Apple Rumored to Follow ChatGPT With Generative AI Features on iPhone as Soon as iOS 18

[dwaite]

Will it be called SiriGPT? Though since Apple loves using + and Pro/Ultra, maybe Siri+, Siri Pro, or Siri Ultra?

Naming wise - Siri or Siri+ if such a thing comes to exist as a subscription.

They'll likely need custom ASICs for server farms as well, for both training and inference. I wonder how far into that design cycle they're already in?

Training you just eat cost to process as fast as possible. Apple wouldn't make an ASIC there unless they were planning to compete as an AI training product.

Inference assumes Apple is running processing in the cloud. Historically they have done machine learning on-device and shared trained/tuned models.

Not really interested in more connected services, paid or not. Wonder if a pared down version could just run locally?

Possibly - but it wouldn't be a ChatGPT style offering at that point. You can run 8b parameter models on a M1/2 Pro/Ultra, but thats going slower and eating way more power than would be acceptable on device.

There's a question of whether having on-device answering of questions based on years-old information is really what an assistant should focus on. I suspect Apple GPT is an analyst pipe dream, personally. Assuming Apple is pumping money into machine learning, it is because they see specific opportunities.

Are they gonna push it back just like an earlier news regarding the AirTag 2?

Very unlikely there will be an AirTag 2 anytime soon. There is a standardization effort around detection that is just starting.

I think AI is a different kettle of fish. It klearns best when it has as braod an input as possible. The stance that Apple appears to take with on device processing/privacy will surely hold them back.

It hasn't in other areas - for example, they have been able train photo detection and personalized friend/family member/pet detection based on public image sets.

Personally I think we have to give up some degree of privacy up to get what we want. Lots seem to say "Oh no I'll immediately switch all that data mining off, which is fine for them, and it works becaisue luckily for them others leave their personal info up for grabs. The system learns more becasue it has access to demographics and it benefits them.

I don't think they'll put an assistant with encyclopedic knowledge on a phone. But phone assistants are there to perform actions and answer current world questions like weather and current events, not to help you write a paper on the civil war.

 

[Premium1]

Will it be called SiriGPT? Though since Apple loves using + and Pro/Ultra, maybe Siri+, Siri Pro, or Siri Ultra?

They are going to call it Siri "Can you say that again" ProUltraGPT

 

[gk_brown]

Apple spent too much on developing the useless Vision Pro when the hype of the VR/AR is already gone. Siri is only good at setting a timer. A businessman can't lead a technology company. Period. He doesn't even know AI is the most important thing to focus on...

Agree on Vision Pro and Siri. Disagree about Tim though. He's not a visionary but he is smart.

 

[wegster]

When you search for the hardware requirements for running ChatGPT models, you come across different estimates including the ones like this: "The largest model, GPT-3 “175B,” requires 600 GB of RAM and a specialized GPU with 250 GB of VRAM.".

For iPhone to be able to run a ChatGPT model, it would need to be foldable (10 times?) and cost (based on Apple upgrade strategies) a few million dollars.

Mostly true but the 175B is the # of hyperparameters, which along with the data type sizes (e.g. 32 bit floaing point for typical GPU, various 16 bit types, 8 bit, ..) can lead to a wide range of base resource requirements. LLMs can also incredibly resource hungry for training, but you can do ‘fine tuning’ which is much lighter weight, and GPT3/4 alternatives with far fewer hyperparams (e.g. LLaMa2 7B - there are a handful of competitors popping up in the reduced resource, reduced hyper-param space) can drop the runtime resource requirements quite a bit. There’s still a lot of work to be done in the field, and I’m unsure offhand how well AS neural compute nodes and/or AS GPU nodes would fare for smallest LLM execution only.

 

[wegster]

How could it be based on Siri even if Apple wanted to?

Siri has nothing to do with AI, as all of its content is human curated.

Eh. What do you believe is doing speech to text for Siri? Regardless of how poor the overall Siri experience is, or how ‘deep,’ it’s certainly doing a variant of NLP(Natural language processing)/speech to text via AI/ML..

 

[Fandroid killer]

Apple plans to start implementing generative AI technology on the iPhone and iPad in late 2024 at the earliest, according to Jeff Pu, an analyst who covers Apple's supply chain for Hong Kong-based investment firm Haitong International Securities.

In a research note on Wednesday, Pu said his supply chain checks suggest that Apple is likely to build a few hundred AI servers in 2023, and significantly more next year. He believes Apple will offer a combination of cloud-based AI and so-called "edge AI," which involves more on-device data processing. He added that patience will be required with Apple's rollout of generative AI, as the company considers how to use and process personal data in a way that aligns with its commitment to customer privacy.

If met, the late 2024 timeframe would mean that Apple could begin rolling out generative AI features starting with iOS 18 and iPadOS 18, but it remains to be seen how it uses the technology. The Information last month reported that Apple plans to incorporate large language models into Siri to let users automate complex tasks, a feature that would involve deeper integration with the Shortcuts app. The report said this feature is expected to be released in an iPhone software update next year, so it could be part of iOS 18.

In August, supply chain analyst Ming-Chi Kuo said Apple's generative AI efforts were "significantly behind its competitors," so it's unclear if the late 2024 timeframe is realistic. It is possible the technology will not be ready until 2025 and onwards.

Generative AI surged in popularity last year when OpenAI released ChatGPT, a chatbot that can respond to questions and other prompts. Google and Microsoft released similar chatbots earlier this year, as more companies race into the space. The chatbots are trained on large language models, allowing them to respond like a human.

In July, Bloomberg's Mark Gurman said Apple was internally testing a "ChatGPT-like" chatbot nicknamed "Apple GPT," but he said the company had yet to "devise a clear strategy" for releasing generative AI tools to the public.

Apple has publicly confirmed its interest in generative AI, and it has mentioned the technology in job listings posted on its website in recent months.

"We've been working on generative AI for years and have done a lot of research," said Apple CEO Tim Cook, in an interview with Forbes last month. "And we're going to approach it really thoughtfully and think about it deeply, because we're fully aware of the not-good uses that it can have, and the issues around bias and hallucination and so forth."

Article Link: Apple Rumored to Follow ChatGPT With Generative AI Features on iPhone as Soon as iOS 18

Knowing apple this is at least 5 years away. Apple will come out and say they revolutionized chatgpt

 

[wegster]

Word. It cannot be emphasized enough that we're still in heavy AI hypesummer. That doesn't mean there's no value to be had in the new tech, but nobody seems to be thinking hard enough about what it actually means for their project beyond marketing... what place are these things going to have when the market finally breaks under the oversaturation? And they all want to go with closed, gross, and heavily logged OpenAI nonsense.
In theory, the 15 Pro/Max is already capable of comfortably running some more lightweight LLMs on-device (7B@4bpw => 3.5GB RAM, more than enough for a Siri-supercharging integration) so Apple has ground to stand on if they wanted to go into the space cloud-free.

lol, always have to hate when going through a thread, typing a longish response mid-thread to get to the next pages to see - hey, someone already said that! If only ‘SiriGPT’ could have scanned the rest of the thread as I was typing in my prior (response to earlier thread posts, but after you posted this)..

But yeah…LLaMa2 and others significantly reduce resource requirements vs chatgpt3/4 and the like and can hold up pretty well on edge/lower compute type devices…and a lot of effort and research is being invested there.

RE: the hype. Also agree. ChatGPT unleashed a huge amount of ‘wow that’s amazing!’ and rightfully so, but given the current LLM limitations, even though a lot of investment $ went in, I’m not convinced of the viable near-term ‘killer app’ for them in the current state is. This is part of the general LLM pattern of not necessarily providing factual data but also others. No doubt they’ll come but not seeing them <now> to the point of being worth companies paying big $ for adoption to solve real business problems… yet.

 

[wegster]

Yes and Apple released Siri 12 years ago….Biggest upgrade came with ios17: it can now set two timers at the same time!

Yet given my phone and watch(or ipad) in close proximity to each other, with multiple alarms set on my phone, still cant reliably turn off ALL alarms….even where only a single device has them set.

 

[wegster]

I would argue that Apple's privacy and security stance is a logical fallacy. Meaning, they claim to maintain customers privacy by not sharing that data, but they're absolutely collecting it for their own purposes.

Yeah, but thats still infinitely better than tracking you <everywhere> and selling huge swaths of your data to anyone purchasing targeted ads (Google, FB, …).

 

[return2sendai]

Last line from Tim, “issues with hallucination” ?

 

[Le0M]

They could use that technology to make a wonderful translation app. Imagine all the synonyms, sample phrases etc it could give, for any given language.

 

[257Loner]

The race to develop Skynet is on! Which tech giant will open Pandora's box?

 

[Tagbert]

You must be kidding.

Confirmed: Apple Caught In Siri Privacy Scandal, Let Contractors Listen To Private Voice Recordings

Atherton Research's Principal Analyst and Futurist Jeb Su unpacks the ramifications of Apple letting outside contractors access private conversations of Siri users.

www.forbes.com

That was a perfect example of “tempest in a teapot”.

 

[svish]

Good to hear about this. Hopefully Siri improves.

 

[CarAnalogy]

Despite Apple having a reputation of being late to the party... they also have a reputation for redefining spaces in tech...

So I think when they do release this... you can be pretty certain they will make a good effort and get most of it right... plus their Neural Engine hardware is amongst the best.

Plus, I trust them on privacy way more than Google.

Yes. But I swear if they claim this requires “the neural engine found only in the iPhone 16 Pro” or something I’m gonna be pissed. For many years now they’ve been talking about the neural engine in their SoCs without a word about wtf it’s for. Only time they really explicitly mentioned it was to tell everyone that their pleb watches can’t handle the new Siri so go out and buy the new one. Well wtf is this neural engine sitting here doing then?

 

[patent10021]

As one fellow on Reddit accurately said,

For a trillion dollar company, it’s actually impressive how bad Siri is. And given how much time they’ve had to improve it, I highly doubt this will make it any better.

You should see the thread about this announcement. It's huge.

 

[Pruus]

I have heard that version 19 will be not only better, bug free, but also faster and less power consumption. PLEASE STOP THESE RUMORS. Apples marketing works again. Don’t be a fool. Let’s wait for a big free version 17.

 

[wanha]

Eh. What do you believe is doing speech to text for Siri? Regardless of how poor the overall Siri experience is, or how ‘deep,’ it’s certainly doing a variant of NLP(Natural language processing)/speech to text via AI/ML..

Sure, but that comparison is like saying fax is like email because they both involve the transmission of information wirelessly.

 

[SevenSeas]

I am extremely content with Apple taking their time incorporating generative AI and LLMs into the ecosystem.

The competition seems to be throwing things at the wall to see what sticks. I’d much rather have a polished, highly-functional, useful generative AI implementation than a half-baked one.

I’m particularly concerned about the data and privacy implications of generative AI and LLMs. As the article alludes to, I believe this is the primary reason Apple is taking much longer than Google, Microsoft, OpenAI, etc. to roll out the headlining AI features.

I would tend to agree, but Siri has been on the street for far too long to as unpolished and behind its competitors as it is.

 

[wegster]

Sure, but that comparison is like saying fax is like email because they both involve the transmission of information wirelessly.

Sure, but your original comment said Siri is not using AI/ML, which is inherently wrong.

The entire implementation end-to-end, e.g. starting from user voice 'request' to what can happen as an action or response, is entirely miserable beyond doubt, but it doesn't mean Siri has zero AI/ML components in place... just they aren't good even at the start (speech to text), let alone further downstream leading to useful results.

I was reminded of this yet again - when once again this morning I woke up before my string of alarms, covered my watch and closed my iPad pro magic KB so the ‘right’ Siri only heard me, told it to shut off all alarms from < a foot away, to have Siri fail miserably three times telling me I have several alarms, with no way to once again scream ALL as a response, so instead going through the whole series again, failing and repeating until ‘she’ finally accomplished the ever-so-complex task of ‘turn off all alarms.’ Thanks for turning a calm wake-up into elevated blood pressure and wanting to scream within minutes, Apple/Siri.

And for some unknown reason, they expect people to use Siri on their Macbooks???? Get real Apple.

 

[chucker23n1]

How could it be based on Siri even if Apple wanted to?

Siri has nothing to do with AI, as all of its content is human curated.

Siri has been AI/ML from the start.

 

[Tagbert]

Siri has been AI/ML from the start.

Siri has been an older gen of AI with a significant amount of hard coded pathways that make updates very complicated.

 

[jonnysods]

I wonder how they will keep privacy as a factor in this. I hope they can pull it off.

 

[doelcm82]

They said the same thing about newspapers telephones and the internet

I rest my case.