OpenAI reasoning models: Advice on prompting