Advanced Prompting Techniques | 进阶提示词技巧

Beyond basic instructions, these techniques unlock the reasoning and planning capabilities of LLMs.

超越基础指令，这些技巧解锁了 LLM 的推理和规划能力。

1. Chain of Thought (CoT) | 思维链

Encouraging the model to "think out loud" before answering. 鼓励模型在回答前“大声思考”。

• Zero-Shot CoT: Just add "Let's think step by step."
  零样本 CoT：只需添加“让我们一步步思考。”
• Few-Shot CoT: Provide examples of reasoning steps.
  少样本 CoT：提供推理步骤的示例。

💻 Code Example: Implementing CoT with Python | 代码示例：用 Python 实现 CoT

import os
from openai import OpenAI

client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))

def solve_math_problem(problem):
    """
    Solves a math problem using Chain of Thought prompting.
    使用思维链提示解决数学问题。
    """
    prompt = f"""
    Q: Roger has 5 tennis balls. He buys 2 more cans of tennis balls. Each can has 3 tennis balls. How many tennis balls does he have now?
    A: Roger started with 5 balls. 2 cans of 3 balls each is 6 balls. 5 + 6 = 11. The answer is 11.

    Q: {problem}
    A: Let's think step by step.
    """

    response = client.chat.completions.create(
        model="gpt-4",
        messages=[
            {"role": "system", "content": "You are a helpful assistant that solves math problems step by step."},
            {"role": "user", "content": prompt}
        ],
        temperature=0
    )

    return response.choices[0].message.content

# Example usage | 使用示例
problem = "The cafeteria had 23 apples. If they used 20 to make lunch and bought 6 more, how many apples do they have?"
print(solve_math_problem(problem))

2. Tree of Thoughts (ToT) | 思维树

For complex problems, explore multiple reasoning paths and backtrack if necessary. 对于复杂问题，探索多条推理路径，并在必要时回溯。

• Process:
• Decomposition: Break the problem into steps.
• Generation: Generate multiple options for the next step.
• Evaluation: Score each option.
• Search: BFS (Breadth-First Search) or DFS (Depth-First Search).
  流程：分解 -> 生成 -> 评估 -> 搜索。
• Use Case: Creative writing, complex math proofs.
  场景：创意写作、复杂数学证明。

3. ReAct (Reason + Act) | 推理与行动

The foundation of AI Agents. Combining reasoning with external tool usage. AI Agent 的基础。将推理与外部工具使用结合起来。

• Loop:
• Thought: "I need to find the weather in Tokyo."
• Action: get_weather("Tokyo")
• Observation: "Sunny, 25°C"
• Thought: "The weather is nice. I can answer the user."
• Answer: "It's sunny and 25°C in Tokyo."
  循环：思考 -> 行动 -> 观察 -> 思考 -> 回答。

4. Self-Consistency | 自洽性

Generate multiple answers using CoT and take the majority vote. 使用 CoT 生成多个答案，然后取多数票。

• Why: Reasoning paths might vary, but the correct answer should be consistent.
  原因：推理路径可能不同，但正确答案应该是一致的。