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...