OMG, I'm reading papers again. What's interesting is that until a few years ago it very much sounds like scientific business as usual. Even the GPT-2 paper "Language Models are Unsupervised Multitask Learners" sounds relatively level headed.