Imagine a world where low information sorts interpret a sampling of possible hi-res reconstructions from low-res security videos as ground truth. That to me is far scarier than the OpenAI and MIRI fear-mongering about GPT-2.