[skeeter2020]

The analogy that jumps to mind immediately is "you get the wikipedia page by default, but have the option to explore the page's metadata (including conversations)." This feels reasonable, and how most people use wikipedia - which isn't surprising since you're often getting an LLM's output from the training on the exact same information. The complaints in the report seem to miss the point IMO, jumping to some form of artificial intelligence and applying it to what is text prediction that unsurprisingly reflects the body of human knowledge on which it is trained.