Apple's Large Language Model Shows Up in New iOS Code

An anonymous reader shares a report: Apple is widely expected to unveil major new artificial intelligence features with iOS 18 in June. Code found by 9to5Mac in the first beta of iOS 17.4 shows that Apple is continuing to work on a new version of Siri powered by large language model technology, with a little help from other sources. In fact, Apple appears to be using OpenAI's ChatGPT API for internal testing to help the development of its own AI models. According to this code, iOS 17.4 includes a new SiriSummarization private framework that makes calls to the OpenAI's ChatGPT API. This appears to be something Apple is using for internal testing of its new AI features. There are multiple examples of system prompts for the SiriSummarization framework in iOS 17.4 as well. This includes things like "please summarize," "please answer this questions," and "please summarize the given text."

Apple is unlikely to use OpenAI models to power any of its artificial intelligence features in iOS 18. Instead, what it's doing here is testing its own AI models against ChatGPT. For example, the SiriSummarization framework can do summarization using on-device models. Apple appears to be using its own AI models to power this framework, then internally comparing its results against the results of ChatGPT. In total, iOS 17.4 code suggests Apple is testing four different AI models. This includes Apple's internal model called "Ajax," which Bloomberg has previously reported. iOS 17.4 shows that there are two versions of AjaxGPT, including one that is processed on-device and one that is not.

TOS Violation ?

[vlad30]

I thought ChatGPT had a rule against using them to train another model or maybe this is a grey area where its only comparing the results

Re:

[dfghjk]

Well that is what the article says and comparing results is VERY different than training, so what is the point of your inflammatory post?

Re:

[zenlessyank]

Karen Alert!

Re:TOS Violation ?

[Entrope]

The actual prohibition [openai.com] is broad:

You will not, and will not permit End Users to: [...]
(e) use Output (as defined below) to develop any artificial intelligence models that compete with our products and services. However, you can use Output to (i) develop artificial intelligence models primarily intended to categorize, classify, or organize data (e.g., embeddings or classifiers), as long as such models are not distributed or made commercially available to third parties and (ii) fine tune models provided as part of our Services;

It's "develop" in general, not necessarily as training input. Using ChatGPT as a basis for comparison or asking it to score output would fail under "develop".

Maybe Apple negotiated a special license.

Re:

[ACForever]

More likely apple is breaking the rules and doesnt care.

Re: TOS Violation ?

[Anonymous Cward]

They have to be developing something which competes with their products and services to fall foul of that, and since Siri pre-dates the existence of OpenAI, Apple legal could argue they are not competing and operate in two completely different sectors. OpenAI is probably glad to rake in the money per API call.

Re: TOS Violation ?

[gnasher719]

Itâ(TM)s not used for training. Apple does all the training of their own software themselves, and then some human takes the output and checks that it is as good or better than ChatGPT.

Re:

[oogoliegoogolie]

I wouldn't bet on it. AI has made most things worse. I find that over the past year or two the accuracy of searches have noticeably lessened. Search terms are ignored even when +forced. I long for the days where + or - switches were religiously obeyed. It's like AI is delibertly trying to tell you that it is 'smarter' than you and knows more about what you want to search for than you.

AI generated written content is easy to spot if you are looking for it. Proper grammar and syntax, but it just doesn't re

Revealed - Training dataset for Apple's LLM

[93 Escort Wagon]

It's basically just several million variations of "here's what I found on the web about ...".

I Smell An Acquisition in the Air

[organgtool]

Apple has traditionally had a habit of acquiring companies to work on technology that's outside of their usual wheelhouse. When they needed multitouch for the iPhone, they purchased FingerWorks. When they needed a personal assistant, they purchased SRI to get Siri. As much as Apple loves doing things in-house, they got a very late start on LLMs and I wouldn't be surprised if they ultimately end up acquiring one rather than trying to get their own up to snuff.

I imagine some people will point out that Apple is rarely the first to market and instead opts to release polished products once the market is a bit more defined. That strategy works extremely well for consumer devices, but LLMs need a LOT of training to provide a polished experience and I just don't see Apple being able to come from that far behind with such a huge barrier to entry. Their best bet would probably be to find a company that has a well-trained LLM, purchase that company, and use Apple's knowledge of chipmaking to develop ASICs that can process the commands of that particular LLM with far greater efficiency than their rivals.

Re:

[PhunkySchtuff]

You seem to be making the assumption that Apple haven't been working on LLMs.
Apple haven't talked about, or demonstrated any work on LLMs, but they are famously secretive and this doesn't mean that they haven't been working on them for a long time.

Re:

[Plumpaquatsch]

So is Google by the way. Including things like "Google" Maps.

Everyone uses that API

[WaffleMonster]

The API is quite popular even for projects and models having nothing to do with OpenAI. It has basically become a generic interface for accessing LLMs.

Highly skeptical

[hdyoung]

Apple does a LOT of things well, but they ALWAYS half-assed Siri. Not that Siri is bad, just limited and mostly irrelevant, and Apple has put in very little work to expand it. Probably because Siri makes them absolutely no money, and I have a hard time seeing how an improved AI would change that.

Unless Apple jumps into the deep end of selling ads and monetizing user data. Which would be an indication Apple is on the long decline. Kind of like when GE got into financials. Dont do it Apple. Your bread and butter is your stellar hardware and software design and build quality. While were here, this switching over to services is a bad idea too - if you stray from your core competency it probably wont end well.

Integrate ChatGPT into Studio or Xcode?

[seoras]

Apple's always harped on about empowering developers, and especially those new to development, to take up Xcode and create stuff.
Given how much I use ChatGPT side by side with both Android Studio and Xcode I'm now wondering when I can expect integration in the same way as git has been.
A chat window next to the code window and the LLM has the code I'm looking at, and even the whole project if I want it to, as context to the questions I'm asking of it.
Has someone done it already?
A quick google search just chucks up guides to writing ChatGPT Apps which isn't the same thing.

Re:

[seoras]

OK, spoke too soon and without looking hard enough.
Found this nice plugin for Android Studio. [jetbrains.com]

Anyone aware of a similar one for Xcode?