高阶提示词工程原理，教学，总结

高阶提示工程

• 注：需要有一定的英文功底

1. Chain-of-Thought (CoT)

概念：CoT是一种逐步推理的方法，通过一系列连贯的步骤引导模型得出结论，就像解一道数学题时的步骤。

类比：想象你在解一道数学题，你需要一步步列出每个计算过程。每个步骤都依赖于前一个步骤，直到最终得出答案。

示例：

Prompt: “What is 15% of 200?”

Thought process:

1. “15% of 200 is calculated by first finding 10%.”
2. “10% of 200 is 20.”
3. “Then we find 5%, which is half of 10%.”
4. “5% of 200 is 10.”
5. “Adding these together, 15% of 200 is 20 + 10, which is 30.”

2. Tree-of-Thought (ToT)

概念：ToT是一种分支推理的方法，每个推理步骤可以生成多个分支，类似于决策树，使模型可以并行探索多个路径。

类比：想象你在探索一个迷宫，你可以同时尝试不同的路径，记录每个路径的进展，直到找到出口。

示例：

Prompt: “Find the optimal route from A to B in the city.”

Thought process:

1. “Consider the route via Main Street.”
2. “Consider the route via Second Avenue.”
3. “Evaluate traffic conditions and travel time for each route.”
4. “Select the route with the shortest travel time.”

3. Graph-of-Thought (GoT)

概念：GoT将推理过程建模为有向图，每个节点代表一个思维步骤，节点之间的边表示推理路径，从而增强模型的复杂推理能力。

类比：想象你在构建一个关系图，图中的每个节点代表一个任务或概念，节点之间的连线表示它们之间的关系和影响。

示例：

Prompt: “Analyze the impact of a new policy on the economy.”

Thought process:

1. “Identify key economic indicators affected by the policy.”
2. “Create a graph with nodes representing each indicator.”
3. “Draw edges showing how the policy impacts each indicator.”
4. “Analyze the overall impact based on the connections and interactions.”

4. Skeleton-of-Thought (SoT)

概念：SoT是一种先生成答案的骨架（主要要点），然后并行扩展每个要点的方法，大幅减少了解码延迟。

类比：想象你在写一篇文章，先列出主要段落的主题句，然后分别扩展每个主题句的内容。

示例：

Prompt: “Write an essay about the benefits of exercise.”

Thought process:

1. “Outline: Introduction, Physical Benefits, Mental Benefits, Conclusion.”
2. “Expand each point in parallel:”
   • “Introduction: Exercise is essential for a healthy lifestyle.”
   • “Physical Benefits: Improves cardiovascular health, builds muscle strength.”
   • “Mental Benefits: Reduces stress, improves mood.”
   • “Conclusion: Regular exercise leads to overall well-being.”

5. Algorithm-of-Thought (AoT)

概念：AoT通过结合搜索算法示例，克服CoT的局限性，增强探索和问题解决能力。

类比：想象你在编写一个程序，使用不同的算法来解决问题，每个算法提供不同的路径和策略。

示例：

Prompt: “Find the shortest path in a graph using AoT.”

Thought process:

1. “Consider using Dijkstra’s algorithm for shortest path finding.”
2. “Evaluate other algorithms like A* and Bellman-Ford.”
3. “Compare the efficiency and results of each algorithm.”
4. “Select the algorithm that provides the optimal path with minimal computation.”

总结论文表

• 想要更系统，专业，深入的了解提示词工程就离不开专业的论文，如果自行阅读困难，可以使用工具

    此处内容已隐藏，请评论后刷新页面查看.

   

  来辅助阅读，当然也可以不用并使用免费的 DeepL翻译等….。这个工具是目前最强的在线学术 AI之一，具备翻译，解释，类比，预测与评估等功能…