What is Prompt Engineering?
Prompt engineering is a term used in the field of natural language processing, and refers to the process of designing and optimising prompts for Large Language Models (LLMs) such as GPT-3, BERT, FLAN-T5 or LaMDA. Good prompts are specific, descriptive, offer context and helpful information, cite examples, and provide guidance about the desired output/format/style etc. Prompt engineering often requires a combination of deep human expertise, machine learning techniques, trial-and-error, and of course patience!
While there are plenty of prompt engineering resources for image generation AI models like DALL-E or Stable Diffusion, this post will focus on LLMs only.
Prompt Engineering Playgrounds & Marketplaces
The OG playground by OpenAI aims to be the native way to test different models like
text-curie-001, and others, for variations in the basic parameters and attributes. It offers a few presets, but is not really intended to be a full-blown playground.
Cohere Playground is a visual interface to test Cohere's LLMs without the need for code. It is simplistic in nature, much like the OpenAI playground, but also allows you to play with embeddings and classifiers.
Promptable is a platform to create prompts, organise them into folders, evaluate them on datasets, and ultimately deploy them behind APIs for seamless integration into your projects. You can version your prompts and track changes to ensure you are always working with the correct version.
Comet's Prompt Playground allows you to iterate quickly with different prompt templates, and understand the impact on different contexts. You can keep track of prompts, responses, and chains, track prompt usage, and iterate with LLMs as part of an emerging LLMOps workflow system.
Everyprompt is another platform that allows you to create and deploy prompts in a sandbox environment. They offer models and datasets for fine tuning models in production applications, and statistics for deployed functions in their Pro plan.
FlowGPT is a platform to share, discover, and learn about useful ChatGPT prompts. The prompts are community contributions, with a leaderboard highlighting the top contributors on the platform. FlowGPT highlights trending prompts across a variety of categories, and allows you to instantly try variations of any given prompt using the ChatGPT API (you need the FlowGPT Chrome extension installed and configured).
PromptLayer is a platform for prompt engineers to visually manage prompt templates, track API requests, usage, and performance.
PromptBase is a popular marketplace for DALL-E, Midjourney, Stable Diffusion and GPT prompts. It allows you to search for top prompts, buy prompts, and even sell your own prompts (PromptBase takes a 20% cut). An eBay for prompts.
PromptHero is another marketplace for ChatGPT, Midjourney, Stable Diffusion, Openjourney and DALL-E prompts. While traditionally focused on image prompts, PromptHero has now jumped onto the ChatGPT bandwagon too.
Promptvine curates the latest ChatGPT products and prompts across a variety of personas e.g. artist, advertiser, storyteller, and more. Admittedly, the prompts are quite basic, and may require a fair bit of tuning for your specific use case.
Prompt Engineering Guides / Courses
The OpenAI cookbook includes sample Python code for completing common tasks with the OpenAI API. Popular topics include text/code writing, website/file Q&A, using vector databases for embeddings, best practices for fine tuning, and more.
Cohere offers useful tips and principles on constructing prompts that will help you get the best generative output.
Developed by Dair AI, prompt engineering guide contains latest papers, learning guides, lectures, references, and tools related to prompt engineering.
Built in partnership with OpenAI, this free course from DeepLearningAI covers the basics of prompt engineering for developers, enabling them to build applications using LLMs quickly.
Dair AI has also partnered with Sphere to develop a new paid course on prompt engineering for LLMs.
AI Proof: Generative AI Unleashed by Metacrafters ($)
Did I miss any? Hit me up on Twitter and let me know.