Did Apple just outthink everyone in AI? A chat with tech analyst Ben Thompson

Thompson’s main takeaway is that Apple is using its leverage as a phone maker to make AI companies — in this case, OpenAI — do much of the expensive and difficult work. And that Apple gets the upside of AI without incurring many of the costs and risks other tech players are taking on.

But it’s worth reading Thompson’s comments at length. The following is an edited excerpt of our conversation.

Peter Kafka: You seemed pretty enthusiastic about Apple’s AI efforts before they announced them on Monday. You seem even more so after the fact. What do you like about what Apple showed off?

Ben Thompson: I think what’s so compelling is actually how little they’re doing.

This entire moment is about generative AI. And I’m not sure people quite realize that Apple’s not really doing any generation.

They do have image generation, which is fairly tightly constrained. But [primarily] they are leveraging this capability to do interesting things that were not possible previously, like tying together different vertical apps.

I thought the most compelling demo was the presenter getting the message from her mom about her plane arriving, finding the dinner reservation that was in an email, getting information about real-time flight data, and incorporating that all together. It just resonates because that’s a real problem that people have.

It felt like Apple at its best: We’re not selling whiz-bang technology; we’re selling thoughtful solutions that are super-obvious once you see them.

I thought that really came through. And it sidesteps a lot of the challenges around generative AI. Like hallucinations, like getting stuff wrong. They’re basically just handing that off to OpenAI. And it’s going to be branded OpenAI. And you’re going to be [told] you’re going to OpenAI. And if it screws up? Well, go talk to OpenAI.

The reason why this is compelling, and why I was optimistic even before the presentation, is that it speaks to their position in the value chain. They own the interface where people conduct their lives. And that gives them the luxury of solving problems only they can solve.

And then having an interface for the folks that want to spend billions of dollars to do these huge large language models, to plug in and sort of take it or leave it — it’s Apple leveraging their position of being the trusted device in people’s lives, and getting everyone to dance to their tune.

Peter Kafka: You mentioned hallucinations, which people understand comes with AI. The Washington Post asked Tim Cook about them, and he said he couldn’t promise there wouldn’t be any. But it sounds like you’re saying the hallucination problems are going to be from the OpenAI queries. And they’ll be OpenAI’s problems.

Ben Thompson: I think the hallucination bit is maybe overrated. It’s something that’s very easy to sort of latch on to, like glue on pizza. It’s very funny and kind of embarrassing.

But if you go to ChatGPT.com, you go into it knowing it might come up with something weird. And actually, 99% of the time, it doesn’t.

And [OpenAI] is very confident, and I think it’s reasonable to be confident it’s going to continue to get better in that regard. And the fact of the matter is 100 million people or whatever it is, are still using it despite these dangers. Which speaks to the tremendous utility that does exist.

Apple doing what Apple does best

Peter Kafka: But back to the airport scenario. That sounds great, right? But what if — whether it’s a hallucination or a more basic error — I rely on Apple to tell me when I’m picking mom up, where we’re going, and they get it wrong? Isn’t the risk there much, much higher than a garden-variety hallucination?

Ben Thompson: Yes. I would say this is the No. 1 risk facing Apple.

I think there are two risks Apple faces. One is some massive AI breakthrough, such that the phone becomes the commodity, as opposed to Apple making the AI the commodity. Like: Who’s commoditizing who? That one is TBD and somewhat out of Apple’s control. But it’s worth mentioning.

The real risk is execution risk. Apple does have the luxury of coming to market later, and they benefited from a huge amount of research and improvements. Like shrinking down these models, giving them high efficiencies, so they can run on-device. They’ve had all those benefits.

What they are proposing to do — to actually orchestrate different apps and different bits of data — no one has done well, yet. Apple’s bet is they can do it well because they have the data, because they are on the device. But there is a real execution risk.

There is so much of AI that demos well and dies down on the edge cases. And there’s a gazillion edge cases. So to the extent this does not work out and does go badly, [the risk] is there.

Apple is not a fast follower here. They are a leader. And whether this is a place that they have the skills and capabilities to lead effectively is by far the biggest open question.

Peter Kafka: Do you take Apple at face value when they say this tech only works on our latest, greatest, highest-end machines?

Ben Thompson: 100%. The No. 1 constraint in running these models is memory. And every single device that supports this has eight gigabytes of RAM as a minimum. That’s just the long and short of it.

I think this is evidence, honestly, about how Apple was late. To develop phones, the whole process runs over several years. I think if they could go back in time, they would have made the base iPhone 15 have eight gigabytes of RAM [so it could run the new AI features].

Will Apple spark more iPhone sales?

Peter Kafka: Do you think this spurs device sales? That’s obviously the question Wall Street’s wondering about.

Ben Thompson: I think so. And I think it doesn’t just spur device sales. I think it potentially spurs higher ARPUs [average revenue per user]. I’m very curious if Apple’s going to actually ever start talking about RAM. Because RAM is the key constraint here. If you got a phone with more RAM, you could theoretically run a better model. Are they going to talk about that such that you don’t just upgrade your phone, but you actually spend more, to get a higher-end phone? There’s definitely the opportunity to do that.

Peter Kafka: You’ve been speculating about whether Apple is paying OpenAI, or OpenAI is paying Apple. There’s no real reporting out there about the arrangement between the two companies.

Does it matter in the end whether Apple is cutting a check for several billion dollars or getting a check for several billion?

Ben Thompson: I don’t think it matters. I think it’s more interesting as to what it says about the relative power in this value chain.

In my initial article, I assumed Apple would be paying OpenAI. After the keynote, and [seeing] the degree to which OpenAI was diminished — it’s just a link, basically, and they’re going to warn you every time you go there, and there was no Sam Altman onstage, and this was all Apple stuff — that made me feel like Apple could have done the exact same presentation without OpenAI. And just say if you want to do chatting, open an app.

Because of that, I think that they’re in a stronger bargaining position. And my guess is there’s just no money changing hands at all.

(Editor’s note: Following my interview with Thompson, Mark Gurman at Bloomberg reported that, as Thompson suggested, neither company is paying the other, though they both imagine the deal could generate revenue by getting iPhone users to sign up for paid services sold by OpenAI.)