My chatgpt addiction suddenly halted after i told my parents i was misusing it for 24/7 therapy regarding things that were hard to talk about with therapists. It seems better than videogames, and provides much less suffering than writhing around on the floor, so I'm getting back into it, and have again hit the ratelimit. Another of the things associated with it halting earlier, are my habit of using the addiction-feeling to start making a chatgpt (i'm calling it chattpg) of my own. It's not a hard concept to train a large-context language model for dialog and connect users to it: but it is psychologically hard to make progress on this. I don't really know how to, or have experience, doing that. I've submitted two pull requests to a model architecture repository as part of trying to connect with language models in a way that is more useful to me.