Tim Cook confirms that Apple has been working on generative AI for years

AppleInsider · August 3, 2023 9:05PM

Apple spent $22.61 billion on research and development through 2023, and CEO Tim Cook confirmed beyond a shadow of a doubt that part of that expenditure is because the company is working on generative AI.

Apple AI research is underway

Many pundits have said Apple is behind in AI, specifically because they do not have a chatbot or answer to tools like Google Bard. However, it seems Apple AI has been being developed for years within the company.

According to a report from Reuters, Apple's growing spend in research and development can be tied to developing AI technology. The publication received a statement from Apple CEO Tim Cook on the matter.

"We've been doing research across a wide range of AI technologies, including generative AI, for years. We're going to continue investing and innovating and responsibly advancing our products with these technologies to help enrich people's lives," Cook said. "Obviously, we're investing a lot, and it is showing up in the R&D spending that you're looking at."

Apple spent $22.61 billion across the fiscal year 2023 on R&D so far, which is reportedly about $3.12 billion higher than this point in the previous year.

Apple is rumored to be working on a ChatGPT-like product internally, though it may never see the light of day. Mark Gurman suggests some kind of Apple AI app could arrive in 2024, though Ming-Chi Kuo believes Apple is further behind than that.

Apple R&D spending over time

Despite these claims and rumors, Apple has already pushed out products running on advanced machine learning models like the iPhone's camera, Siri, and more. An update in iOS 17 brings a transformer language model to the autocorrect system, which is the same base technology used in AI chatbots.

While Apple may not announce a SiriGPT anytime soon, if ever, expect more advancements from the company in the area across its platforms. Many expect the developer tool Xcode could be Apple's next target for adding an AI tool.

Read on AppleInsider

twolf2919 · August 3, 2023 9:28PM

Apple has been bringing out some amazing AI-based solutions in the last couple years, including object recognition in the Photo app. But Siri advances continue to be pathetic. If Siri is an example of Apple's AI prowess, it's definitely behind on the AI curve. I'm a big Apple fan - have all their devices - and was very enthusiastic when Siri first came out. But over the years, Siri has moved inches, while competitors have moved miles - I can't believe a company that spends $22b on R&D can't make its voice assistant more useful in people's day-to-day lives. So disappointing.

chasm · August 3, 2023 9:40PM

twolf2919 said:

I'm a big Apple fan - have all their devices - and was very enthusiastic when Siri first came out. But over the years, Siri has moved inches, while competitors have moved miles - I can't believe a company that spends $22b on R&D can't make its voice assistant more useful in people's day-to-day lives. So disappointing.

Siri will NEVER EVER be as “smart” as other voice assistants, so you’d better get over this right now. The reason? Because Siri doesn’t heavily rely on the collected data about you that encompasses more than even your parents know about you.

Siri continues to work very well for me, because I know what it can do and I don’t have much need outside of it helping me organise my day/life. I am fully aware that if you ask it some random question, probably phrased poorly, it will punt to web results. This is because it doesn’t have access to everything you have ever searched for or asked about ever in your life to help piece together what you actually want.

Maybe my needs are unusually simple, but I appreciate that Siri isn’t spying on me to get “smarter,” because it currently does 95+ percent of what I want it to do. I think the main area where Siri actually needs work is on accent recognition, and I could certainly see where Siri may in the future ask you to train it to your voice with more phrases than it does now.

It is much harder to develop a system that uses “AI” (in quotes on purpose) without scraping and selling all your personal data, which is why progress on Siri has been very slow (though it has in fact improved a great deal in recent years, at least in my use of it) compared to the companies that raid your brain for your every thought. But this is like praising the child who answers the algebra question by looking over another student’s shoulder, and failing the child who did all the work themselves because it took too long IMO.

lolliver · August 3, 2023 10:18PM

chasm said:

twolf2919 said:

I'm a big Apple fan - have all their devices - and was very enthusiastic when Siri first came out. But over the years, Siri has moved inches, while competitors have moved miles - I can't believe a company that spends $22b on R&D can't make its voice assistant more useful in people's day-to-day lives. So disappointing.

Siri will NEVER EVER be as “smart” as other voice assistants, so you’d better get over this right now. The reason? Because Siri doesn’t heavily rely on the collected data about you that encompasses more than even your parents know about you.

But this is like praising the child who answers the algebra question by looking over another student’s shoulder, and failing the child who did all the work themselves because it took too long IMO.

Exactly this! Siri does what it is meant to do and does it pretty well, without spying on us. Understanding accents does need some improvement and I’m sure there are other areas it could and will improve.

After spending a week at a friends house that had several Alexa devices though I am not sold on the idea that Alexa is good at answering broader knowledge questions. While it will answer questions rather than stating “here’s something I found on the web”. The answers were not always accurate and way too often it wasn’t even specific to the question asked. Alexa would regularly misunderstand the question but still give an answer anyway. And often that answer was inaccurate for the question it thought was asked.

My friend would ask Alexa a question, tell her to stop part way through as it got the question wrong and then they may have to repeat this process several times before finally getting Alexa to correctly understand the question (but not always get the answer correct).

I started pulling out my phone and searching the web every time my friend started speaking to Alexa and was usually able to get the answer much sooner and from more trustworthy sources.

GerfnitAuthor · August 3, 2023 10:45PM

My primary problem with Siri is it doesn't remember context. So if I ask about where someone lives or the address of a company and it gives me the answer, I can't request, "Take me there" because it doesn't know what "there" means. I'm afraid I've gotten spoiled by GPTchat engines with which I can make backward references.

coolfactor · August 4, 2023 1:22AM

GerfnitAuthor said:

My primary problem with Siri is it doesn't remember context. So if I ask about where someone lives or the address of a company and it gives me the answer, I can't request, "Take me there" because it doesn't know what "there" means. I'm afraid I've gotten spoiled by GPTchat engines with which I can make backward references.

I believe the next version of Siri — coming with iOS 17 — can chain queries together. Combined with saying just "Siri", instead of "Hey Siri", the experience of interacting with the assistant should be much more natural and enjoyable.

Has this improved Siri been available in the iOS 17 betas, or do the betas still use the same Siri as the rest of us?

noelos · August 4, 2023 1:52AM

I'm sorry - but the excuses about Siri being crap because of privacy concerns just doesn't hold water.

"Hey Siri - play Scarborough Fair". Now I have "Scarborough Fair" by Simon & Garfunkel in my library, and I'd guess this is the version that 99.9% of people would want 99.9% of the time and it's a track I've also asked Siri to play literally hundreds of times because it helps my toddler get to sleep.

And yet, every time it is a gamble if she will play this, or some other rando version, or some other rando song completely. Siri learns NOTHING and doesn't even use the contextual knowledge available to make a reasonable guess of what you want.

Or what about a simple query while I'm driving and I idly ask what year some actor died. "I can't answer that while you're driving". Is that to preserve my privacy or is it just because they haven't put in the work to make Siri be able to answer simple queries in natural language?

chasm · August 4, 2023 2:55AM

noelos said:

I'm sorry - but the excuses about Siri being crap because of privacy concerns just doesn't hold water.

"Hey Siri - play Scarborough Fair". Now I have "Scarborough Fair" by Simon & Garfunkel in my library, and I'd guess this is the version that 99.9% of people would want 99.9% of the time and it's a track I've also asked Siri to play literally hundreds of times because it helps my toddler get to sleep.

And yet, every time it is a gamble if she will play this, or some other rando version, or some other rando song completely. Siri learns NOTHING and doesn't even use the contextual knowledge available to make a reasonable guess of what you want.

Or what about a simple query while I'm driving and I idly ask what year some actor died. "I can't answer that while you're driving". Is that to preserve my privacy or is it just because they haven't put in the work to make Siri be able to answer simple queries in natural language?

Okay, let’s go over this:

As I mentioned, if you issue requests poorly, you can get bad results. Your S&G request is a good example of this. If you want a specific artist’s version of a song, you need to include the artists’ name in the request. A thousand acts (at least) have covered this song, so saying “Hey Siri, play Scarborough Fair by Simon and Garfunkel” will probably get you the right version, though I admit there’s a possibility you might get a live version by them — but if you add the phrase “from my library” I think you’d hit 100 percent accuracy.

As for your driving and asking questions query, I’ll remind you that Siri is not a chat bot. It is MUCH more restricted by Apple while you’re driving so that you don’t get distracted, and this is by design. Apple does not want you asking Siri random questions while you are driving, because that indicates that you are distracted or want to be distracted.

You may find that too restrictive, and I think you have something of a case, but when it comes to driver distraction I’d rather Apple err on the side of being over cautious and thus more limited. When I’m NOT in motion, suddenly Siri can answer a question like that handily. I just asked it “what year did actor John Wayne die,” when I was at home, and it gave me a very full answer, including his age and location when he died as well as the date.

I am certainly NOT saying there isn’t any room for improvement in Siri; quite the opposite! I’m just explaining why it behaves in some of the ways it behaves.

CheeseFreeze · August 4, 2023 3:23AM

That’s damage control by Tim Cook. Super vague and nothing to show for.

CheeseFreeze · August 4, 2023 3:25AM

coolfactor said:

GerfnitAuthor said:

My primary problem with Siri is it doesn't remember context. So if I ask about where someone lives or the address of a company and it gives me the answer, I can't request, "Take me there" because it doesn't know what "there" means. I'm afraid I've gotten spoiled by GPTchat engines with which I can make backward references.

I believe the next version of Siri — coming with iOS 17 — can chain queries together. Combined with saying just "Siri", instead of "Hey Siri", the experience of interacting with the assistant should be much more natural and enjoyable.

Has this improved Siri been available in the iOS 17 betas, or do the betas still use the same Siri as the rest of us?

That’s just a bandaid solution. Siri is a dead-end in the world of LLM AI.

exceptionhandler · August 4, 2023 3:33AM

GerfnitAuthor said:

My primary problem with Siri is it doesn't remember context. So if I ask about where someone lives or the address of a company and it gives me the answer, I can't request, "Take me there" because it doesn't know what "there" means. I'm afraid I've gotten spoiled by GPTchat engines with which I can make backward references.

I thought I remembered Siri was supposed to remember context, and had used that feature on occasion. I think over time Siri has changed or at least I don’t use the scripted version in the 2011 keynote when Siri was introduced:

https://m.youtube.com/watch?v=agzItTz35QQ

I remembered Siri being a lot better then than now, but still prefer it over others, or not having one at all.

FileMakerFeller · August 4, 2023 4:30AM

chasm said:

noelos said:

I'm sorry - but the excuses about Siri being crap because of privacy concerns just doesn't hold water.

"Hey Siri - play Scarborough Fair". Now I have "Scarborough Fair" by Simon & Garfunkel in my library, and I'd guess this is the version that 99.9% of people would want 99.9% of the time and it's a track I've also asked Siri to play literally hundreds of times because it helps my toddler get to sleep.

And yet, every time it is a gamble if she will play this, or some other rando version, or some other rando song completely. Siri learns NOTHING and doesn't even use the contextual knowledge available to make a reasonable guess of what you want.

Or what about a simple query while I'm driving and I idly ask what year some actor died. "I can't answer that while you're driving". Is that to preserve my privacy or is it just because they haven't put in the work to make Siri be able to answer simple queries in natural language?

Okay, let’s go over this:

As I mentioned, if you issue requests poorly, you can get bad results. Your S&G request is a good example of this. If you want a specific artist’s version of a song, you need to include the artists’ name in the request. A thousand acts (at least) have covered this song, so saying “Hey Siri, play Scarborough Fair by Simon and Garfunkel” will probably get you the right version, though I admit there’s a possibility you might get a live version by them — but if you add the phrase “from my library” I think you’d hit 100 percent accuracy.

As for your driving and asking questions query, I’ll remind you that Siri is not a chat bot. It is MUCH more restricted by Apple while you’re driving so that you don’t get distracted, and this is by design. Apple does not want you asking Siri random questions while you are driving, because that indicates that you are distracted or want to be distracted.

You may find that too restrictive, and I think you have something of a case, but when it comes to driver distraction I’d rather Apple err on the side of being over cautious and thus more limited. When I’m NOT in motion, suddenly Siri can answer a question like that handily. I just asked it “what year did actor John Wayne die,” when I was at home, and it gave me a very full answer, including his age and location when he died as well as the date.

I am certainly NOT saying there isn’t any room for improvement in Siri; quite the opposite! I’m just explaining why it behaves in some of the ways it behaves.

This is a situation where the good intentions result in a worse outcome. What's more distracting, listening to an informative answer or being told you can't have the answer? My frustration level certainly grows with the latter.

I would further argue that the decision about whether or not I have the capacity to make the query and accept the response is a matter for me the human and not the decision trees implemented by the software. The motion sensors can certainly tell if I'm gridlocked in traffic; in my opinion that should be a factor in the algorithm. Plus, Siri has no problem reading me the latest text message from a person - which is a similar level of cognitive effort to absorb.

Even if I'm wrong in my perceptions of my ability in a given situation (and, let's face it, that's a common mistake we humans make), a better response would be "It's not appropriate for me to answer that while you're driving. Would you like me to: 1. Remind you of this query once you stop driving, 2. Tell you the answer anyway, or 3. Put the answer in a note that you can review later?" - since people generally only remember the first and last thing they hear from you, burying the "problematic" option in the middle of the list is a subtle "force" to the "better" options.

humbug1873 · August 4, 2023 5:45AM

People just don't get it. Apple is rarely a techno-feature company but intensely focus on user/useable features. Oh look we've got this new techno-gimmick you can ... use (for what exactly we're not sure) ... that's not Apple.
I would guess they're probably working on a Siri Improvement, that can hold a conversation (eventually having in part generative AI working in the background) .. and that you can sell to the average user (and that is what an average user would care about).
... and beyond that, they won't talk about it until it is done.

Or to put it in hype cycle terms Apple will enter during the rise phase, work through the peak phase (usually in quiet) and then suddenly show up with something useful during the through of disillusionment phase and prbably be the reason we'll eventually rise to the plateau of productivity, because something truely useful comes out of it ... or will silently let it die if nothing useful comes out of it.

Where are we with generative AI!? Probably at the heights of the Peak going to fall down soon. Because essentially right now its more of a toy.

forgot username · August 4, 2023 10:33AM

Apple should just license GPT-4. It’d be cheaper, faster and more powerful than Siri.

avon b7 · August 4, 2023 11:27AM

CheeseFreeze said:

That’s damage control by Tim Cook. Super vague and nothing to show for.

To be honest, it does sound like an effort to get this technology into the mindset of Apple users/watchers.

If they've been working on it for years and said nary a peep, why come out with such a vague claim now?

Someone at Apple obviously thinks it is a good idea to gain some mindset here and not have it sound like they have nothing comparable in the pipe.

Not that GPT is massively important right now. The real challenges are in the backend and industrial applications.

spheric · August 4, 2023 11:45AM

GerfnitAuthor said:

My primary problem with Siri is it doesn't remember context. So if I ask about where someone lives or the address of a company and it gives me the answer, I can't request, "Take me there" because it doesn't know what "there" means. I'm afraid I've gotten spoiled by GPTchat engines with which I can make backward references.

Oh hell yes.

100% real exchange (albeit in German), tried a dozen times:

"Hey Siri, call Hansi, mobile"

"I don't have a mobile number for Hansi. Do you want to call landline or iPhone?" (bonkers #1)

"Call iPhone."

"Okay! Whom do you want me to call?"

That eventually changed (after YEARS) and finally started working some time this spring. There's a bunch of things that make Siri completely and utterly USELESS in German, though — and that's definitely a language problem, because switching Siri to English for test purposes worked —, that I've pretty much entirely stopped using it except for reading messages to me and occasionally making a call.

I can't imagine wanting a HomePod that's built around (German-language) Siri.

AppleZulu · August 4, 2023 12:28PM

humbug1873 said:

People just don't get it. Apple is rarely a techno-feature company but intensely focus on user/useable features. Oh look we've got this new techno-gimmick you can ... use (for what exactly we're not sure) ... that's not Apple.
I would guess they're probably working on a Siri Improvement, that can hold a conversation (eventually having in part generative AI working in the background) .. and that you can sell to the average user (and that is what an average user would care about).
... and beyond that, they won't talk about it until it is done.

Or to put it in hype cycle terms Apple will enter during the rise phase, work through the peak phase (usually in quiet) and then suddenly show up with something useful during the through of disillusionment phase and prbably be the reason we'll eventually rise to the plateau of productivity, because something truely useful comes out of it ... or will silently let it die if nothing useful comes out of it.

Where are we with generative AI!? Probably at the heights of the Peak going to fall down soon. Because essentially right now it’s more of a toy.

Bingo. Even here on an Apple-centric site, people seem surprised when Apple isn’t first to the market with some new thing. Then they’re critical when Apple enters the market “late” with a thing. Finally, they forget all that when Apple’s version of said thing becomes the gold standard, because they came at it from a different direction that makes it actually useful, instead of just a novelty.

eriamjh · August 4, 2023 1:47PM

A proper assistant like Siri should remember everything I ever said, what day, what topic, etc.

Siri, how many calls did I get yesterday from non-contacts?

Siri, what’s the number of the call I got yesterday around 4-5pm? Please create a contact for it and call it Bob Smith from Scams are Us.

Privacy is fine. Encrypt everything. End to end.

avon b7 · August 4, 2023 2:52PM

spheric said:

GerfnitAuthor said:

My primary problem with Siri is it doesn't remember context. So if I ask about where someone lives or the address of a company and it gives me the answer, I can't request, "Take me there" because it doesn't know what "there" means. I'm afraid I've gotten spoiled by GPTchat engines with which I can make backward references.

Oh hell yes.

100% real exchange (albeit in German), tried a dozen times:

"Hey Siri, call Hansi, mobile"

"I don't have a mobile number for Hansi. Do you want to call landline or iPhone?" (bonkers #1)

"Call iPhone."

"Okay! Whom do you want me to call?"

That eventually changed (after YEARS) and finally started working some time this spring. There's a bunch of things that make Siri completely and utterly USELESS in German, though — and that's definitely a language problem, because switching Siri to English for test purposes worked —, that I've pretty much entirely stopped using it except for reading messages to me and occasionally making a call.

I can't imagine wanting a HomePod that's built around (German-language) Siri.

It's been a while since we've even bothered with Siri. We actually stopped when they modified the Spanish voices (Spanish Spanish) and they sounded far worse but the situation you describe was common.

One of our biggest problems was its non existent ability to pick out different languages in speech. In our case Spanish, Catalan and English are peppered throughout our interaction (names, places etc).

danox · August 4, 2023 3:16PM

AppleZulu said:

humbug1873 said:

People just don't get it. Apple is rarely a techno-feature company but intensely focus on user/useable features. Oh look we've got this new techno-gimmick you can ... use (for what exactly we're not sure) ... that's not Apple.
I would guess they're probably working on a Siri Improvement, that can hold a conversation (eventually having in part generative AI working in the background) .. and that you can sell to the average user (and that is what an average user would care about).
... and beyond that, they won't talk about it until it is done.

Or to put it in hype cycle terms Apple will enter during the rise phase, work through the peak phase (usually in quiet) and then suddenly show up with something useful during the through of disillusionment phase and prbably be the reason we'll eventually rise to the plateau of productivity, because something truely useful comes out of it ... or will silently let it die if nothing useful comes out of it.

Where are we with generative AI!? Probably at the heights of the Peak going to fall down soon. Because essentially right now it’s more of a toy.

Bingo. Even here on an Apple-centric site, people seem surprised when Apple isn’t first to the market with some new thing. Then they’re critical when Apple enters the market “late” with a thing. Finally, they forget all that when Apple’s version of said thing becomes the gold standard, because they came at it from a different direction that makes it actually useful, instead of just a novelty.

Apple’s path as a vertical computer company gets it right in a more practical manner, when compared to their competition, Apple simply over the years gets too much stuff, right over the long-haul, iMac, OS X, iPod, iPhone, iPad, Apple Maps, Metal, Face ID, Touch ID, Apple Watch, AirPods, Studio Display, Pro Display XDR, Apple Silicon, VisionOS and the Apple Vision Pro (which features the new R1 co-processor, nothing wrong with that pathway, the only criticism is not getting they’re all in one Mac’s back on track with the latest M series SOC’s being made available to the public to have at least a chance of buying one.

Alexa is on life-support, Cortana is dead, and Google, keeps begging Apple to support their undercover spy schemes, Apple however, is on the better more profitable, user-friendly, usable, constant, more practical pathway.

The lack of constancy among the primarily software only companies, Google, Microsoft, and Meta, make it is easy to dabble in the latest flash, fad tech buzzword of the year, this year, it appears to be AI that will save all. Next year, I think a more pedestrian down to earth co-processor (the R1) made by Apple will be the star.

https://arstechnica.com/gadgets/2023/03/with-amazon-alexas-future-in-peril-fire-tvs-offer-a-glimmer-of-hope/

https://www.tomsguide.com/news/its-official-microsoft-is-killing-cortana-in-favor-of-ai

avon b7 · August 4, 2023 3:49PM

danox said:

AppleZulu said:

humbug1873 said:

People just don't get it. Apple is rarely a techno-feature company but intensely focus on user/useable features. Oh look we've got this new techno-gimmick you can ... use (for what exactly we're not sure) ... that's not Apple.
I would guess they're probably working on a Siri Improvement, that can hold a conversation (eventually having in part generative AI working in the background) .. and that you can sell to the average user (and that is what an average user would care about).
... and beyond that, they won't talk about it until it is done.

Or to put it in hype cycle terms Apple will enter during the rise phase, work through the peak phase (usually in quiet) and then suddenly show up with something useful during the through of disillusionment phase and prbably be the reason we'll eventually rise to the plateau of productivity, because something truely useful comes out of it ... or will silently let it die if nothing useful comes out of it.

Where are we with generative AI!? Probably at the heights of the Peak going to fall down soon. Because essentially right now it’s more of a toy.

Bingo. Even here on an Apple-centric site, people seem surprised when Apple isn’t first to the market with some new thing. Then they’re critical when Apple enters the market “late” with a thing. Finally, they forget all that when Apple’s version of said thing becomes the gold standard, because they came at it from a different direction that makes it actually useful, instead of just a novelty.

Apple’s path as a vertical computer company gets it right in a more practical manner, when compared to their competition, Apple simply over the years gets too much stuff, right over the long-haul, iMac, OS X, iPod, iPhone, iPad, Apple Maps, Metal, Face ID, Touch ID, Apple Watch, AirPods, Studio Display, Pro Display XDR, Apple Silicon, VisionOS and the Apple Vision Pro (which features the new R1 co-processor, nothing wrong with that pathway, the only criticism is not getting they’re all in one Mac’s back on track with the latest M series SOC’s being made available to the public to have at least a chance of buying one.

Alexa is on life-support, Cortana is dead, and Google, keeps begging Apple to support their undercover spy schemes, Apple however, is on the better more profitable, user-friendly, usable, constant, more practical pathway.

The lack of constancy among the primarily software only companies, Google, Microsoft, and Meta, make it is easy to dabble in the latest flash, fad tech buzzword of the year, this year, it appears to be AI that will save all. Next year, I think a more pedestrian down to earth co-processor (the R1) made by Apple will be the star.

https://arstechnica.com/gadgets/2023/03/with-amazon-alexas-future-in-peril-fire-tvs-offer-a-glimmer-of-hope/

https://www.tomsguide.com/news/its-official-microsoft-is-killing-cortana-in-favor-of-ai

How is any of that not applicable to other manufacturers?

Apple Maps was a direct reaction to Google Maps and an utter disaster at launch. Google Maps remains the gold standard along with other maps options.

Apple maps will not get me to some places as it can't find them in the first place. On other occasions it will think it's found the place but its hundreds of kilometres away. It's got better over the years but still isn't up to Google Maps in many places.

The less said about Siri the better.

IM? It's Whatsapp all the way in most of the world.

Mail? I wonder how many Apple users don't have a Gmail account.

Home Automation? Why do you think Matter/Thread are on the lips of Apple users?

What happened to the notch? What replaced it? Something that has been around on other devices for years.

The TV? The car? The periscope lens? Where are they?

Nothing in your statement hasn't been provided by other vertically integrated manufacturers over the years.

Both in terms of hardware and software.

Today (just a couple of hours ago) NearLink was announced. Scarce on details at the moment but obviously a lot of work has gone into it.

https://sparrowsnews.com/2023/08/04/huawei-nearlink-technology-intro/amp/

Tim Cook confirms that Apple has been working on generative AI for years

Comments