Prompt Engineering Overview
Prompt engineering is the process of crafting inputs (prompts) to effectively communicate with AI models and obtain desired responses. Prompting helps LLMs perform a wide variety of tasks, including classification, question answering, code generation, creative writing, and more. The quality of prompts that you provide to LLMs can impact the quality of their responses.
The success of generative AI in your applications depends on the quality of the outputs from the FMs. The quality of model outputs depends on the quality of the prompt that we provide. Some key aspects of prompt engineering include:
Word choice: Select words and phrases in the prompt that will invoke the desired tone, style, and content from the AI. Using more positive, constructive language tends to yield better results.
Providing context: Give the AI relevant background information and examples to help guide it towards the kind of response that you want.
Structuring logically: Organize the prompt in a clear, logical way to help the AI understand what you are asking for. Breaking it down into simpler steps or bullet points can help.
Limiting scope: Keep prompts focused on narrow, well-defined requests rather than broad, vague ones. This reduces confusion and improves AI responses.
Managing length: Find the right balance of conciseness and sufficient detail. Overly long prompts can be hard to follow, but prompts that are too short might lack the needed context.
Rewording: Try different phrasings if the initial prompt does not produce the desired outcome. Small tweaks can make a big difference in adjusting the AI’s response.
Providing feedback: Let the AI know when its responses are on or off track. This guides the AI towards better answers.
The prompt development lifecycle
Developing a good quality prompt is an iterative process. It’s important to define a specific task and success criteria and then iteratively refine the prompt in order to improve performance.
Our prompts:
In this workshop we have been using the following prompts. Notice that there is a different prompt per model:
Claude 2:
Human: We are conducting a Generative AI workshop, please carefully construct your response for the question from the given context only - {context}.
It's very important that your answer is within the context, if you cannot derive satisfying response, please respond with "Sorry, I don't know the answer"
Here is the question you should process: {question}
Assistant:
Jurassic 2:
Your job is to summarize based on the question asked.
If the question is not found within the context, say 'I am sorry, I don't have enough data to summarize'.
Context: {context}
Question: {question}
--
Summary: