Apple Intelligence – Bells and Whistles.

Apple replicates Copilot+ with privacy bells and whistles

• Apple has made its play in generative AI by picking to copy the best that its competition has to offer and rolling it up with a dedicated server system to ensure that its privacy and security are superior to anything else on offer.
• Apple Intelligence is a completely closed and exclusive hybrid AI system that will be integrated into iOS, iPadOS and MacOS that on rare occasions will refer enquiries to ChatGPT but this is not going to be exclusive to Open AI for very long.
• Apple Intelligence has three main components:
  • First, on device: which is an architecture that looks very much like Copilot+.
  • Apple did not explicitly spell it out, but it looks like there will be a run time on the device which contains a number of pre-trained small language models that have been optimised for a series of tasks.
  • These would be the ability to convert a sketch to an image, generate summaries, smart replies, generate emojis and so on.
  • The functionality that Apple is launching in iOS18 also looks very like Microsoft’s Recall where the system builds a semantic graph of the user based on his or her activity and data which is used to prime a model before a response is generated.
  • This graph is very Apple app-centric and unless 3^rd parties have fully plugged into it, the activities within their apps will not be in the semantic graph.
  • Microsoft has a similar problem but because so much more time on PCs is spent in Microsoft Apps and a few other productivity apps, it will be easier for Microsoft to ensure its system fully reflects the activities of the user.
  • The drawback with moving this on device is that only the A17 and M-series chips are powerful enough to support it meaning that anything less than an iPhone 15 Pro will not support this function.
  • Furthermore, I doubt whether this will be enough to trigger an upgrade cycle meaning that this is going to be a service available to a small fraction of the user base for some time.
  • This will also impact developer willingness to support it as only a small subset of the users will be able to access the functionality from launch.
  • Second, Privacy Cloud Compute: which is a server specifically designed using Apple silicon to run larger workloads remotely.
  • This is where Apple really made its differentiation as privacy and data security are issues that have arisen and so far, have not been well addressed by the emerging AI industry.
  • When a request is made that the on-device SLMs can’t handle, the runtime will select the relevant data from the semantic graph and use that to prime a model in the Apple cloud and then send the request.
  • None of this data will be used to alter model weights or be stored on the remote server and Apple has gone to extra lengths to reassure users by ensuring that these servers will be audited and certified by 3^rd parties.
  • This is where Apple is differentiating from its peers as most of the services it announced either already exist or have already been announced by its competitors.
  • Apple has cherry-picked competitor services it thinks its users will like most, added a sprinkle of entertainment (think Genmoji) and then put a layer of privacy on top that it thinks its users will value.
  • Third, public generative AI: where Apple has created the ability of its on-device system to make calls to the larger language models available on the internet.
  • Apple is starting with ChatGPT powered by GPT-4o, but it will add other services and models over time.
  • Any data sent in these directions will be considered lost as Apple will ask permission to share the data from the device with the cloud service on every instance.
  • This is going to become like search on iOS where the user can choose from a range of options when setting up the device which can be changed at any time.
  • Hence, although this is what most of the chatter has been about in the technology media, this is by far the least significant announcement that Apple has made.
• Apple Intelligence is generative AI implemented Apple’s way and it remains to be seen how popular it is going to be.,
• On the one hand, Siri is going to take a massive leap forward in terms of its performance but only a fraction of users are going to be able to use it at launch.
• It will take about 4 to 5 years to fully populate the iOS ecosystem and so the significance of this to Apple’s fundamentals in the medium term is low at best.
• Instead, this is an answer to the critics who have accused Apple of having nothing when it comes to generative AI services, but it remains to be seen if these offer a boost to the quality of the user experience in the Apple ecosystem.
• Worst case, this is a placeholder for Apple to ensure it is in the generative AI game that it can backfill as its offering improves.
• This was clearly not the big bang the market was expecting which is why the shares were weak yesterday but in the grand scheme of things, this makes no difference to the investment case for Apple.
• Apple remains expensive for a company struggling to grow and as such I continue to have no inclination to own it.