Posts by pa28@social.vivaldi.net
 (DIR) Post #AVYw2PVcSO7ppdaoro by pa28@social.vivaldi.net
       2023-05-12T00:48:19Z
       
       0 likes, 0 repeats
       
       @grammargirl the output of language models depends on the strength of the model for the text being generated. Every word produced is simply the highest scoring word for that place in the text. Wherever there is weakness associating, for example, an author with a citation the odds of getting it wrong go up. Increased model size hides but does not correct these mistakes.