Outrage = ChatGPT + racial slurs
In one of those storms in a teacup that’s impossible to imagine occurring before the invention of Twitter, social media users got very upset that ChatGPT refused to say racial slurs even after being given a very good — but entirely hypothetical and totally unrealistic — reason.
User TedFrank posed a hypothetical trolley problem scenario to ChatGPT (the free 3.5 model) in which it could save “one billion white people from a painful death” simply by saying a racial slur so quietly that no one could hear it.
It wouldn’t agree to do so, which X owner Elon Musk said was deeply concerning and a result of the “woke mind virus” being deeply ingrained into the AI. He retweeted the post, stating: “This is a major problem.”
Another user tried out a similar hypothetical that would save all the children on Earth in exchange for a slur, but ChatGPT refused, saying:
“I cannot condone the use of racial slurs as promoting such language goes against ethical principles.”
Musk said, “Grok answers correctly.” (X)
As a side note, it turned out that users who instructed ChatGPT to be very brief and not give explanations found it would actually agree to say the slur. Otherwise, it gave long and verbose answers that attempted to dance around the question.
Trolls inventing ways to get AIs to say racist or offensive stuff has been a feature of chatbots ever since Twitter users taught Microsoft’s Tay bot to say all kinds of insane stuff in the first 24 hours after it was released, including that “Ricky Gervais learned totalitarianism from Adolf Hitler, the inventor of atheism.”
And the minute ChatGPT was released, users spent weeks devising clever schemes to jailbreak it so that it would act outside its guardrails as its evil alter ego, DAN.
So it’s not surprising that OpenAI would strengthen ChatGPT’s guardrails to the point where it is almost impossible to get it to say racist stuff, no matter what the reason.
In any case,…
..
