I feel like, with the rise of Jarvis and other derivates of OpenAI, we might benifit from a plugin that detects AI-generated content, as it starts to grow rampant on blogospheres like Medium. Does such a tool exist - and if not, would any of you guys be interested in creating one with me?
1 Like
Really good idea. Consistent with safe, non-misleading, accountable uses of AI. Unfortunately, I don’t have the technical ability to help build the plugin, but depending on your jurisdiction I might be able to guide you to relevant legal advice, if needed.
AI-generated writing (without human editing) follows some predictable patterns (triple rephrasing, oxymora, …) and a general underspecificity in word choice and description. It can’t be too difficult to set up some parameters and determine the chance that a text is AI-generated, which is, in fact, how Bunyip works (but based on GPT-2).
The intent would not be to watermark OpenAI’s content (or to limit creative uses of the engines) but rather to give users a tool to protect them against fake news, hoaxes, misleading information, bot speech in fora or whatsapp groups etc
1 Like
A Voight-Kampff test for text 
1 Like