[ninjha01]

Ben Thompson of Stratechery raised a good question - why isn't Apple partnering with Microsoft instead of OpenAI? Microsoft has the weights, can provide the requisite infra, is in the process of derisking the OpenAI collaboration via defacto aquihires...

OpenAI is losing people, has a CEO with a mixed reputation, irritates a famous actress who beat Disney in court...

[gwern]

You're looking at it the wrong way: Altman really wants this deal. For a long time, accelerated by the OA coup, he has been searching for ways out of the MS trap, where OA becomes an appendix to MS, dependent on MS infrastructure completely. ('Above you, below you, around you...') Apple is one of the few viable alternatives to play off against MS. (This is the same as why Anthropic gets investment from Google & Amazon, as their anti-OA.) The mere fact of any Apple deal at all is a lot of leverage against Nadella. Think of it as like North Korea playing off China and Russia against each other; it is bad to have only 1 patron, but it is good to have 2.

EDIT: ah, there we go: https://www.businessinsider.com/satya-nadella-sam-altman-ope... Mission. Accomplished.

[ninjha01]

I understand the motivation from the OAI side. what does apple gain though?

The purported google deal made a lot more sense to me. OAI is terrified of google, and Apple could mitigate Pixel differentiating against iPhone. Google would get a bunch of training data as well.

[kromem]

Apple's AI strategy is edge computing, which is very smart given their processor advantage.

But edge LLMs for general purpose aren't there yet and won't be for some time, so a partnership with a leading centralized LLM provider makes a lot of sense.

Do simple low hanging fruit LLM work like actual autocomplete at the edge on the device, and kick more difficult workloads to the AI partnership.

If the tech plateaus your edge computing will eventually catch up and you won't need the partnership but won't have lost market share in the meantime, and if it doesn't you'll continue to split the workloads.

[ninjha01]

But this is equally true for partnering with Microsoft or Google. Why go for OAI over the other bigger, actually owning infra players?

[kromem]

One possibility is regulatory oversight.

Apple and Microsoft striking a deal is going to get more attention and review than Apple and a much smaller company in market cap and share.

[layer8]

Apple likely plans to use their own infrastructure [0], and why should they add a major competitor as a middleman?

[0] https://www.reuters.com/technology/apple-power-ai-servers-wi...

[talldayo]

> Apple will deliver some of its upcoming artificial intelligence features this year through data centers powered by its own in-house chips

Emphasis mine. I don't think there's any chance that means ChatGPT will run on Apple Silicon. It seems more likely that they will use it to power their internal CoreML/MLX-based research, anything more would kinda be a waste of money and electricity.

[ninjha01]

I can’t imagine apple has chip production scaled to the point it could support all the inference queries coming from all the iPhones

[c1sc0]

They don’t need to if they can process a significant chunk of the queries on-device. Llama3-level inference works fine on M2-level chips today and the M4 is already in a mobile device.

[consumer451]

Maybe branding/brand recognition? Outside of tech circles, it's likely that very few people know anything bad about Altman, compared to the hundreds of millions who know about "ChatGPT."

[ninjha01]

The brand recognition is surely flowing from the iPhone to ChatGPT.

They’ll have to rebrand siri though