It reads to me that locking yourself / AI up in a room with only yourself to converse with causes it to go MAD / out of touch with reality. If you interact wit

It reads to me that locking yourself / AI up in a room with only yourself to converse with causes it to go MAD / out of touch with reality. If you interact with people / fresh data regularly you'll avoid that. I'm not sure how much is down to evolving requirements and data formats vs degradation of the training data. The idea sounds common-sense but it's also triggering a bullshit warning for me, because I'd expect more consistency per AI model, and differences to occur on new model interations - regardless of the dataset.