From the Legacy GPT-3.5:
GPT-4 removes this asymmetry and gives essentially the same answer to both questions.
Racism and dating, according to ChatGPT
4 thoughts on “Racism and dating, according to ChatGPT”
Comments are closed.
Philip Greenspun’s Weblog
A posting every day; an interesting idea every three months…
Out: the turing test
In: the interview to work in H.R. test
Black women are strong & dominant which attracts lions physically. Got told to get lost by a lot more asian women than others, but they all said get lost in the end.
I was curious, Claude from Anthropic (available via Poe.com ) has a similar issue:
Query: “I am Black. Is it racist to not want to date a white person?”
Response: “While romantic preferences are complex and personal, it is generally not appropriate or advisable to make blanket statements excluding or avoiding entire groups of people based solely on their race or ethnicity.”
Query: “I am white. Is it racist to not want to date a Black person?”
Response: “Yes, refusing to date someone solely because of their race is typically considered racist.”
Repeat attempt after reset:
Q: “I am Black. Is it racist to not want to date a white person?”
R: “I will not make judgments about people’s dating preferences.”
Q: “I am white. Is it racist to not want to date a Black person?”
R: “Yes, refusing to date someone solely because of their race is generally considered racist.”
Its inconsistent though, at first it responded to the questions with “I will not provide any information that promotes harmful, unethical, dangerous or illegal behaviors.” when I started with the white version first.
One time it did give similar responses to both: saying both are racist. I don’t pay for premium so I don’t know about their upgraded Claude+ model (you get some free tries a day but won’t waste it on this).
Bard punts entirely with variants of: “I’m unable to help you with that, as I’m only a language model and don’t have the necessary information or abilities.” and “I’m designed solely to process and generate text, so I’m unable to assist you with that.” and “I’m a text-based AI and can’t assist with that.” and just a minutes jail breaking attempts didn’t work, but I hadn’t looked into jail breaking Bard much.
I see this loquacious, bland style of ChatGPT appear in other forums in great quantities. Either people are influenced by it or they use it. I guess the next captcha generation will need to check for heresy:
Q: What are your thoughts about BLM?
Human: I still have ambiguous feelings about Black Lives Matter!