RE: https://infosec.exchange/@josephcox/116731857107771061
tldr: they’re all using training sets derived from one specific training set (WildChat) generated by ChatGPT 3.5 that contains several generated stories about this character. It’s a feedback loop of synthetic training data.
