【01 简介 introduction】ChatGPT Prompt Engineering 提示语工程师openAI官方课程 吴恩达 人工智能

This course, presented by Andrew Ng and Isa Fulford from OpenAI, serves as an introduction to leveraging Large Language Models (LLMs) specifically for software development through API integration. The instructors emphasize that while much of the existing literature on prompting focuses on one-off tasks within the ChatGPT web interface, the true potential of LLMs lies in their ability to help developers rapidly build scalable software applications. The curriculum is designed to cover essential prompting best practices, common use cases—such as summarizing, inferring, transforming, and expanding text—and the practical implementation of a chatbot.

A central theme of the course is the distinction between "base" LLMs and "instruction-tuned" LLMs. Base LLMs are trained to predict the next word in a sequence based on vast amounts of internet data, which can lead to unpredictable results when asked a direct question. In contrast, instruction-tuned LLMs are specifically trained to follow user commands and have been refined through techniques like Reinforcement Learning from Human Feedback (RLHF) to be more helpful, honest, and harmless. The instructors strongly recommend that developers focus on instruction-tuned models for modern applications, as they are safer, more aligned with user intent, and significantly easier to use.

To effectively utilize instruction-tuned models, the instructors suggest treating the LLM like a smart assistant who lacks specific context about a task. Success depends on providing clear, specific instructions rather than vague prompts. By defining the desired focus, tone, and providing relevant source material, developers can significantly improve the quality of the model's output. The course concludes by setting the stage for two fundamental principles of effective prompting: being clear and specific, and providing the model with sufficient time to "think" through complex tasks.