Apple Spending a 'Tremendous Amount of Time and Effort' on AI, Details Coming 'Later This Year'

[G5isAlive]

Apple has released several bits over the passed several months that shows what they’re working on and the direction they’re headed…

- Ferret which is a multi-model (image, text) system that has surpassed even ChatGPT 4 Vision in detail and accuracy
- Apple’s mlx framework which allows LLM’s to be built and run on Macs
- and their work in getting LLM’s to run on devices with much, much less memory by offloading part of it to flash memory.

It would seem to me that Apple’s main interest is getting “AI” on device rather than from the cloud, which is why we haven’t seen anything from them yet. But like the article mentions, I’ sure we’ll see something at WWDC this year. (Their text prediction system is iOS 17 is based off the same transformer models used by most of the LLMs.)

And Siri works fine for me, but I’ve been using it regularly for years. Maybe it just understands me better?

Thanks for the detailed explanation for reasons of hope for the future. Much appreciated. I am obviously not an expert. And like I said, Siri works fine for me with limited tasks. Been using it since it was released. But going to stand by my disappointment that it hasn't evolved much over 12 years and is still limited in what it could do. It understands my words fine. That it fits your needs entirely is great for you, but I wish it could do more. it is very picky on one's phraseology. I can get it to work, but often it takes recalling a very specific phrase. For example if I am driving on a preprogrammed route, but then ask it 'how far to x', because I am trying to figure out if its worth the bother, it will go to the web to try to find information on x, not distance, and often will start talking about something a continent away. No, I am driving. You know where I am. Context counts. But if I ask for directions to x, then it correctly finds the x that is nearby and plots a route. which, if I then decide 'never mind' then I have to cancel the whole route and start over. Sorry, thats just stupid. And it doesn't quite have music albums down yet (at least not reliably on the first try). But sure, it will tell me the time and the weather fine, as well as turn on and off lights when I ask it too.

And I agree, I much prefer my AI to be local because I often travel in areas with bad cell phone reception.

So, to summarize, I think Siri could use some improvement. You don't? because thats the gist of my message. Not if it works fine or not.

 

[9081094]

whatever Apple comes up with there are a few certainties:

A. It won’t come free
B. Tim Cook says that you’ll see this only coming from Apple
C. Whatever it is it will get leapfrogged by the competition in less than three months.

 

[obviouslogic]

whatever Apple comes up with there are a few certainties:

A. It won’t come free
B. Tim Cook says that you’ll see this only coming from Apple
C. Whatever it is it will get leapfrogged by the competition in less than three months.

A. Apple has released both their mlx framework and ferret model as open source.
B. The way in which uses technology is usually an Apple only thing - until everyone decides to do it to.
C. Apple’s Ferret model has leapt over and is more advanced than Chat GPT’s!!! I doubt anyone else will be able to top that in a few months time. In fact some of things it is capable of is the first of its kind.

 

[NT1440]

whatever Apple comes up with there are a few certainties:

A. It won’t come free
B. Tim Cook says that you’ll see this only coming from Apple
C. Whatever it is it will get leapfrogged by the competition in less than three months.

If it’s on-device AI, three months isn’t near enough time for the competition to change how their server dependent models work. Apple will be at least a year ahead.

Up until that white paper a couple weeks back the industry seemed in consensus that a good on-device model simply wasn’t feasible 🤷‍♂️

Seeing as it’s going to integral to the OS…it absolutely is going to be free.

 

[flybass]

A. Apple has released both their mlx framework and ferret model as open source.
B. The way in which uses technology is usually an Apple only thing - until everyone decides to do it to.
C. Apple’s Ferret model has leapt over and is more advanced than Chat GPT’s!!! I doubt anyone else will be able to top that in a few months time. In fact some of things it is capable of is the first of its kind.

Do you have any specific articles about ferret? From what I could find online, ferret seems mostly different from chatgpt vs more advanced.

 

[Rookbird¥]

Apple has drawn a line in the sand on not allowing voice recognition to be sent to servers.

Google/Amazon/etc don’t have that Hangup because their assistants are designed to get you to use their services…which if it’s free means it’s just data collection for their ad platforms.

I understand that and I applaud Apple for that but I think there was still a lot they could have done. I think the next step is a dedicated chip specifically for voice recognition. I have designs on how a dedicated chip could be created that could greatly advance voice recognition.

 

[NT1440]

I understand that and I applaud Apple for that but I think there was still a lot they could have done. I think the next step is a dedicated chip specifically for voice recognition. I have designs on how a dedicated chip could be created that could greatly advance voice recognition.

But that’s exactly the type of thing the NPU is designed for.

I think we should all revisit this topic in June and reflect back on just how long Apple has been working in this space quietly and out of the spotlight.

 

[MacFarmer]

At least in the apple of old, there were flexible teams. I’d guess, that team is long gone, but not because it was fired, but because it was dissolved and had to work on another project. That might also be the reason why bugs don‘t get fixed for years or new features often don‘t see any updates until they are forgotten.

ah, yes, I think you are right. I don't remember which app there was a big discussion, but it stagnated for years. Maybe it was notes or some other non-core app. But it had bugs, and was updated may years later with new features and other stuff. There was some interview with guy or some leak from inside that devs are moved around quite a bit, and that there are no dedicated teams or even a person, for some small apps. As you say, this might be the same with Siri.