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.
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.”
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
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.