From “giving instructions” to “expressing needs”
| Strategy Type | Definition and Objectives | Applicable scenarios | Example (for inference models) | Advantages and risks |
|---|---|---|---|---|
| Command-driven | Directly give clear steps or format requirements | Simple task,need to execute quickly | “Write a quick sort function in Python. and output should include comments.” | The results are accurate and efficient Limit the model’s autonomous optimization space |
| Demand-oriented | Describe the problem background and goals, and use the model to plan the solution path | Complex problems require autonomous model reasoning | “I need to optimize the user login process. Please analyze the current bottleneck and propose three solutions.” | Stimulate deep reasoning of the model Need to clearly define the boundaries of requirements |
| Blending Mode | Combine the requirements description with key constraints | Balancing flexibility and controllability | “Design a three-day tour plan for Hangzhou, including the West Lake and Lingyin Temple, and keep the budget within 2,000 yuan.” | Balance goals and details Avoid excessive constraints |
| Heuristic Questions | Ask questions to guide the model to think proactively (such as “why” and “how”) | Exploratory questions, requiring model explanation logic | “Why did you choose gradient descent to solve this optimization problem? Please compare it with other algorithms.” | Triggering the model’s self-explanatory capabilities May deviate from the core goal |
