I wasn’t talking about privacy. Their AI “strategy” is a hot mess. Apple Intelligence is garbage.
If referencing the "Apple Intelligence" announcement, that really was not an AI 'strategy". That was mainly an attemp to capture AI hype in their stock price than any technical strategy.
Apple had a strategy toward this space before "chatGPT" became some mania thing. It was don't mention 'AI' as much as Machine Learning and apply ML to useful things. Part of the twist here was not Apple shift but the usurping of 'AI' to mean 'chatbot'. Those are no where near real equivalent (interchangable) terms.
Siri has been somewhat in the same boat as Alexa. Substantively large userbase so a radical change is just about as disruptive as potentially helpful. They also were composed in a context where being correct vast majority of the time was a desired feature. ChatGPT opened to the door a system being acceptable if it occasionally spits out complete rubbish from time to time because succeeds more often than it fails. IF Siri doesn't know is basically says it doesn't. ChatGPT will just blow smoke at you ( How many letter 'r' appear in Strawberry ... 2. )
Siri's problems are more so tactics that strategy. Apple has chased some stuff that probably sucked up more resources than should have been allocation. Does the world need yet another AI visual slop generator ? Probably not. They seemingly stop investing in classic Siri because new foundation was 'just around the corner' . They should have done both and skipping visual slop (take the visual slop money and keep making classic Siri improvements). That is more a tactical move than a strategic one.
Similarly the push to do more Siri processing on device when possible. That isn't a bad strategy. Making the most gigantic , maximal power consuming cloud 'brain' possible is disrupting the grid. It is a 'loop' the deep pocketed tech companies and funds are chasis because that also creates a high barrier to entry. Apple chasing to dump all their spare cash into jumping off that cliff isn't a good strategy. That is more of a training issue than a necessary inference issue. Apple was never going to 'own' training.
Now they’re desperately hoping Google will bail them out. First it was on device, then private cloud compute, now Google servers. They’re obviously just making this up as they go along.
A decent amount of that is based on privacy constraints/objectives that Apple has self imposed. Another aspect of that isn't shifting Apple strategy but 'scope creep' on the what is the minimal set of 'Ai" features they feel they need to deploy. There is cloud AI doing made up images. movies clips from amateur scripts . write my 30 page essay for class and don't get caught, etc.
There should be no particular end user difference between running on PCC or spilling over to Google Cloud service is there is a demand spike. Depending upon how long the 'porting' process is to PCC for the google model If the porting process is so long that Google is on to model n+1 by the time Apple+Google have ported model n then that would be a problem. There is next to nothing in these announcement that says that is the case one way or another.
The more times they port the more of the port process will get an 'coding expert system' applied to more aspects of the problem. The chance that the port stays really long over the very long term are relatively low. (asuming keeps Apple kept their PCC hardware competitive over time. )
Using Google's model. from a using 3rd party library; not a major shift. Apple doesn't exclusively write their own C compiler. The unix utilities used to help run the system's ... lifted from BSD Unix ... that Apple doesn't license peoples stuff to implement things is smoke. In the context of creating largest piled higher and deeper stack of models.... the corollary to that is suck in as much data from the world as possible. Apple cashes this search advertising revenue sharing checks but they don't do the work ( so can claim king of privacy). Substantially similar issue here.
They’re being reactive instead of proactive and it shows. They look like they are flailing and have no real strategy.
Chasing the media hype trail is almost always reactive.
Strategy.
Apple doesn't want to be in the mega hyperscale server business. No change in strategy here at all. Current iCloud web services are provisioned in part on those vendors across the world.
Device first and only walk processing up as necessary. No real change in strategy here at all. It also happens to be a lot 'greener' and scalable if on average the workloads are distributed in size.
www.nytimes.com
