Apple Intelligence, Apple Intelligence, Apple Intelligzzzzzzzzzzzzzzzz

[ShuggyCoUk]

Autonomous driving has to solves issues of evaluation of the state of the world around it from limited sensor data. yes so do humans but there are plenty of humans who are utterly (legally and physically) unable to drive cars who are still considered (non artificial) general intelligences...

If you put a human behind a screen with a textual interface and ask them to solve problems (novel or otherwise) based purely on text in and out then it's entirely possible for them to be a General Intelligence through that filter. If an artificial entity did well in the same environment enough it would be considered (by many) to be an AGI.

Therefore self driving is not a requirement to solve AGI. I agree it is almost certainly something where the capabilities inherent in making any AGI would be massively helpful, I posit it is a much much simpler problem in general because it is amenable to altering the rules of the game (if it turned out that we needed a legal requirement that all vehicles indicated their direction of intended travel by some V2V method and that clinched it to the point we could deploy it widely then we may well add that requirement to all new cars - effectively redefining the problem and shifting the goal posts - but to the end goal it's immaterial except in cost/time)

If you want to pick a specific task that's more likely to be a strict subset of AGI consider being able to describe a game in human language of any sort expressible through the textual interface and being able to play the game, and over time play the game well (better than 50% of the population). It's fine if it needs to ask questions to understand the game. The Alpha system used for Go/Chess and bunch of others is neat, but requires humans to explain (in code terms) the rules of the game. There's some usage of 'just playing' with a built in scoring function and the indication of success. But someone explaining it is rather different...

 

[gabemaroz]

there are plenty of humans who are utterly (legally and physically) unable to drive cars who are still considered (non artificial) general intelligences...

The blind are limited obviously by their ability to see and the elderly usually by their ability to react in a timely manner. The speed issue was tangentially addressed earlier but there is an implicit assumption that AGI would be fast. So what happens if it’s not? Moore’s law has tended to help overcome technical obstacles in the past, but polynomial time Big-O algorithms are dog slow no matter how fast the hardware gets.

Self driving is not a requirement to solve AGI. I agree it is almost certainly something where the capabilities inherent in making any AGI would be massively helpful

To clarify, I’m not implying it is a step to AGI but just an overall simpler problem. If driving a car is something AGI should be able to do (a subset), then it would make more sense to put the current resources directed at autonomous vehicles toward AGI.

Where I disagree is that I think neither self-driving, nor AGI will be achieved using the paradigms being used. More training and more data is not the path forward.

I recommend reading the criticism (and recommendations) of Gary Marcus on LLMs.

Consider being able to describe a game in human language of any sort expressible through the textual interface and being able to play the game

I like your example but I think requiring language is unnecessary. For example, imagine teaching a child rock, paper, scissors. It’s almost purely symbolic by nature.

 

[Bonusround]

It's two distinct problems. Don't get caught up in the fact that the programs used to solve these problems both employ "neural networks". Photoshop and Excel both employ sort algorithms and data structures.

The blind are limited obviously by their ability to see and the elderly usually by their ability to react in a timely manner. The speed issue was tangentially addressed earlier but there is an implicit assumption that AGI would be fast.

You need to define "fast": controlling the speed of a 70 mph vehicle, where milliseconds matter, or "fast" where I take an extra sip of coffee while AGI compiles its treatise about the Question to the Answer?

So what happens if it’s not? Moore’s law has tended to help overcome technical obstacles in the past, but polynomial time Big-O algorithms are dog slow no matter how fast the hardware gets.

A good point. It could happen. To date, state-of-the-art neural nets have proven miraculously parallelizable. But there are indications that some of the easy scaling may be ending.

To clarify, I’m not implying it is a step to AGI but just an overall simpler problem. If driving a car is something AGI should be able to do (a subset), then it would make more sense to put the current resources directed at autonomous vehicles toward AGI.

Thank you for the clarification. You've asked, repeatedly, whether self-driving will be a subset of AGI. I'll say no – driving a car or navigating the physical world at all is not something an AI will need to do in order to qualify as "AGI".

To wit: the MCP was trapped inside Dillinger's desk that whole time but still caused plenty of havok.

Where I disagree is that I think neither self-driving, nor AGI will be achieved using the paradigms being used. More training and more data is not the path forward.

It seems to be working thus far, as evidenced by myself and others on this thread. Self-driving is real, today. No, it is neither complete in its capabilities nor universal in its distribution – which is to be expected. It is getting better – detectably better – by means of greater training.

I recommend reading the criticism (and recommendations) of Gary Marcus on LLMs.

Will read, thank you. To your superset/subset question, I suspect what we eventually dub "AGI" will draw from (include) a multitude of LLMs, but won't be one at the topmost level.

I like your example but I think requiring language is unnecessary. For example, imagine teaching a child rock, paper, scissors. It’s almost purely symbolic by nature.

AGI will be a tool created by humans. It will need to understand our languages. That said, an AI that started from infant-level intelligence and learned languages would be a solid step in the direction of AGI.

 

[VirtualWolf]

I upgraded to an M4 iPad Pro and saw that the "Apple Intelligence" model is automatically downloading itself despite me not wanting it. I can't do anything to switch it off until it's downloaded, but it also appears to have gotten itself stuck, it's been sitting there saying "Downloading..." since last Thursday.

(Not that I'm particularly complaining, just more evidence that Apple's QA appears to have gone down the shitter in recent years, and "AI" in general is a garbage fire to begin with.)

 

[lloyddean]

Likely remote work effects in action or inaction.

 

[xoa]

Agreed. You see autonomous driving at the inflection point of an S-curve and I see it at the edge of a cliff.

No big argument overall but fwiw:
1. The original contention I disagreed with was specifically about iPilot05's assertion that people wouldn't tolerate any safety issues with FSD vehicles. You jumped in after that with some other stuff, but to be clear I wasn't really wading into the rest of the discussion just that specific point about that specific area and then to some extent some other reflections on Apple.

2. I still struggle how you can see the "edge of a cliff" for something that is providing real world commercial value in production right now. It doesn't need to conquer everything tomorrow or next year or next five years to be a big deal or be progressing at a rapid clip.

Only time will tell but I think we’ve exhausted that avenue of discussion for the here and now.

Sure, and cateye has been very patient with us all on this. But I do think there is one thing really worth noting in relation to Apple and Apple Intelligence as well as other efforts: regardless of how any of you feel about Waymo or other attempts on an absolute scale vs some future envisioned FSD standard, what I do think is completely objective on a relative scale is that they've been massively, massively more successful and productive then Apple. And not for lack of funding either! According to the New York Times article on its death Project Titan cost in excess of $10 billion, to produce nothing shipping. Actually reflecting back that's a really fucking wild number and failure that should have gotten far more attention then it has. That's like, the entire budget to develop Starship. Waymo "closed a Series C" (from Alphabet granted) about 3 months ago which brought their grand total capital raised to around $11 billion. So exact same rough area.

But Apple wasn't able to convert that remotely as efficiently, which to me points at the sorts of internal organizational problems we've remarked upon for a few years now. It's notable that Waymo is not a "Google" subdivision either. Even if it is under the overall Alphabet umbrella and ultimate control, it has its own budget and internal control and structure and independence. A lot of Apple's internal ways of doing things that work extremely well for high end consumer electronics don't to work as well for other areas of business. If they're really serious about this (and other emerging areas) I wonder if they wouldn't be better served by creating an actual independent subsidiary for it that won't be subject to the same internal Apple politics and DNA (and can also do things like be more open/collaborative while maintaining a firewall vs what Apple would prefer kept secret).

 

[cateye]

Sure, and cateye has been very patient with us all on this.

It's a great conversation, I'm really loathe to interrupt when things are interesting, even if it has strayed well beyond Apple's universe. But yes, let's look for ways to re-attach all this to what Apple is (or may) do with Apple Intelligence.

 

[wicker_man]

Has anyone found how to completely remove this junk from our iOS devices?

Apple’s flagrant disregard of users’ choices here is staggering, when a forced feature that quite a number of people do not want is foisted up0on them, taking up valuable space with Apple’s well-known skimpy attitude of endowing their basic devices with enough storage.

I hope someone out there is readying a class-action suit that would force them to provide what they should’ve done at the start: a complete opt-out by default with complete removal of all models.

As Steve used to say, ask them, then ask them again.

 

[KhaineBOT]

I upgraded to macOS 15.3 and under Privacy and Security I saw Apple Intelligence Report, and its enabled.

The description of what it is, is clear as mud. When you try and disable it, its even more confusing. Does anyone know what it's about?

 

[wicker_man]

I upgraded to macOS 15.3 and under Privacy and Security I saw Apple Intelligence Report, and its enabled.


View attachment 101333

The description of what it is, is clear as mud. When you try and disable it, its even more confusing. Does anyone know what it's about?

Looks to be similar to the current Siri dictation history etc. but obviously scaled up to the next level where all conversational stuff and also actions are recorded.

 

[Louis XVI]

Welp, my first experiment with iOS 18.3‘s visual intelligence went about as expected. To try out the new ability to identify plants and animals, I held down the camera button, pointed the phone at my dog in a well-lit room at a distance of about four feet, and hit the “search” icon. The phone promptly offered up a whole bunch of links to articles about chihuahuas. Which is kind of cool, except that my dog is a Cretan hound, which doesn’t look a whole lot like a chihuahua, and is about five times as big.

Ah well, maybe it’ll work in 18.4….

 

[gregatron5]

Did it ask to offload to ChatGPT or Google Image Search? I haven't pushed-and-held that button in ages.

 

[Louis XVI]

Did it ask to offload to ChatGPT or Google Image Search? I haven't pushed-and-held that button in ages.

There are icons for both. I chose Google, as I haven’t enabled ChatGBT on my phone, and don’t plan to.

 

[gregatron5]

OK, so that's Google's fault for the misinformation and transitively Apple's for using Google. It still offloads the actual work to services I don't wish to use, so I won't be using that "feature." Thanks for the response!

 

[dspariI]

Is this not built on the Visual Lookup that's been in Photos for the last few years? I've had good results using that for plants and birds at least. If not, the right thing has usually been in the visually similar images that get suggested.

 

[gregatron5]

It is not, which is both baffling and maddening.

 

[xoa]

Man, they're pushing this pretty hard . I have all auto updates and checks turned off on my 15.2 Mac and have since setup but it's still hitting me with constant "do you want to install this update now in the middle of work or try tonight".

 

[ShuggyCoUk]

Looks to be similar to the current Siri dictation history etc. but obviously scaled up to the next level where all conversational stuff and also actions are recorded.

https://support.apple.com/en-gb/guide/iphone/iphe3f499e0e/ios

 

[effgee]

Man, they're pushing this pretty hard . I have all auto updates and checks turned off on my 15.2 Mac and have since setup but it's still hitting me with constant "do you want to install this update now in the middle of work or try tonight".

This has, very much, been rubbing me the wrong way as well. Every time I walk away from my desk for a few minutes, I come back, the MBP wakes up and this fucking notification is back.

Not cool, Tim Apple. Not cool at all.

So, out of pure spite, because I can, and for shits and giggles, I looked up the URLs for Apples software update servers, conveniently provided by Tim Apple himself, and created a set of Little Snitch rules to block all outgoing traffic to them. Just to see what happens, and if perhaps this fucking nag-dialog might just stop appearing and pestering me.

Not a permanent solution, of course, but I'm feeling resentful towards Tim today, and will see how/if that goes.

Spoiler: URLs Apple says have to do with macOS updates

gdmf.apple.com
gg.apple.com
gs.apple.com
ig.apple.com
mesu.apple.com
skl.apple.com
swcdn.apple.com
swdist.apple.com
swdownload.apple.com
swscan.apple.com
updates-http.cdn-apple.com
updates.cdn-apple.com
xp.apple.com
gdmf-ados.apple.com
gsra.apple.com

 

[xoa]

This has, very much, been rubbing me the wrong way as well. Every time I walk away from my desk for a few minutes, I come back, the MBP wakes up and this fucking notification is back.

Not cool, Tim Apple. Not cool at all.

So, out of pure spite, because I can, and for shits and giggles, I looked up the URLs for Apples software update servers, conveniently provided by Tim Apple himself, and created a set of Little Snitch rules to block all outgoing traffic to them. Just to see what happens, and if perhaps this fucking nag-dialog might just stop appearing and pestering me.

Not a permanent solution, of course, but I'm feeling resentful towards Tim today, and will see how/if that goes.

Spoiler: URLs Apple says have to do with macOS updates

gdmf.apple.com
gg.apple.com
gs.apple.com
ig.apple.com
mesu.apple.com
skl.apple.com
swcdn.apple.com
swdist.apple.com
swdownload.apple.com
swscan.apple.com
updates-http.cdn-apple.com
updates.cdn-apple.com
xp.apple.com
gdmf-ados.apple.com
gsra.apple.com

Yeah it's pretty irksome, Microsoft-ish even. I don't mind setting the defaults to be aggressive about it, it is good for people to pay attention to. But this is a machine for work I'm working on and restarting and bringing everything back up is non-trivial and not something I want to do in the middle of things. If the operator turns the setting off that should be that. I do have Little Snitch but after that spell where it looked like Apple might exempt their own stuff and thus couldn't be trusted I stuck that list in my firewall instead, but same idea. We shouldn't need to do such things though at all. Sigh.

 

[effgee]

Well, I should've known that blocking one (outbound) way wasn't going to be enough... the notification returned this afternoon. Probably some flag or summit on Apple's end that'll continue the notification intrusions bombardment 'delivery' until the update's been completed. But I have since followed Xoa's example and added the fifteen URLs to my router's blocklist, and, having woken up the machine twice since, Tim Apple's uninvited, obnoxious seizure of my MBP's notification system appears to have stopped.

Cautiously optimistic that I'll be able to keep this set of for-now-at-least superfluous bullshit 'innovation of a mostly decorative nature', along with the unsolicited 'direct marketing' thereof, at bay for a few more weeks until I have a better idea of how this all shakes out.

If I want to play with an LLM in the meantime, I'll just install and run an instance of DeepSeek inside a VM. Whodathunkit, Chinese researchers offering users more of a choice than Apple. T'is truly the best timeline, this. S(He) who finds the /s gets to keep it.


ETA: ... it is the following day. Et voilà, no more nags. /tinyvictories

 

[Rocketpilot]

Is this not built on the Visual Lookup that's been in Photos for the last few years? I've had good results using that for plants and birds at least. If not, the right thing has usually been in the visually similar images that get suggested.

Wait are they actually deprecating a machine learning feature that actually works for one that ... doesn't?

I must admit I worry about people using any plant identification tool for distinguishing culinary mushrooms for Agatha Christie plot-advancing toadstools...

 

[jediatzinger]

I have an iPad Pro (M4) and iPhone 15 Pro Max. Last night I upgraded both to iOS 18.3. I had decided back when 18.0 was released I was going to wait for the “beta” tag to be dropped from Apple Intelligence before jumping in, but in 18.3 it appears it’s enabled by default.

Not cool, but it’s been displaying in Settings > Apple Intelligence & Siri
”Downloading support for Apple Intelligence. Connect iPad (or iPhone) to Wi-Fi and power while models are downloading.”

Hmm. It’s been exactly like this for 24 hours…? No status or percentage displayed? How long do you wait?

edit: I went into Messages and clicked on Genmoji and see “Downloading support for Genmoji. Once downloaded this iPad will be able to use Genmoji”

 

[GaitherBill]

Just turned this off on my daily drive iPad Pro M4.

Let’s see if I notice a tangible difference.

I still don’t know why I want any of this mining my actions.

 

[jediatzinger]

I'm likely going to turn it back off, but since it was enabled after installing 18.3 I thought I'd take a look... this morning it's still displaying on both my iPad and iPhone the same message as before. If it needed to download more I'd like to see a % of completion or an ETA when the AI stuff would be ready. I'm not a typical user (I'm a sysadmin), but IMHO this is a terrible first experience to Apple Intelligence. I shouldn't be wondering what's going on and/or when I'll be able to use certain features.

edit: fixed a few spelling errors.

 

[cateye]

IMHO this is a terrible first experience to Apple Intelligence. I shouldn't be wondering what's going on and/or when I'll be able to use certain features.

It wasn't appreciably different in the early betas and access to upstream resources was queued (for good reason, probably). Some people got approved (and access to things like Image Playground) in minutes. For others, it took hours, or even longer. There was no indication in the UI as to how long the wait would be, or why. Peak Apple.

 

[Bonusround]

Puzzling, isn't it? They have the CDN figured out for the OS updates themselves... why are the model downloads bottlenecked?

Not that this explains/excuses the download fails at all, but I wonder: are they distributing models specific to (tweaked for) the neural engine variants across their devices?

 

[jediatzinger]

/facepalm

Eating lunch 15m ago I was browsing the web with Safari and all of the sudden a pop up appeared and says "Apple Intelligence Is Ready" followed by "Image Creation Tools Are Here".

Guess it took a few days, but come on Apple... I don't mind waiting, but let the users know what's going on (% until ready, etc)

 

[Honeybog]

Donning a tinfoil hat, you kind of have to wonder if Apple will count that forced period of Apple Intelligence being enabled when they report Apple Intelligence users.

 

[cateye]

Of course they will. Not out of any nefarious purpose, I don't think, just tradition for such things. Just like someone enjoying a free AppleTV+ or Apple Music trial is a "subscriber" and counted as such, so anyone who has Apple Intelligence enabled at any point in a given reporting period is probably considered a "user."

 

[gabemaroz]

Well by that logic, Songs of Innocence by U2 is the number 1 album of all time by "purchases."

 

[Bonusround]

Of course they will. Not out of any nefarious purpose, I don't think, just tradition for such things. Just like someone enjoying a free AppleTV+ or Apple Music trial is a "subscriber" and counted as such, so anyone who has Apple Intelligence enabled at any point in a given reporting period is probably considered a "user."

"Whatever it takes, team, so long as our customer sat remains best-in-class." *

* (as measured by the firm we handsomely pay to measure such things)

 

[BadtzMaru]

So obviously I turned it off, but I don't remember where it stored that 7GB of training data, it was in some library/container location... maybe it was in this thread. Anyone remembers? Thanks in advance!

 

[cateye]

Not that this is a huge surprise, but Apple has confirmed that the more personally-contextual Siri, the "Apple Intelligence" feature that had the potential to be the most interesting or at least useful, is delayed:

Siri helps our users find what they need and get things done quickly, and in just the past six months, we've made Siri more conversational, introduced new features like type to Siri and product knowledge, and added an integration with ChatGPT. We've also been working on a more personalized Siri, giving it more awareness of your personal context, as well as the ability to take action for you within and across your apps. It's going to take us longer than we thought to deliver on these features and we anticipate rolling them out in the coming year.

(emphasis added)

It was originally expected to debut in some form by iOS 18.4, but "in the coming year" is a bit of a weasel phrase suggesting it may not be fully available until iOS 19.

 

[irfoton]

Must be the result of the “fixer” being assigned to the AI/Siri group. She clearly saw issues with the roll out.

 

[gabemaroz]

Not that this is a huge surprise, but Apple has confirmed that the more personally-contextual Siri, the "Apple Intelligence" feature that had the potential to be the most interesting or at least useful, is delayed:

My hope is that the “delay” is akin to what AirPower experienced…

 

[Dr_Lex]

This story on the BBC website yesterday about a garbled transcription can’t help public perception of the unreliability of AI-derived services.

https://www.bbc.co.uk/news/articles/c0l1kpz3w32o


“Egg freckles?”

 

[educated_foo]

This story on the BBC website yesterday about a garbled transcription can’t help public perception of the unreliability of AI-derived services.

https://www.bbc.co.uk/news/articles/c0l1kpz3w32o

It's worth pondering the absolute word salad Apple's "AI" transcription feature produced:

"Lots of ___ CT from work line over line trail of you will just be told to see if you have received an invite on your car if you've been able to have sex and not what should we call and just keep trouble with yourself that'd be interesting you piece of **** give me
a call _ customer _ will have to help us thank you..."

You'd think that, after they did speech-to-text, they would run the text through a grammar checker, or an LLM, or something that would recognize the result as gibberish. But then again, there's no room for QA in Apple's write-commit-ship pipeline.

 

[cateye]

Oof, this headline:

Apple pulls ad that showed Siri being useful after delaying feature that could make Siri useful

There's a chum-in-the-water feel to some of the follow up reporting to Apple's announcement, with both Gurman and Gruber indicating that the earliest they think we'll see an "improved" Siri is well into 2026 and that the level of dysfunction in its development is so high that it may require throwing out what's been done and starting over entirely. From the MacRumors article:

Bloomberg's Mark Gurman says that the ‌Siri‌ features "won't be released until next year at the earliest." Some people in Apple's AI division think the features could even be "scrapped altogether and rebuilt from scratch.

There are some concerns that fixing ‌Siri‌ could require more powerful hardware, which would mean Apple needs to reduce its feature set or make the models run more slowly on current devices.

Wonder if this will rise to the level of John Giannandrea (head of AI/machine learning) deciding to "leave Apple to spend more time with his family"?

 

[Bonusround]

Wonder if this will rise to the level of John Giannandrea (head of AI/machine learning) deciding to "leave Apple to spend more time with his family"?

Only if Apple’s Upper Executives wish to out themselves as genuine bozos… if, as has been reported, the decision to ship AI in iOS 18 was a premature, top-down directive.