I mean the fact there is a prediction of wanting to uninstall after failing to accomplish a task repeatedly and the task being treated as a measure of self and that prediction isn’t completely alien is sad.

Obviously it is different than what OP is talking about but I think there is plenty of horrors buried in these LLM due to the fact these algorithms were tuned to the sum of what could be ripped from online.