Mastodon Feed
Boosted by jwz:
davidgerard@circumstances.run ("David Gerard") wrote:
the big problem with that Adversarial Poetry versus AI paper is that they did half the work with chatbots and not humans.
they translated prompts into poems using chatbots.
then they judged how well the poems jailbroke the chatbots using other chatbots to do the judging?!
Looking over the work, I can see no reason it requires chatbots to do these jobs. This was gratuitous, lazy and should be a discredit to these researchers.