Key Principles
1.Model selection
• Prioritize the selection based on the task type rather than the popularity of the model (e.g., choose the reasoning model for a math task, and the general Model).
2.Prompt design
• Reasoning model: concise instructions, focus on the goal, and trust in its internalization ability. (“Just say what you want”).
• General model: structured and compensatory guidance (“Fill in what is missing”).
3.Avoid Mistakes
• Do not use “heuristic” hints (such as role playing) on the reasoning model, which may interfere with its logical main line.
• Do not “over-trust” the general model (such as asking complex reasoning questions directly and verifying the results step by step).
