Apple Says iOS 18's ChatGPT Integration on Track for Later This Year

[Dark_Omen]

1. Apple is still on its publicly announced timeline.
2. If you were developing software at Apple's scale you probably would've told us.
3. Releasing features as they're ready instead of trying to prepare everything at one is responsible modern software development at large scale.

Apple doesn't release numbers, but I think a conservative guess is 100,000,000 lines of code (not counting Xcode and related tools) at 3,500 developers.

It won't take long if they use Chat GPT to write the code. 5 minutes, tops.

 

[obviouslogic]

apple is behind everyone else in AI. so much they had to rename A.I lmao.

of course it's going to take them a while to get the basic AI stuff out. plus they have iPhone 16's to sell you

Getting tired of this... "A.I. = LLM" therefore, if you don't have an LLM you're behind in A.I.

Apple decided to go down a different path and concentrate on on-device models - and they've succeeded, even surpassing everyone else in that regard. Apple being "behind" is just from the perspective that Apple doesn't throw technology against the wall and see what sticks. They are extremely thoughtful on how they use and implement technologies. Apple Intelligence is a clear example of that. They didn't just create an LLM and plug it into their operating systems, they created disparate features throughout the system... and, instead of wasting resources and creating their own LLM to compete against all the others, they decided to allow users a method to use their "favorite" LLM's if they so choose.

I also think it was a wise decision to separate Apple Intelligence features from other A.I. features in the system, so that "Apple Intelligence" mainly refers to those things that generate content or perform a task. So things like, object detection, scene detection, voice recognition and replication, text/handwriting recognition and replication, sound recognition, language translation, etc. are all not part of it.

 

[winxmac]

It's going to be an AirPower moment... Announced to happen at a certain date but may never happen at all...

However, for anyone looking to make use of Apple Intelligence in non-beta form, it should be ready around mid-2025 probably by WWDC 2025...

It's not for me anyway...

 

[CrysisDeu]

I perfer the lets do it right vs ship it so we don't miss our made up deadline.

Or actually plan things correctly

 

[SkweeBop]

Greeeaat. Now my students can look up wrong answers directly from their iphone

 

[CrysisDeu]

I just don’t understand. Apple boasts about their large server model beating gpt4, why do we still need gpt integration then?

 

[CrysisDeu]

Getting tired of this... "A.I. = LLM" therefore, if you don't have an LLM you're behind in A.I.

Apple decided to go down a different path and concentrate on on-device models - and they've succeeded, even surpassing everyone else in that regard. Apple being "behind" is just from the perspective that Apple doesn't throw technology against the wall and see what sticks. They are extremely thoughtful on how they use and implement technologies. Apple Intelligence is a clear example of that. They didn't just create an LLM and plug it into their operating systems, they created disparate features throughout the system... and, instead of wasting resources and creating their own LLM to compete against all the others, they decided to allow users a method to use their "favorite" LLM's if they so choose.

I also think it was a wise decision to separate Apple Intelligence features from other A.I. features in the system, so that "Apple Intelligence" mainly refers to those things that generate content or perform a task. So things like, object detection, scene detection, voice recognition and replication, text/handwriting recognition and replication, sound recognition, language translation, etc. are all not part of it.

Lol Apple’s ML model is way under performing. The keyboard auto suggestion and auto correction is just okay. Voice dictation is horrible.

When other phones have auto subtitle and translation systems wide, iOS 18 with focus on AI still doesn’t have it. Even on the pro side, fcpx does’t have auto subtitles

And magic eraser just coming this year? Lmao

 

[xmarcuswildx]

Shareholders got ahold of those rumors huh, Tim?

 

[gk_brown]

I work in engineering and I have very tight and stringent deadlines.

I work in engineering too, and tight "deadlines" tend to produce bugs and inferior products.

 

[obviouslogic]

Lol Apple’s ML model is way under performing. The keyboard auto suggestion and auto correction is just okay. Voice dictation is horrible.

When other phones have auto subtitle and translation systems wide, iOS 18 with focus on AI still doesn’t have it. Even on the pro side, fcpx does’t have auto subtitles

And magic eraser just coming this year? Lmao

Show me how their model is under performing.

OpenELM

Apple releases OpenELM, a slightly more accurate LLM

It's not the fastest machine learning model, but you can't have everything

www.theregister.com

half the tokens, yet more accurate.

Again, Apple is focused on on-device models… and they are ahead of everyone else in that regard.

show me something other than anecdotal heresay.

 

[obviouslogic]

I just don’t understand. Apple boasts about their large server model beating gpt4, why do we still need gpt integration then?

those are performance metrics; speed and accuracy for their models. Apple wants no part in “world” LLMs. The models they want to ship are tailored for their products and users… outside of that, users have the option of using ChatGPT.

they made that clear at WWDC.

Again, just like their SoC’s they optimized it for efficiency - to run on device. So they ended up with very efficient and performant techniques.

 

[dropadrop]

I work in engineering and I have very tight and stringent deadlines. I am unsure why Apple just lets things get delayed so willy nilly. Seems like they’re very laissez-faire with deadlines in engineering.

I work in product and believe most deadlines are made up and self imposed.

It’s different if you are a contractor signed up to build a bridge on a schedule. In this case the contracts will result in penalties if you miss the deadline.

Many engineering projects finnish at the deadline. For a product launch the story starts at the deadline, and the quality of the release will greatly affect what the story is.

I think it takes guts to delay something like this and applaud Apple for it. I think it shows a good example to other product houses that the initial quality matters.

 

[CrysisDeu]

those are performance metrics; speed and accuracy for their models. Apple wants no part in “world” LLMs. The models they want to ship are tailored for their products and users… outside of that, users have the option of using ChatGPT.

they made that clear at WWDC.

Again, just like their SoC’s they optimized it for efficiency - to run on device. So they ended up with very efficient and performant techniques.

Yeah yeah yeah. Their large model is definitely just a LLM. Siri is also a chatbot to some extent

 

[svish]

Good to know. Expecting it by November.

 

[MAlbi]

All the current available AI is just monetized SaaS of questionable value to me as end user and customer. AI that stands for Advertised Increment.
Untill AI creators and content owners come to conclusion about terms on how the AI can use that content and process that content to my benefit, this will be toy for one rainy afternoon and semi-successful search engine.
Those developers who do not care about anything and feed AI, they will produce something usable. But that is going to be blocked.
And again, why would I need 10 12 14 GB RAM iPhone if I can access AI on the internet with iPhone SE 1gen. I am looking forward what Apple is going to bring to the table and convince me that I am wrong. Except for privacy that now is almost good but not very good or perfect at all.

 

[The Game 161]

apple is behind everyone else in AI. so much they had to rename A.I lmao.

of course it's going to take them a while to get the basic AI stuff out. plus they have iPhone 16's to sell you

They were behind

fair to say what they are offering this year they will be ahead or at least level with google

 

[hagar]

They were behind

fair to say what they are offering this year they will be ahead or at least level with google

What? Gemini is available in 40 languages. Apple will reach this … never. They’ll offer 5-6 languages and then get bored.

 

[The Game 161]

What? Gemini is available in 40 languages. Apple will reach this … never. They’ll offer 5-6 languages and then get bored.

And? Are we really thinking this is a big deal that it’s offered in a lot of languages? If you speak English what other languages it uses is hardly giving you a better al experience or matters

And for record Apple has said they will be using others like Gemini as well going forward.

 

[Tagbert]

What? Gemini is available in 40 languages. Apple will reach this … never. They’ll offer 5-6 languages and then get bored.

I'm hopeful that they will be able to integrate more languages more easily that they ever could with Siri. Siri's mix of hard-coded and Natural Language Processing code-base was very hard to modify and expand. The new Siri uses LLMs and LLMs are much better about handling languages that the older code could ever hope for.

 

[garranhado]

All buzz about Apple AI and we have all of AI things today with OpenAI (and others) for free and running on old devices and multi language support.

Apple should fully integrate OpeanAI on OSs by now.

 

[sdf]

It won't take long if they use Chat GPT to write the code. 5 minutes, tops.

I know you’re joking and it was funny, but I admit I cringed at the thought. 😀

 

[Grey Area]

Getting tired of this... "A.I. = LLM" therefore, if you don't have an LLM you're behind in A.I.

Apple decided to go down a different path and concentrate on on-device models - and they've succeeded, even surpassing everyone else in that regard. Apple being "behind" is just from the perspective that Apple doesn't throw technology against the wall and see what sticks. They are extremely thoughtful on how they use and implement technologies. Apple Intelligence is a clear example of that. They didn't just create an LLM and plug it into their operating systems, they created disparate features throughout the system... and, instead of wasting resources and creating their own LLM to compete against all the others, they decided to allow users a method to use their "favorite" LLM's if they so choose.

I also think it was a wise decision to separate Apple Intelligence features from other A.I. features in the system, so that "Apple Intelligence" mainly refers to those things that generate content or perform a task. So things like, object detection, scene detection, voice recognition and replication, text/handwriting recognition and replication, sound recognition, language translation, etc. are all not part of it.

Agreed. LLMs have become the current hype after dotcom > web 2.0 > social media > blockchain > NFT > metaverse (probably missed a few buzzwords). Everyone feels the need to dabble a bit in LLMs to show they are "not behind in AI". At the same time it is not at all certain LLMs are going anywhere. Sure, they are impressive compared to what was possible ten years ago, but they quickly reached the point of diminishing returns - despite throwing ever larger datasets and data centers at them, performance remains stagnant and unreliable, and actual applications are missing. I think the technology is lacking some fundamental ingredients and facing challenges that cannot be overcome by the brute force of sheer quantity and money. OpenAI is the main driver behind the hype, yet I see no indication that they know how to move LLMs beyond funny chatbots. Without some major breakthrough in the next 1-2 years I think interest in LLMs will dwindle. Apple is smart not to associate themselves too closely with what could very well be another passing fad.

 

[con2apple]

My tip: don't upgrade until spring. When the promised functions have been delivered with iOS 18.1 and 18.2 and the worst bugs removed.

Just like every year.
Apple promises too much and therefore delivers unfinished software. March/April upgrades are safer. Which doesn't mean that all the promised features will be integrated by then.

 

[spazzcat]

My tip: don't upgrade until spring. When the promised functions have been delivered with iOS 18.1 and 18.2 and the worst bugs removed.

Just like every year.
Apple promises too much and therefore delivers unfinished software. March/April upgrades are safer. Which doesn't mean that all the promised features will be integrated by then.

18.1 most likely will come out in October and 18.2 before the end of the year.

 

[dropadrop]

And? Are we really thinking this is a big deal that it’s offered in a lot of languages? If you speak English what other languages it uses is hardly giving you a better al experience or matters

And for record Apple has said they will be using others like Gemini as well going forward.

Apple does not usually offer services in countries if it’s not localized. I have the ”one” service bundle here in Finland and it’s missing news (could be licensing in part) and fitness…

I’d be more then happy to use them in english as it’s my primary language, but that’s not an option.