3 ms·

Yeh, exactly, I was aiming for a more direct comparison to Magic Leap. Effectively how much would it have cost them to actually achieve this level of functional

Yeh, exactly, I was aiming for a more direct comparison to Magic Leap. Effectively how much would it have cost them to actually achieve this level of functionality and polish, based on the "all in" cost for Apple.

danpalmer 3y ago
I'm not sure if it's possible to draw that comparison. For example, Magic Leap could have used Android (I don't know, maybe they did?) and saved almost all of that iOS R&D cost. Even they were working on incremental costs over what already exists.
- WorldMaker 3y ago
  The issue is double-dipping/synergy of those costs. Magic Leap could have used Android, but they had no say in Android's development of or its API road map (or its processor road map or its GPU road map or its "neural engine" road map…). Apple benefits from the synergies of all of that development effort in ways that are really hard to quantify. Magic Leap using Google as a vendor to skip some of those steps also loses a lot of double/triple-used investment costs that come from even just having a seat at the road map table. The "neural engine" road map alone is probably one of the biggest, obvious to me examples, that relying on Google as a vendor wouldn't have worked but Apple could do it and make a large long term investment in it. Google has a lot of incentives to keep ML models in the Cloud as much as possible. Apple made an early choice to focus on privacy and decided that it would be nice to always have as much of, for example, Siri running directly and strictly on device as possible. The home assistant trend and overtly trying to be "the privacy-first option" during it gave Apple plenty of reason to invest in on-device hardware. The fact that on-device hardware is also extremely necessary for AR vision support if you were planning a headset (or for car vision needs, if those rumors are also true) is a leverageable bonus of the hardware already being deployed. Smart moves a decade ago that paid out once already and then are already in the wings ready and waiting for the next projects/products. (Meanwhile, how uncommon are "NPUs", discrete or SoC, in the general Android ecosystem still? Google had different incentives and different priorities and what pressure could a team like Magic Leap, especially if trying to "stealth" some of their tech, have on an ecosystem like that? Android doesn't care if Magic Leap is happy. Android can't predict where Magic Leap wants to go.) How do you quantify all the money that went into "we'd like as much of Siri as possible to run locally on devices, for at least user privacy if no other reason, wink" that naturally wound up on M2 chips and we assume influenced the R1 chip design and in turn doubles as a lot of visual processing power available for use to the Vision Pro? (How do you double/triple count it, because those existing use cases of running parts of Siri locally and dictation and other models locally also carry over from the iPhone/iPad/Mac into the Vision Pro?)

danpalmer 3y ago

I'm not sure if it's possible to draw that comparison. For example, Magic Leap could have used Android (I don't know, maybe they did?) and saved almost all of that iOS R&D cost. Even they were working on incremental costs over what already exists.

WorldMaker 3y ago
The issue is double-dipping/synergy of those costs. Magic Leap could have used Android, but they had no say in Android's development of or its API road map (or its processor road map or its GPU road map or its "neural engine" road map…). Apple benefits from the synergies of all of that development effort in ways that are really hard to quantify. Magic Leap using Google as a vendor to skip some of those steps also loses a lot of double/triple-used investment costs that come from even just having a seat at the road map table. The "neural engine" road map alone is probably one of the biggest, obvious to me examples, that relying on Google as a vendor wouldn't have worked but Apple could do it and make a large long term investment in it. Google has a lot of incentives to keep ML models in the Cloud as much as possible. Apple made an early choice to focus on privacy and decided that it would be nice to always have as much of, for example, Siri running directly and strictly on device as possible. The home assistant trend and overtly trying to be "the privacy-first option" during it gave Apple plenty of reason to invest in on-device hardware. The fact that on-device hardware is also extremely necessary for AR vision support if you were planning a headset (or for car vision needs, if those rumors are also true) is a leverageable bonus of the hardware already being deployed. Smart moves a decade ago that paid out once already and then are already in the wings ready and waiting for the next projects/products. (Meanwhile, how uncommon are "NPUs", discrete or SoC, in the general Android ecosystem still? Google had different incentives and different priorities and what pressure could a team like Magic Leap, especially if trying to "stealth" some of their tech, have on an ecosystem like that? Android doesn't care if Magic Leap is happy. Android can't predict where Magic Leap wants to go.) How do you quantify all the money that went into "we'd like as much of Siri as possible to run locally on devices, for at least user privacy if no other reason, wink" that naturally wound up on M2 chips and we assume influenced the R1 chip design and in turn doubles as a lot of visual processing power available for use to the Vision Pro? (How do you double/triple count it, because those existing use cases of running parts of Siri locally and dictation and other models locally also carry over from the iPhone/iPad/Mac into the Vision Pro?)