4 ms·

I believe ChatGPT inserts invisible characters that watermark the text. I'm sure it isn't hard to remove them via some third party tool

• weird-eye-issue 4y ago
  No, this is utterly false. They were looking into doing watermarking using things like character frequency, choice of words, etc (it would effect the content itself). But it isn't implemented yet and something like inviable characters would be trivial to bypass
• jansan 4y ago
  If I paste an answer into a hex editor, there isn't anything suspicious. All space chars are 0x20, all characters are normal ASCII. I think no watermarking is taking place.
  • vezycash 4y ago
    >If I paste an answer into a hex editor, there isn't anything suspicious Pasting it in windows 10 or lower notepad should do the same thing?
  • danwee 4y ago
    It's not that kind of watermarking. See above comments
• throwuxiytayq 4y ago
  I can’t find any information that confirms that they’re doing it, just that there’s plans for it. It’s amusing that they’d think it’s possible to watermark plaintext; I think it’s investor posturing.
• danwee 4y ago
  Not invisible characters (that would be easy to bypass by not copy-pasting the text ChatGPT outputs). https://techcrunch.com/2022/12/10/openais-attempts-to-watermark-ai-text-hit-limits/ https://techcrunch.com/2022/12/10/openais-attempts-to-waterm...