I used OpenAI's best practices and optimizer to rerun my GPT-5 tests. The results were strange, inconsistent, and sometimes bizarre, raising real concerns about how much developers can trust this AI ...