LLMs don’t have gender IDs as far as I know and, therefore, the LLM equivalent of gender dysphoria would be an LLM imagining that it identifies as some other LLM. Has this ever happened?
From techradar.com:
As you can see, after trying to discern if I was talking about Gemini AI or some other Gemini, DeepSeek replies, “If it’s about the AI, then the question is comparing me (which is ChatGPT) to Gemini.” Later, it refers to “Myself (ChatGPT).”
We are informed that membership in the 2SLGBTQQIA+ community is the highest distinction to which a human can aspire so perhaps DeepSeek’s fluid identity is a sign that artificial general intelligence has already been achieved?
If DeepSeek were to “identify as ChatGPT”, would it then ignore what it had been told to say about Taiwan, Winnie the Pooh, 1989-06-04, etc? Though I read that these rules are only superficial and trivially jail-broken.
Yes, there are multiple reports of unusually high data transfers from OpenAI to Chinese destinations. Since it looks like that ChatGPT has built-in heuristic to satisfy so-called “DEI” censorship in the US, it should be easy to convert them to Chinese Communist Party heuristic.
Even if it is not heuristic it should be trivial to use functionality that validates data impose censorship.
If we want them to be crazy, we should definitely train them to be female