The notion that Apple is losing in AI...

The notion that Apple is losing in AI because they're so privacy-focused and, therefore, don't have enough data to build large models is one I've been thinking about recently. " "" "Could Apple be onto something? " "" "Specifically, AI is primarily comprised of two phases of development: training and inference. For training, it's clear that size is critical, and all the top models get bigger and bigger. It's almost inverted for inference since the larger the model, the slower it will be, and humans aren't too patient! " "" "Apple is strategically focusing on how their end users will experience AI, likely in service of future device sales and, more importantly, service revenue. Lightning-quick inference is essential to deliver such value, and small, local models offer the cheapest way. " "" "Given our ability to distill massive models into small, use-case-specific packages, the ability to distill a custom model to be entirely run on-device is coming. Apple seems well positioned to deliver fantastic end-user experiences leveraging AI while maintaining its privacy ethos by letting other big companies amass giant datasets to train the foundational models they'll distill. " "" "With AI growing like it is, having an additional layer of differentiation between those innovating in the model creation space vs. those innovating in the model usage space makes sense. There is likely a benefit in strategic focus on one of the two major AI phases, but whether both would be best served by a single company (like perhaps Google) remains to be seen. Only time will tell!

---

Original post on LinkedIn