“Prompt Engineering” sometimes feels like persuasive penmanship in the guise of yet another trendy phrase in the AI space.
But last week, OpenAI released their official Prompt Engineering Guide with Six Strategies for getting better results. And it’s been a real game changer - these are really solid strategies with detailed tactics and examples and the traction with LLM developers has been incredible.
I created a playground with Streamlit where you can put in your original prompt and experiment with the different strategies to up-level your prompts.
How does it work
I created prompt templates for each strategy. The prompt template for each strategy takes in your original prompt and uses the tactics and few shot examples given in the guide to generate an improved prompt. The prompt templates are stored in an AIConfig (json-serializable config) so you can run the prompts against models like Gemini, gpt-4, gpt-3.5-turbo, PaLM, or a model of your choice.
Streamlit App: https://openai-prompt-guide.streamlit.app/
Github: AIConfig Repo