Today’s preprint paper has the best title ever: “Adversarial Poetry as a Universal Single-Turn Jailbreak Mechanism in Large Language Models”. It’s from DexAI, who sell AI testing and compliance ser…
  • blakestaceyAEnglish
    9·
    15 days ago

    AI preprints : peer review :: Scott Aaronson : cops

    • David GerardOPMAEnglish
      7·
      15 days ago

      hey hey hey!! there’s always making it a Conference Proceeding

    • lagrangeinterpolatorEnglish
      6·
      15 days ago

      You’d think peer review would make things better here, but big ML conferences have to deal with an absurd amount of submissions these days. NeurIPS this year got over 21000. The system they use for reviews is that anyone who submits a paper is required to review a certain number of other papers. So yeah, your ML paper is getting reviewed by other people who happen to submit their own papers. Who are competing with you to get their own papers accepted. Yeah, no problems there.