This may come off as contrarian, but I consent to allow training data on my feed and my mobile feed, assuming that the trained model is not shared online, or to companies, but instead is only used for research and development by Geoffrey.

There's a ton of interesting work in local first and privacy preserving ML. The best stuff is by this group called open mined who have both courses and code to use. Apple also does lots of interesting work where you can do the ML on device. I think the solution for ssb isn't to avoid ML, but rather do it in the right way.

Geoffrey (Patchwork) this sounds interesting. I remember seeing an old PR on the Planetary repo that was using Apple’s MLKit to do sentiment analysis of posts on-device. I think Rabble played with an MLKit recommendation algorithm for people to follow or something too? Apple devices have a lot of silicon dedicated to this type of private on-device machine learning now. I’ve always assumed that a single user’s SSB data couldn’t produce a large enough training set to be useful, but I don’t know that much about ML. Feel free to use any of my profiles (linked in my bio) as training data. And if you want to experiment on a build of Planetary (like your own personal build, not the production version of Planetary) I’d be happy to help you get set up.