3 comments

[ 3.6 ms ] story [ 15.1 ms ] thread
I think your point about research is valid, but once that is done and you have a rough understanding of a solution, you just want things to work. You dont want a bunch of ways to do one thing, you want a single opinionated solution. Part of the information encoded in these models is from trainers deciding which answer is best, and that is information itself which isnt necessarily online.
Everyone works in their own unique way! I think a big part of the promise of this sort of AI is that it can tailor itself to what you need :)
I would like an AI assistant that can leverage personal/work emails(via API), texts, phonecalls, photos (friends social media posts with me tagged), movies/music I like, spotify likes, netflix likes, browser history, health data, youtube history, etc. - to train on and answer (and be able to ask it) personal questions about me. When was the last time I played softball? Use it to determine interests and hobbies (ranked list)

This may be a security and privacy nightmare. However, I do believe that Google, Microsoft and more have already begun this process. So why should we not have detailed access of all combined accounts?