Apple Intelligence
I think I have been pretty vocal what I think abut companies pushing "AI" into every single product they can find. And also about the practices how some of the companies training and selling LLMs operate.
To be blunt, most of it is a dumpster fire of missing security, lack of controls and absence of quality assurance on top of a ton of copyright infringement. You can obviously not generalise the whole industry, but we have seen enough of the above examples to raise questions about the models being used and operational practices.
Apple published two documents absolutely worth reading. Introducing Apple’s On-Device and Server Foundation Models and Private Cloud Compute: A new frontier for AI privacy in the cloud.
The security research article does an excellent job showing how hard it actually is to build secure infrastructure. And it appears Apple took many of the lessons they learned building security into enduser products and applied it to their cloud.
One of the most interesting paragraphs beside the on-device metrics in the machine learning article is the one about the data used to train their models.
We train our foundation models on licensed data, including data selected to enhance specific features, as well as publicly available data collected by our web-crawler, AppleBot. Web publishers have the option to opt out of the use of their web content for Apple Intelligence training with a data usage control.
Not a fan of the crawler being opt out (but I understand the reasoning behind it), but it seems like they base it on licensed data. Which is a pretty good step in the right direction.
From what I have seen so far Apple is doing a lot of things better than others. Better in this context still does not mean good in general. I also would have hoped for some insights into the environmental impact as Apple is known to stress this talking point whenever they can.
In my opinion it is a bit too early to judge the final product and I am not sure how forced Apple felt to join the AI playing field right now. Public opinion is already poisoned by those who came before them, so their standing is rough from the get go. But historically they also rarely let the market force them to release anything they did not consider "good enough".
The big differentiator so far seems to be that Apple understood that LLMs will show their strength augmenting day to day activities. Not changing or replacing them. Make common tasks easier and faster to execute is far more realistic and useful than trying to make the little cat picture viewer in our pockets intelligent enough to operate independently on our behalf.