Post AcsbnPQVVnMRH20JOa by hasanahmsd@twit.social
 (DIR) More posts by hasanahmsd@twit.social
 (DIR) Post #AcsbnPQVVnMRH20JOa by hasanahmsd@twit.social
       2023-12-16T20:58:59Z
       
       0 likes, 0 repeats
       
       If this is true then the gpt 4 and 3.5 model have gotten dumber since launch . With jives with faster output , shorter outputs and less accurate outputs in order to save money . Sort of makes sense when Gemini pro hard answers are almost as good as gpt4 despite being a smaller model . @jeffjarvis @leo @simon
       
 (DIR) Post #AcsbnQPTr68MK8X0Qy by simon@fedi.simonwillison.net
       2023-12-16T21:38:35Z
       
       0 likes, 0 repeats
       
       @hasanahmsd @jeffjarvis @leo I can't remember where now but I've seen that 20bn number convincingly debunked, it was a mistake in the paper, not an accidental leak of a closely guarded secret