Apple will reportedly pay OpenAI with 'exposure' | Hacker News Reader

Apple will reportedly pay OpenAI with 'exposure'

windowscentral.com

14 pointspjmlp32 comments

They're also paying with free user data (I know apple has invested a lot in privacy etc. here, but the bodies of your queries still end up going through OpenAI)

>but the body of your queries still ends up going through OpenAI

depends on whether there's an opt-out/agreement to not use the data for training. Even the free version of chatgpt has an opt-out for training data, although they still say they're going to retain the data for 30 days.

training is only one of many uses for the data.

I think I remember from the first WWDC presentation that Apple made some agreement with OpenAI that forbids them from using the queries for training and anything else. I'm not 100% sure if I remember it correctly though, maybe someone else can give a more confident recall of it.

Even if it's not used explicitly for "training", it's still valuable for model evaluation, A/B testing, etc.

(e.g. "do users engage more with model A or B?")

And to be clear, I'm not accusing Apple or OpenAI of being "evil" here wrt. handing user data, I just think it's a potential factor in their partnership.

> , but the bodies of your queries still end up going through OpenAI

Only the queries that Siri deems to need extra help and only after user approval..

and we know how well users opt-out of things like this or understand / care enough to opt-out / opt-in

Sure, they generally just tap/click whatever they always tap click to get the modal out of the way.

In either case, data is sent to OpenAI only in specific contexts, not all the time from all the AI interactions in the OS, which was the distinction I was making.

Will Apple save the data, so that they can use it to build their own AI systems?

Apple: we're the gatekeepers, you get exposure until we've figured out this AI thing.

It's between ChatGPT being on there or something else like Claude... OpenAI cannot afford to let the masses know there is something out there that is just as good

Gemini has on device

It is ironic, as so are offered artists and writers whom the C-suite want to replace with AI.

Beyond the data, OpenAI is getting access to the most valuable users on the planet. Everyone using this will see that "ChatGPT" is what's used when Siri is not smart enough. It really puts their brand out there in a big way. They want to be a household name, like Google.

It's also smart for Apple because they can slowly improve Siri (one can dream) so that it falls back less and less on ChatGPT. They can also make a deal with Google (Gemini) and others, so that they can start giving users the choice of what LLM to use, which means they can start charging Google or OpenAI another 10Bn++ for being at the top of the list, or being the default choice entirely.

Apple users will want to use OpenAI regardless of the deal with Apple.

I'm guessing it's more likely that OpenAI just want's the data that hundreds of millions of users are searching for vs. the much smaller group of 'technical' users niche users who are already using ChatGPT.

So, OpenAI - Microsoft - signed a contract to provide their service in exchange for exposure. Not a deal I'd take, but I'm not a corporation with billions of dollars in market value and the brand name when it comes to AI.

Apple absolutely sucks when it comes to pushing small developers and artists into bad contracts. But we're talking about Microsoft here, so it's not like they have no power to push back on contract terms they don't want.

I think this is broadly explicable by GPT etc. still being in the market growth stage of development in general. Grabbing market share and cutting off competitors is the name of the game at this point in the cycle. Once they're happy with their share they'll start tightening the bolts in the name of profit maximisation.

Yup, that seems like a reasonable explanation to me too. Corporations are, as a stereotype, greedy (or their investors are), and they would not do something like this with no reason.

OpenAI's problem is that it's not at all clear where the profit is going to come from. Average users aren't going to pay a subscription just to make the occasional query.

I think GitHub Copilot is a good LLM-based product with a fairly large market. I'm concerned that it might be the only good product out there, still.

Apple seems to have a good strategy to me - Use an existing 3rd party until a better path emerges (ideally, the Apple silicon path).

How much capital would it take for them to run leading-edge inference for all of their customers? Why not outsource that risk until it's obviously worth taking on?

It's very similar to Sherlocking.

https://en.wikipedia.org/wiki/Sherlock_(software)

No it’s not. That’s something else. It’s similar to Apple Maps/Google maps

It's different but similar. In both cases Apple allows an application on their platform and after a while they switch the application for one of their own. In both cases they abuse the power of being a platform owner.

I get that Apple wants in on this hype, for the first time they felt they can't afford to come years later like usual. But still strange, afaik they never exposed a partnership / a third party company like this before for core functionality.

Google maps? Google search?

Is the headline possibly a reference to this Oatmeal comic?:

“Exposure” https://theoatmeal.com/comics/exposure

I doubt it's that comic specifically, as it's a way of doing things as old as time. Doing it for the clout, doing it for credit, doing it for (anything other than money) is not a new trend some web comic started.

I'm shocked someone like unOpenAI would make this kind of agreement though

Given that Google has to pay tens of billions of dollars to Apple for 'exposure', this seems like great deal for OpenAI.

That's not just exposure, though. Google fingerprints those users, prioritizes showing them sponsored content, and routes them to other ad-enabled Google services. Google pays so their exploitative and inferior experience can be default.

Finally a title so simple to answer what I’ve been too lazy to read into the details on but was curious about.

"exposure [of user data]"

Reply on news.ycombinator.com