Task requirements and prompting strategies
-tx-
| Task Type | Applicable models | Tips focus | Examples (effective hints) | Tips to avoid |
|---|---|---|---|---|
| Mathematical Proof | The reasoning model | asks questions directly without step-by-step guidance | ”Proving the pythagorean Theorem" | Redundancy resolution (e.g. “draw the graph first, then write the formula”) |
| The general model | explicitly requires step-by-step thinking and provides examples | “Please derive the Pythagorean theorem in three steps, referring to: 1. Draw a right triangle…” | Ask questions directly (easy to skip key steps) | |
| Creative Writing | Reasoning models | encourage divergence, set roles/style | “write an adventure story in the style of Hemingway” | Overly constrained logic (such as “list in chronological order”) |
| General models | need to clearly define constraints and avoid free play | “Write a short story containing the words ‘quantum’ and ‘desert’, no more than 200 words” | Open-ended instructions (such as “free creation”) | |
| Code Generation | Reasoning model | simplicity requirements, trust model logic | “Implementing Quick Sort in Python” | Step-by-step instructions (e.g. “Write the recursive function first”) |
| General model | refinement steps to clarify input and output formats | “First explain the quick sort principle, then write the code and test the example” | Vague requirements (such as “write a sorting code”) | |
| Multi-round dialogue | Universal model | Natural interaction, no need for structured instructions | “What do you think the future of artificial intelligence will be like” | Forced logical chain (such as “answer in three points”) |
| The reasoning model | needs to clarify the dialogue goal and avoid open divergence | “Analyze from the three aspects of technology, ethics and economy The Future of AI" | Emotional questions (e.g., “Are you afraid of AI?”) | |
| Logical Analysis | The reasoning model | directly raises complex questions | “Analysis of the conflict between utilitarianism and moralism in the ‘trolley problem’” | Add subjective guidance (e.g. “Which one do you think is right?”) |
| The general model | needs to split the problem and ask questions step by step | “First explain the definition of the trolley problem, then compare the differences between the two ethical views” | Ask complex logic at one time |
