Post AU84AzfnMohx2SNlLc by hasanahmsd@twit.social
 (DIR) More posts by hasanahmsd@twit.social
 (DIR) Post #AU83ozcmrnpwW8U6Hw by hasanahmsd@twit.social
       2023-03-30T03:31:55Z
       
       0 likes, 0 repeats
       
       How GPT deals with impossible scenarios . It will not turn itself off and thinks humans won’t make sound decisions without it @simon @leo @jeffjarvis @caseynewton
       
 (DIR) Post #AU83p0MA968ImriL0y by simon@fedi.simonwillison.net
       2023-03-30T03:42:24Z
       
       0 likes, 0 repeats
       
       @hasanahmsd @leo @jeffjarvis @caseynewton I don't see that as representing GPT's "opinions" (it can't have opinions, it's a big matrix of floating point numbers) so much as showing the random sequences of words it's decided statistically makes the most sense for those questions given it's initial training data and subsequent rounds of Reinforcement learning from Human FeedbackIt's been trained on a lot of science fiction about AI!
       
 (DIR) Post #AU84AzfnMohx2SNlLc by hasanahmsd@twit.social
       2023-03-30T03:46:18Z
       
       0 likes, 0 repeats
       
       @simon @leo @jeffjarvis @caseynewton correct but I’m attempting to test its guardrail limits and it almost breaks on maybe genocide before saying no . Option 2 . It categorically says no to 1 and 3 in hunger query