I worked on some next next gen AR glasses, I saw and used both orion and the prototype version of the rayban meta display glasses

Orion suffered from the problem that plagued it throughout its production: shifting goals and poor product management. The display team promised and then didn't deliver, twice.

One of my friends showed me a demo in meta display glasses (the production one) that allowed you to take a youtube recipe video, feed it into the dipshit machine and it'd spit out a step by step guide for you to follow. That was really nice to use.

The demo I worked on was taking some research glasses and attaching SLAM, always on audio and basic VLM descriptions of what you are looking at and dumping it into a database. You'd be surprised at how useful that context was. at the end of the day you could say "In the meeting I had with x what were the actions" Sometimes it even worked. Because everything that you saw or did was geocoded (ie it knew which room you were in) and linked to a calendar, even though facial and voice recognition wasn't allowed, you could get something akin to a decent executive assistant.

Making it secure and private is a big fucking challenge. I put in face and screen blurring, but thats not really enough. There is OCR based on eye gaze, which is hard to censor. The problem is there is more than enough context stored to infer lots of things.

The problem is not just securing the storage, its finding a decent sharing mechanisms that allows you to perform normal actions without asking a million questions or over sharing.