If you want to learn Prompt Engineering but don’t know where to find suitable learning materials, here are some high-quality resources I know of. They are constantly updated and you are welcome to add to them.
Prompt Engineering Guide promptingguide.ai This is an open-source learning resource for Prompt Engineering. It explains all aspects of prompt engineering step by step and is available in multiple languages. I also contributed a few pages of translations to the Chinese version. This site is suitable for a quick and systematic browse to get a general understanding, and you can return to it from time to time.
Anthropic's Prompt Engineering resources Prompt Engineeridocs.anthropic.com/en/docs/build-…//t.co/CG6hMfozth Anthropic's documentation website is quite good, and there is a very important part related to Prompt Engineering, which is mainly combined with its own Claude model. It can be regarded as the best practice of Claude: - Instructions should be clear and direct - Use examples - Chain thinking - Use XML to structure input and output - Set roles for Claude - Others Although the above best practices are aimed at Claude, they are basically applicable to most models, but there are also some differences to note. For example, Claude can have many examples and the effect is very good. This is because its context length is very long and it follows the instructions very well. The same effect may not be good when used on other models. In addition, Claude is specially fine-tuned for XML and supports it quite well, but for other models, XML may not be the best format.
Anthropic Prompt Library docs.anthropic.com/en/prompt-libr… It has a series of commonly used prompt libraries. By studying these prompt examples, you can roughly understand how to write prompts best for different scenarios.
Anthropic's Prompt Genedocs.anthropic.com/en/docs/build-…SGgNHx Anthropic provides a powerful prompt generator with very good quality. I often use it myself and then adjust it based on the results. However, please note that this prompt generator costs money, about $0.3 per time, and you need to register as an Anthropic developer to use it.
Anthropic courses github.com/anthropics/cou… Anthropic's open source Prompt tutorial includes four courses: 1. Anthropic API Basics Course - teaches the basic points of using the Claude SDK: obtaining API keys, operating model parameters, writing multimodal prompt words, streaming responses, etc. 2. Prompt Word Engineering Interactive Tutorial - step-by-step detailed guidance on key prompt technologies. 3. Google Vertex Version 4. Real-World Prompt Engineering Course - learn how to apply prompt technology to complex real-world prompts. In particular, the Real-World Prompt Engineering course is very good. It first reviews the best practices in its official documentation, and then combines several common application cases, step by step from wrong practices to common practices to best practices, with jupyter notebooks, so that you can see the execution results of each optimization solution. The several cases included are very representative: - Extracting structured information from medical reports - Generating summaries for customer calls - Customer service after-sales support
Anthropic Cookbook github.com/anthropics/ant… This is a code sample biased towards Anthropic developers, not just Prompt Engineering, but also some other content, such as Embedding, fine-tuning, etc. If you are a developer, you can find some related ones to take a look, otherwise it doesn't matter.
deeplearning.ai hdeeplearning.aiThis is a deep learning course taught by Professor Andrew Ng. It seems to be free. The course not only has teaching videos, but also an environment for running code. You can write code and execute it according to the course. Many of the courses are very suitable for AI beginners.
Generative AI for Everyone is a very suitable coursdeeplearning.ai/courses/genera…y understand generative AI
ChatGPT Prompt Engineering for Developers deeplearning.ai/short-courses/… An introductory tutorial video on Prompt Engineering that combines ChatGPT and several common cases
OpenAI's Prompt Engineering Learning Resources OpenAI has also released many high-quality Prompt Eplatform.openai.com/docs/guides/pr…ources. Prompt Engineering Guide https://t.co/gzS425xS5a Its Prompt Engineering document contains various best practices for using prompts. The six strategies summarized are very representative: 1. Strategy 1: Write clear instructions 2. Strategy 2: Provide reference text 3. Strategy 3: Split complex tasks into simple subtasks 4. Strategy 4: Give the model more time to "think" 5. Strategy 5: Use external tools 6. Strategy 6: Systematically test changes I have compiled them before:
OpenAI's Prompt examples platform.openai.com/docs/examples There are many commonly used prompts listed here, such as translation, doing math problems, interview preparation questions, explaining code, etc., and each prompt example not only has code, but also includes how to use the code to call
OpenAI Cookbook cookbook.openai.com OpenAI's Cookbook is a treasure trove for learning to call various OpenAI APIs. You can find codes for many common cases, which not only contain detailed instructions but also code examples.
Generative AI for Beginners github.com/microsoft/gene… Microsoft is also very active in releasing tutorials. They have a set of "Generative AI for Beginners" which is very good and has been updated. Now it has been released to V2 with a total of 18 lessons, and it comes with a Chinese version.
Prompt Engineering for Generative AI developers.google.com/machine-learni… Google also has a lot of AI tutorials, including quite a few videdevelopers.google.com/machine-learni…n to the above prompt engineering, there are many other AI-related tutorials, which you can find on its website.
NVIDIA Deep Learning Institute learn.nvidia.com/en-us/training… NVIDIA has made so much money selling graphics cards through AI, so it stands to reason that it should invest a little in AI courses. It does have an AI teaching video website, but most of the courses on it are paid.
IBM Technology youtube.com/@IBMTechnology IBM has a channel on YouTube called IBM Technology, which has many high-quality AI teaching videos. Each episode is not long and is very rewarding to watch.
Stanford CS25 youtube.com/watch?v=P127jh… CS25 is a Stanford course called "Transformers United" that teaches Transformers. It's now in its fourth edition, and each time they invite industry professionals to give lectures. For example, last semester, Jason Wei and Hyung Won Chung from OpenAI, and Loubna Ben Allal from Hugging Face were guest lecturers.
Deep Learning by 3Blue1Brown youtube.com/@3blue1brown 3Blue1Brown is known to many people for his easy-to-understand videos that make complex theories relatively easy to understand. He has recently been serializing courses related to Deep Learning, and has already reached the 7th lesson, all with high-quality Chinese subtitles.
CS50 pll.harvard.edu/subject/artifi… Harvard University's renowned CS50 course is free and open to the public. It also includes AI-related courses. Notably, they introduced an AI teaching assistant, depicted as a little rubber duck, inspired by the common rubber duck debugging method used by programmers.
freeCodeCamp.org hyoutube.com/@freecodecampThis is a YouTube channel that collects various development-related free video tutorials, many of which are AI-related teaching videos with good quality.
LangChain hyoutube.com/@LangChainLangChain's YouTube channel is well done, and it often posts some high-quality teaching videos. Although they are all centered around LangChain, apart from the Langchain part, many of them are universal, especially those related to RAG.
Andrej Karpathy youtube.com/@AndrejKarpathy Andrej Karpathy is the former co-founder of OpenAI and the former head of Tesla's autonomous driving. His YouTube channel has many professional AI-related teaching videos of high quality, such as teaching you to write a GPT from scratch. I heard that in some universities his videos are used as auxiliary teaching materials.
This is a temporary stwitter-thread.com/t/183048702913…date collection: