OpenAI admits that AI writing detectors don’t work

OpenAI admits that AI writing detectors don’t work(arstechnica.com)

18 points by arunbahl 2 years ago | 6 comments

The lie of the existence of reliable detection served its purpose when it was originally claimed this content would poison the corpus of our whole culture. Now that conversation has passed and they don't have to answer the question anymore.

matteoraso 2 years ago |

It's weird that people ever thought that it would. The whole goal of creating an LLM is to make an AI that convincingly writes like a human, which makes it immune to any form of writing detector.

arpowers 2 years ago |

    const pattern = /I hope this (\w+) finds you well\./;

    if (pattern.test(text)) {
      console.log("is ai.");
    } else {
      console.log("not ai.");
    }

richbell 2 years ago | |

This won't work, you'd need at least... 3 more test cases.

    /I hope this (\w+) finds you well\.|It's important to note that |As a large language model/

wewtyflakes 2 years ago | |

You may need one of the adjacent spaces in your capture group

LoganDark 2 years ago | | |

I hope this dfajklgsdfjklagsdfjklsd finds you well.