At first, the rumors of Apple negotiating with Google (and, apparently, OpenAI) to use their generative AI-capable LLMs (Gemini and GPT/DALL·E, respectively) don't make any sense. We've spent the last couple of months hearing that Apple is brewing something magnificent, the next breakthrough in AI, something only a company with the combined DNA of Jobs, Ive, Cook, and Srouji can manage. Something like the M1 chips, but AI. We even had a timeline: WWDC 2024, this June. It should've been the second coming of Siri, the OG AI assistant, a redemption arc she surely deserves.

While all of that might yet be true, here we are getting intel from Gurman himself that Apple is considering using external help from its biggest rivals to serve its users' generative AI needs. The consensus is Apple will use its own models via Siri to reliably provide a limited range of interactions, while say Gemini Ultra or something similar will power the use cases the model can't or, more likely, won't handle.

Now, tech X's (formerly a bird) reaction of doom and gloom is understandable: surely, this signals a white flag from the company that just can't compete on the same level as its competitors, right? It's not the first time Apple has given up on a project years into making, like the heavily rumored "Apple Car" or the actually announced "AirPower" wireless charger. But I don't think that's the case here. Just hear me out.

Apple is famous for refusing to release anything that might be considered half-baked. I know some people think Vision Pro or the OG Watch were half-baked, but I disagree: those category-defining gadgets were pretty good for first-generation products and, more importantly, were completely usable and reliable for their intended use cases. Same with the OG iPhone, iPad, etc.

Now, even the more bullish AI evangelists among us (looking at the mirror now) have to admit that our best LLMs at the moment aren't there yet. They are unstable, unreliable, extremely biased, and sometimes outright dangerous. Also, they're slow and power-hungry. Half-baked is an understatement, in this case, they're at a tech preview stage. Now, even in this state, they might be extremely useful for some use cases, but they're not ready to be an Apple product.

But Apple needs a generative AI solution for those use cases, and they need it fast—like, this year fast. Now, they can train something good enough in a couple of months (Grok was born in 4), and considering the number of AI companies Apple has acquired in the last couple of years, they probably have models way better than that. But even if they reach the ballpark of Gemini Ultra, GPT 4.5, and Claude-3, it's still not good enough. Not for Apple at least.

So what do they do? They do them, obviously, and train a highly controlled and reliable subset of an AI model and ship it as the new Siri, while outsourcing everything else to a 3-rd party model or 3. This way, anything stupid that happens while generating content with AI, like black Founding Fathers and Asian Nazis, will be blamed on said 3rd party, while Apple will have the most reliable AI assistant on the market for once.

Touché, Mr Cook.

P.S. The image is generated by Dall-E.