3 ms·

The only thing noteworthy about this is how small-scale this is, and that the perpetrators don't even bother/have the means to set up their own infrastructure.

chatmasta 2y ago
The noteworthy thing is that OpenAI is reporting it. They’re signaling that they are proactively monitoring and investigating this activity, and that they’re willing to work with federal agencies while self-policing their negative externalities. This is all part of an ongoing conversation with lobbyists about “safe AI,” and it’s ultimately done to show that OpenAI is making an effort to mitigate the risks that regulators claim it creates. But there’s also another signal, which is what they’re not broadcasting: “ChatGPT can be used for propaganda, it works in Persian too, and we’re happy to sell to the DoD.”
- lancesells 2y ago
  It's a good thing for them to report this, but for a company that decided against watermarking their output they are kind of complicit. I know they might lose their teens and college demographic if they do it, but if it's truly this world-changing tool that they claim than not watermarking is scraping the serial number off the guns they're selling.* * Maybe a shitty analogy
  - mondobe 2y ago
    How exactly would they watermark textual content?
    - rolph 2y ago
      with fonts and typos.
      - Something1234 2y ago
        With unicode characters. Greek question mark anyone instead of semicolon?
    - lancesells 2y ago
      https://www.theverge.com/2024/8/4/24213268/openai-chatgpt-text-watermark-cheat-detection-tool https://www.theverge.com/2024/8/4/24213268/openai-chatgpt-te...
- bfLives 2y ago
  > and it’s ultimately done to show that OpenAI is making an effort to mitigate the risks that regulators claim it creates. I’d go one step further — the subtext is that only Open AI is willing to do this, and that LLMs are just too dangerous to open source/not another with regulation that only big companies have the resources to adhere to.
- r3db34rd 2y ago
  [dead]

chatmasta 2y ago

The noteworthy thing is that OpenAI is reporting it. They’re signaling that they are proactively monitoring and investigating this activity, and that they’re willing to work with federal agencies while self-policing their negative externalities. This is all part of an ongoing conversation with lobbyists about “safe AI,” and it’s ultimately done to show that OpenAI is making an effort to mitigate the risks that regulators claim it creates. But there’s also another signal, which is what they’re not broadcasting: “ChatGPT can be used for propaganda, it works in Persian too, and we’re happy to sell to the DoD.”

lancesells 2y ago
It's a good thing for them to report this, but for a company that decided against watermarking their output they are kind of complicit. I know they might lose their teens and college demographic if they do it, but if it's truly this world-changing tool that they claim than not watermarking is scraping the serial number off the guns they're selling.* * Maybe a shitty analogy
- mondobe 2y ago
  How exactly would they watermark textual content?
  - rolph 2y ago
    with fonts and typos.
    - Something1234 2y ago
      With unicode characters. Greek question mark anyone instead of semicolon?
  - lancesells 2y ago
    https://www.theverge.com/2024/8/4/24213268/openai-chatgpt-text-watermark-cheat-detection-tool https://www.theverge.com/2024/8/4/24213268/openai-chatgpt-te...
bfLives 2y ago
> and it’s ultimately done to show that OpenAI is making an effort to mitigate the risks that regulators claim it creates. I’d go one step further — the subtext is that only Open AI is willing to do this, and that LLMs are just too dangerous to open source/not another with regulation that only big companies have the resources to adhere to.
r3db34rd 2y ago
[dead]