I’m getting strong Déjà vu 2024 vibes.
I feel like I’m back two years ago, awaiting Apple’s response to ChatGPT.
Apple never intended to have a “response” to the likes of ChatGPT, or any other LLM service. This is clearly obvious by Apple’s decision to not invest in huge AI data centers over the past few years. Google has spent almost 200 billion, OpenAI had plans to spend upwards of 1.4 trillion by 2030 (recently cut to 600 billion). Apple’s capex spending last year was $14 billion, that’s all their spending, not just anything related to AI.
Instead, Apple has concentrated their efforts (R&D) towards on-device LLM’s and they have actually succeeded in this area. Not only with their “local” foundation models, but also by including a neural engine on almost every device since 2017 with the release of the A11 Bionic.
I’d wager that Apple may have tried to shoehorn in a new set of “advanced” features based on LLM’s into Siri, and realized it didn’t work as well as they intended, so they scrapped it and had to revamp all of Siri. This is the reason for the long delay. With all the criticism of Siri, they can’t really afford to release a “new” Siri that didn’t live up to promises.
And the deal with Google to license their Gemini model, was their realization that forcing users to offload more complex requests to ChatGPT is/was a horrible experience. Gemini saves Apple a ton of money and buys them time to build up their own foundation model for their “Private Cloud Compute”, while also allowing Apple to provide a more seamless experience.