I can't make GPT-4 generate a wrong answer for many of these.

What is the author doing wrong when using GPT-4?