I am 100% convinced that the closed source model developers, like Anthropic, purposely instruct their models to hallucinate and/or partially ignore instructions so that you need to repeatedly reprompt them in order to increase token burn, thereby pumping up their revenue.
72โค๏ธ17๐2๐ค2๐1๐1๐ฏ1