The industry is moving quickly, but the latest research suggests that small models (13 or even 7B) can outperform large models when trained on a narrow input. So it is possible that Apple will end up with a mixture of small experts doing a variety of tasks on-device. This will both perform well and preserve privacy because everything is done on device.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple Rumored to Follow ChatGPT With Generative AI Features on iPhone as Soon as iOS 18
- Thread starter MacRumors
- Start date
- Sort by reaction score
The problem with Siri isn’t the AI, it’s the connection.
HomePods or CarPlay are constantly responding with “sorry, I’m having trouble connecting right now” for something like a request to turn the lights off while my Echos work 100% of the time.
HomePods or CarPlay are constantly responding with “sorry, I’m having trouble connecting right now” for something like a request to turn the lights off while my Echos work 100% of the time.
Whatever they call it, I would LOVE to be able to put Siri to good use finally. It's already hooked way into my email, calendar and everything else but it's as dumb as a bag of rocks. Give it more powers and give it some common sense. Let me actually DELEGATE tasks to it, the way you would with an actual human assistant.
"Siri, go onto Google Flights and look up round trip non stop trips to LA right after Thanksgiving, leaving in the afternoon if possible. Also look up car rentals and hotels with a max of $200/night."
"Siri, find a easy few outdoor activities for next Saturday morning, and lunch options nearby."
"Siri, go onto Google Flights and look up round trip non stop trips to LA right after Thanksgiving, leaving in the afternoon if possible. Also look up car rentals and hotels with a max of $200/night."
"Siri, find a easy few outdoor activities for next Saturday morning, and lunch options nearby."
Not everyone is using these, especially because they are extremely hard to get. Google uses TPUs, for example. I'd expect Apple to use their own neural engine.
I’m not clear how Apple can compete in generative AI with non-privacy focused companies. As I understand it, AI requires a ton of learning data—more the better, but Apple refuses many types of personal data, which means Apple is inherently at a disadvantage. Not clear on how big of one.
Of course it’s possible Apple doesn’t intend to compete, but just provide their users with a portion of the functionality found on ”competing” systems. But people will compare regardless.
Of course it’s possible Apple doesn’t intend to compete, but just provide their users with a portion of the functionality found on ”competing” systems. But people will compare regardless.
Sometimes you do need to check what version you're running, and asking Siri could be more convenient than digging through the settings menu. First world problems, I know, but why WOULDN'T Siri be able to tell us this?
Answer: If I ask Siri what version of iOS the phone is running, it answers "You're Running iOS 17.0.3". First world problem solved.
But if I ask "What model iPhone is this?", Siri simply doesn't answer, not even with a quippy "Interesting question". That's weird.
I just hope, the will not do the same crap like microsoft with windows 11. Microsoft is controls now over AI, what theyr users write and read (across all programs) and will even cancel their microsoft licenses, if they want to.
It would be very sad to use apple products with linux...
It would be very sad to use apple products with linux...
AI is going to revolutionize miseducation and propaganda.
Like always, then they show everyone else how it should have been done.
Worked for the phone, the mp3 player the watch and so on..
I actually think Apple Maps is much better than Google Maps these days.How are they going to come up with anything remotely resembling AI if they cannot get the basics right?I am rooting for Apple but Siri is a complete mess.
View attachment 2298500
For Apple building a generative AI feels a bit like Apple trying to compete with Google Maps with their very own Apple Maps. Despite their efforts Google Maps is and probably always will be light years ahead.
It doesn’t need to work as long as people are mesmerized by that iridescent orb.How are they going to come up with anything remotely resembling AI if they cannot get the basics right?
View attachment 2298500
For Apple building a generative AI feels a bit like Apple trying to compete with Google Maps with their very own Apple Maps. Despite their efforts Google Maps is and probably always will be light years ahead.
I mean, they certainly haven’t shown people how voice assistants should be done in the last 10 years.
I mean, it is the future. It’ll be able to automate so many things and save so much time. For better and worse.
I just tried the following:
I’d argue it actually did respond in the style of Siri.
“as the company considers how to use and process personal data in a way that aligns with its commitment to customer privacy.”
In plan English they are going to access EVERYTHING on your phone. Email accounts, social media accounts, bank accounts and any other service you log into. It will be Apple who decides how they use it.
Can’t wait to see how those TOS read.
Lol, totally not wrong.Large language models would create a whole lot of different ways to say "I don't know what you mean by _______. Here's what I found on the web for _________." Synonyms for words in those lines are abundant and there are many ways to restate the same confusion.
Would be great to just get to Siri is reasonably useful and correct more times than not as a solid start.
Generative in context of speech or text to visual, like Midjourney and friends can be a lot of fun, but LLMs are a somewhat different yet related beast as trying to respond with 'authoritative answers' (e.g. chatbot, Bing, Bard, search summary results, etc.) as the problem is beyond traditional 'AI Bias' which is generally down to the source software input data, for example, AI/ML apps built for 'crime detection' depending on the training set may improperly 'featurize' a racial attribute with an improperly high weighting leading to exhibit a racial bias in that certain skin tones may be higher weighted versus the actual activity in progress.
For LLMs, they generally operate on patterns of words and phrases, e.g. given a single word, what is the most probably next word or set of words, and given a phrase, what is the most likely <next phrase> based on it's data used for training, which is effectively 'swaths of the internet' with nothing in there whatsoever evaluating on if the data source, or 'constructed response' is actually in fact, true or not.
'Recipes' using poisonous ingredients are but one of the symptoms of the problem, although obviously the above is a huge simplification and plenty of people continue to work on LLM and LLM variants, but it certainly is not 'general AI' (although also being worked on).
So - should be interesting. Let's see if Siri actually understands us better + some fun with generative AI Art would be fun. A whole lot of 'let's see how this works out' on the rest of it...
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
With Apple's QA efforts on products getting worse and worse the thought of all the issue with buggy Apple AI is scary. Apple you're the biggest tech company in the world invest in a world class QA team.
It’s definitely both.
Also, confidently mishearing: “Hey Siri, 17 minute timer.” — “1 hour and 10 minutes starting now.”
Current gen Siri on device, and SiriGPT requires cloud connection? I could get behind that.
For training, yes, although there are various ongoing attempted challenges. Inferencing is a different story and there are a handful of lower power options out there, and while I haven't fallen down this rabbit hole yet/had time to play with it, presumably Apple Silicon 'Neural Engine' nodes should be sort of ready-made for inferencing tasks.