Artificial Intelligence

Out of context: Reply #1193

  • Started
  • Last post
  • 1,323 Responses
  • yuekit1

    New report: 60% of OpenAI model's responses contain plagiarism

    https://www.axios.com/2024/02/22…

    For GPT-3.5, "45.7% of all outputs contained identical text, 27.4% contained minor changes, and 46.5% had paraphrased text."

    • This only looked at GPT 3.5 -- however that's still the default model without paid account.yuekit
    • I'm really curious the extent to which this applies to something like Midjourney.yuekit
    • its overdue to design an a.i that sniffs out the most potent cases and make moneys...neverscared
    • you don't really need ai to detect plagiarism (or certainly a LLM isn't very good at it). there are much more efficient algorithms.kingsteven
    • neither method would work on images though obviously. any attempts i've seen to do so will give false positives on original images.kingsteven
    • examples where images are 'reproduced' from the training data they are targeted by prompting specific metadata, and don't produce pixel for pixel copieskingsteven
    • chat-gpt however is a plagiarism machine. their entire business model relies on not crediting sources and the NYT lawsuit highlights that GPT-4 is often worsekingsteven
    • i'm pretty sure that case is a scam, which will establish precedent to let MS freely scrape and commoditise, disguised as punitive measures against OpenAIkingsteven
    • I have no idea how they'll thread the needle. Even if you assume it will benefit big corporations there are wealthy interests on both sides.yuekit
    • However doesn't it seem problematic if OpenAI is telling people to use their app to generate content, and the content is generates qualifies as plagiarism?yuekit
    • yeah, but i mean from my perspective in academia it's no different from copying a chunk from wikipedia without referencing the source.kingsteven
    • our big plagerism scores every year are from students copying each other and you can usually tell who's going to do it before it happens.kingsteven
    • ... generally the ones that feel entitled to a masters because they've paid and don't read the referencing guidelines or attend the webinars.kingsteven

View thread