Hey Siri, Who Powers You Now?

How Apple might handle their partnership with OpenAI...
Hey Siri, Who Powers You Now?

We're now just hours away from the keynote at WWDC where all things Apple AI – sorry, 'Apple Intelligence' – will be unveiled. And while there have been endless reports on both the high-level of Apple's work and thinking here as well as seemingly some very granular details about new functionality, as far as I can tell, no one has yet nailed down perhaps the key to the entire equation: how is Apple actually utilizing OpenAI technology in iOS 18? Is this a "brain transplant" for Siri, as it were, or something else?

Update: it seems to ultimately be "something else", more below...

The latest report from Tripp Mickle for The New York Times yesterday just leads to more confusion:

In recent months, Adrian Perica, Apple’s vice president of corporate development, has helped spearhead an effort to bring generative A.I. to the masses, said two people with knowledge of the work, who asked for anonymity because of the sensitivity of the effort.

Mr. Perica and his colleagues have talked with leading A.I. companies, including Google and OpenAI, seeking a partner to help Apple deliver generative A.I. across its business. Apple recently struck a deal with OpenAI, which makes the ChatGPT chatbot, to fold its technology into the iPhone, two people familiar with the agreement said. It was still in talks with Google as of last week, two people familiar with the conversations said.

That has helped lead to a more conversational and versatile version of Siri, which will be shown on Monday, three people familiar with the company said. Siri will be powered by a generative A.I. system developed by Apple, which will allow the talking assistant to chat rather than just respond to one question at a time. Apple will market its new A.I. capabilities as Apple Intelligence, a person familiar with the marketing plan said.

First and foremost, a rare moment in the spotlight for Corp Dev!1 Second, this is yet another outlet saying the OpenAI deal is done but that Google talks remain ongoing. This would clearly seem to indicate that Apple doesn't just want one partner for whatever they intend to release with third-parties. And this matters because it's key to how it might be implemented/rolled out. These partners will help "lead to a a more conversational and versatile version of Siri" which would certainly seem to indicate that OpenAI is at least partially powering Siri under the hood.

And later on...

Enter OpenAI, which has positioned itself at the forefront of the generative A.I. movement with ChatGPT. Apple plans to complement what OpenAI offers with technology that it developed in-house to do select iPhone tasks. Its system will help Siri set timers, create calendar appointments and summarize text messages.

The company also plans to promote its revamped Siri as more private than rival A.I. services because it will process many requests on iPhones rather than remotely in data centers. Apple’s privacy focus proved to be a sticking point during negotiations with OpenAI and Google because it wanted to limit what iPhone data partners received, a person familiar with the negotiations said.

So this almost reads as if OpenAI is powering Siri with Apple simply augmenting what Siri can do with their own apps, locally. But again, Apple is still apparently holding talks (or was recently) with other partners. Have they developed a way to simply mix-and-match different flavors of AI behind-the-scenes? That would be interesting, but complex. Especially from the product standpoint of how you surface what is going on to the user – especially important to Apple undoubtedly do to the 'hallucination' risk here. And how do you handle the "opt-in" here?

That has led many to wonder if Apple wouldn't instead ship a stand-alone ChatGPT app/service for that technology specifically. But there hasn't been a lot of rumors/reporting talking about this aspect/possibility.2

Trying to square this circle a bit, it might be possible that Apple has struck a deal with OpenAI to help power certain Siri queries, but that this might actually not be ChatGPT, but something else developed by Apple and OpenAI separately. While it's easy to conflate the two, remember that OpenAI is the company while ChatGPT is just one product.

I go back to something Sam Altman wrote the day GPT-4o was unveiled:

Our initial conception when we started OpenAI was that we’d create AI and use it to create all sorts of benefits for the world. Instead, it now looks like we’ll create AI and then other people will use it to create all sorts of amazing things that we all benefit from. 

We are a business and will find plenty of things to charge for, and that will help us provide free, outstanding AI service to (hopefully) billions of people.

Certainly Apple and OpenAI were in the midst of chatting when he wrote this...

So again, it's entirely possible we don't hear the word "ChatGPT" uttered during the keynote (or perhaps just a quick shout-out in noting that "OpenAI, which ushered in a revolution in this field with the launch of their ChatGPT service..." or the like). And instead, we get something more like "we've worked with our friends at OpenAI to use their latest large language models to help augment Siri, to make her smarter when it comes to what you may be looking for..."

Such a move may also allow Apple to run any OpenAI large language models on their own servers, thus allowing them to tout security, versus running in Microsoft's cloud, as ChatGPT does. Perhaps this is why Microsoft is apparently not thrilled with any OpenAI/Apple partnership (aside from giving the company another very credible consumer AI competitor).

On the other hand, running such an LLM on their own servers would be both expensive and require a lot of capacity. In their earnings report last quarter, Apple seemed to indicate they weren't going down that path with regard to AI spend to the extent that others were... Could they run something without training it? What would be the point of that? Is there some hybrid option here?

Perhaps Apple baked all of this in such a way that while OpenAI is the key partner touted on stage today, over time, they intend to bring in other partners to augment Siri. Perhaps more granular LLMs for specific tasks in specific fields, and the like. And, of course, over time, Apple would aim to build up their own models to the point where they rely on third-parties less and less (though this won't be mentioned in the keynote, obviously).

It’s possible that Apple may look to offer the improved Siri as a service, analysts said. By charging $5 a month for people to use the talking assistant, the company could generate $4 billion to $8 billion in annual sales, according to Morgan Stanley.

This feels less likely unless there was something special this premium version of Siri could offer – like the aforementioned tailored LLM integration, perhaps? On the other hand: Services, Services, Services. And this would allow Apple to give Wall Street a clear path to AI monetization. But I think touting the notion of upselling devices will be the play here – i.e. you'll need newer Apple hardware to access Apple Intelligence. While perhaps not a long term answer to Apple's growth issues (a nice law of large numbers problem to have, but still a problem), the prospects of an upgrade "super cycle" – the same general game plan as with AI PCs, sorry, "Copilot+ PCs" – in the near term would undoubtedly please Wall Street.

Update: In Apple's keynote at WWDC today, they announced a partnership with OpenAI which will allow users to opt-in to send queries to ChatGPT when Apple's own AI doesn't provide what you may be looking for. More thoughts below....

Apple’s Artificial Approach to ‘Apple Intelligence’
Apple is doing “AI” their way, in ways likely good and well, TBD

1 But was it really Perica spearheading this effort since this report last week by Aaron Tilley for The Wall Street Journal had Apple SVPs John Giannandrea and Craig Federighi clearly spearheading the effort...

2 Related is how Apple would serve up generative content such as images. Still just via Siri? Perhaps with a note that "this was generation by our partner OpenAI" or the like?