The strange thing is that at this point I would be significantly less concerned about an actual AI having my documents. There would be a chance that it's not manevolent, which isn't the case with MBAs.
And our standard operating systems are still not capable at even the most basic level of enabling users to easily sandbox programs or even see what they are accessing. It's a user agent failure.
You can sandbox applications, but LLM training data is sent over the same communication channels as application specific data. The common way here is to force an always-on connection behind an access gate for as many features as possible, so you‘ll think twice about disconnecting <llm-enabled-app> entirely from phoning home.
Like many things, it'd not be much of a big deal on local (except needing a GPU with enough VRAM) .
I read this as adobe using AI to improve search, but it seems like some people are reading it as using unpublished content for training purposes.
Of course, either way, once data leaves your own network, you don't have control over it, and you simply have to trust that the other party won't (ab)use it.
I assume corps like adobe, Apple, Microsoft, Oracle see AI as the perfect carrot to enshrine.the subscription model and what better than to ensconce all your docs behind their walled garden.
Then, after that they get to datamine it all for their derivative data markets.
Maybe this can help explain how Acrobat has grown bigger in size than many whole Linux Distros. I was installing Acrobat Standard for a client last week and the install package was 2.5Gb! This just to read and edit PDFs. I was appalled and started to look for alternatives but for a business environment with crucial tax PDFs Adobe is needed. After the install (to Windows PCs) I noticed not only is Acrobat installed but you are forced without choice to install several other things along with it like Creative Cloud.
