I tested GPT-5.2 and the AI model's mixed results raise tough questions ...