the issue of a lack of standardization for how command-line switches work, but broadly speaking it's not difficult to have a list of common (or even uncommon) commands and how their args work.
(spends a few minutes researching...)
This project evidently exists, and I think it's even fairly well supported in e.g. Debian-based systems: https://github.com/scop/bash-completion.
"Hunt the verb" means that the user doesn't know which commands (verbs) exist. Which a neophyte at a blank console will not. This absolutely is a problem with CLIs.
Many users like myself enjoy a good manual and will lean into a CLI at every opportunity. This is absolutely counter to the value proposition of a natural language assistant.
Every time I try to interact with one of these llm gatekeepers I just say what I want and hope it figures out to send me to a person. The rest of the time I’m trying to convince the Taco Bell to record a customer complaint about how its existence itself is dystopian.
For older tools, sure. Newer tools eschew man pages and just offer some help flag, even though there are excellent libraries that generate manpages like https://crates.io/crates/clap_mangen or https://crates.io/crates/mandown (for Rust, but I am sure most languages have one) without requiring you to learn troff.
An AI wrapper typically has few actual capabilities, concealed behind a skeuomorphic “fake person” UX. It may have a private list of capabilities but it otherwise doesn’t know if it knows something or not and will just say stuff.
It really needs to be 100% before it’s useful and not just frustrating.
> but there is no invitation to guess, and no one pretends you don’t need the manual
which is basically what you're saying too? the problem with voice UIs and some LLM tools is that it's unclear which options and tools exist and there's no documentation of it.
- almost every search field (when an end user modifies the search instead of clicking one of the results for the second time that should be a clear signal that something is off.)
- almost every chat bot (Don't even get me started about the Fisher Price toy level of interactions provided by most of them. And worse: I know they can be great, one company I interact with now has a great one and a previous company I worked for had another great one. It just seems people throw chatbots at the page like it is a checkbox that needs to be checked.)
- almost all server logs (what pages are people linking to that now return 404?)
- referer headers (you product is being discussed in an open forum and no one cares to even read it?)
We collect so much data and then we don't use it for anything that could actually delight our users. Either it is thrown away or worse it is fed back into "targeted" advertising that besides being an ugly idea also seems to be a stupid idea in many cases: years go by betweeen each time I see a "targeted" ad that actually makes me want to buy something, much less actually buy something.
Very charitable, but rarely true.
If you ask Siri about playing some music, it will go the dumb route of finding the tracks that seems to be a close linguistic match of what you said (if it correctly understood you in the first place) when in fact you may have meant another track of the same name. Which means you always need to overspecify with lots of details (like the artist and album) and that defeat the purpose of having an "assistant".
Another example would be asking it to call your father, which it will fail to do so unless you have correctly filled the contact card with a relation field linked to you. So you need to fill in all the details about everyone (and remember what name/details you used), otherwise you are stuck just relying on rigid naming like a phone book. Moderately useful and since it require upfront work the payoff potential isn't very good. If Siri would be able to figure out who's who just from the communications happening on your device, it could be better, but Apple has dug itself into a hole with their privacy marketing.
The whole point of an (human) assistant is that it knows you, your behaviors, how you think, what you like. So he/she can help you with less effort on your part because you don't have to overspecify every details that would be obvious to you and anyone who knows you well enough. Siri is hopeless because it doesn't really know you, it only use some very simple heuristic to try to be useful. One example is how it always offer to give me the route home when I turn on the car, even when I'm only running errands and the next stop is just another shop. It is not only unhelpful but annoying because giving me the route home when I'm only a few kilometers away is not particularly useful in the first place.
And thinking back over it, more than half the failures are complete - e.g., it likely never activated at all. Very few are "it set a timer, but for the wrong time".
But the point is none of that is intrinsic or interesting to the underlying idea, it’s just of annoying practical relevance to interfacing with APIs today