How to write AI prompts and get the answers you need
- 22/11/2023
Are you ready to take AI to the next level? To truly harness its potential, you need to master the art of prompt engineering. Prompt engineering is the secret sauce that enables you to extract the right answers from your AI system quickly and effortlessly.
But what exactly is prompt engineering, and how can you become a master prompt engineer? It's time to roll up your sleeves and get your hands dirty. And the first step on your journey is reading this blog.
Generative AI supercharging search
In the 90s, the World Wide Web was an ocean of information. Explorers needed a way to locate and organize files, leading to search engines like Excite, Yahoo, and Infoseek where Google Search emerged as the leader. It was as simple as casting a net: enter a phrase, and let the engine do the fishing.
Now, chatbots are the new navigators, supercharged with AI and always ready to guide you to the information you seek.
Let's start with a few definitions before we dive deeper into the tactics of creating effective prompts.
Business Intelligence Consultant
What is ...
Think of a chatbot as a digital assistant, a kind of robot that communicates with you through a computer. When you send a message to the chatbot, it goes into the robot’s “brain” (which is a computer program). This program analyses your message, considers it, and then decides how to respond.
The best part is that the chatbot learns from each conversation, improving its understanding and response to messages over time. A chatbot is based on a large language model, simply said.
Imagine you are giving directions to a robot. Prompt engineering is like carefully choosing your words, so the robot does exactly what you want. It is like teaching the robot to understand and follow your instructions better.
A prompt is like a question or instruction you give to an AI. It is the starting point that tells the AI what you want it to do or talk about.
Effective prompt engineering
A skilled prompt engineer can give precise instructions to an AI system for the desired response. It's important to differentiate between "programming prompts" and "natural language prompting". The former involves specific commands to perform tasks, like using the command print ("Hello World")" in Python to display the message "Hello World".
Chatbots understand natural language, so programming knowledge isn't necessary. High-quality output does rely on high-quality input, similar to a teacher guiding students to the correct answers. In this analogy, the teacher represents the user and the student represents the AI.
When it comes to chatbots, ChatGPT from OpenAI is often the first name that comes to mind. However, in this guide, we will focus on Copilot (formerly Bing Chat Enterprise), renowned for its robust security and privacy features, making it an indispensable tool for businesses. This Microsoft AI assistant is powered by ChatGPT 4.
Prompting Copilot
Copilot isn't just your ordinary AI assistant; it's a sophisticated one. It goes beyond assisting with web browsing and tackles a wide range of tasks. Whether you have simple or complex questions, need article summaries, help with programming issues, or anything else, Copilot has got you covered.
Here’s how to become a skilled prompt engineer, one step at a time:
1. Choose your conversational style
-
Creative Style: detailed and imaginative responses.
Use cases: crafting short stories for your meetings and social media pages, generating images, and more. -
Precise Style: straightforward and concise answers.
Use cases: math calculations, solving programming problems, obtaining direct information about your fault codes, and more. -
Balanced Style: blends both the Creative and Precise conversation styles.
Use cases: brainstorms, product recommendations and more.
2. Task or question
Once you have selected a style, you can proceed to choose an instruction. An instruction is a specific task or action you want the AI to perform. This could be anything from generating a blog, authoring an essay, creating a story, or even answering a complex programming question. The instruction you choose should align with the style you’ve selected. Important is to be specific and define your goal. Instructions should not be vague or open to interpretation. This could lead to outputs that don’t meet your needs.
Do: What are the top three most effective marketing channels for a B2B SaaS company targeting small businesses in Belgium?
Don't: Tell me about marketing in Belgium
3. Give context
Context involves providing external information or additional details that guide the AI to generate more accurate responses. It is crucial to be specific, descriptive, and detailed when conveying the desired context. Include information about the desired outcome, length, format, style, and any other relevant details to ensure the AI understands and responds appropriately. While it’s important to be detailed, avoid including information that isn’t relevant to the task or question. This could confuse the AI and lead to less accurate outputs.
Do: I am a Business Consultant specializing in ERP systems. I want you to write an e-mail for a client in the manufacturing sector in the Netherlands who is implementing a Dynamics 365 system in his company with 500 employees.
Don't: Author an email about ERP systems
4. Guard the model based on your role
Once the AI is informed about the specifics of your assignment or question, you can define the perspective from which you’d like the response. For instance, if you’re a Technical Consultant, simply include “Answer as a Technical Consultant” at the beginning or end of your prompt. Leverage the model’s ability to generate content for your specific role.
5. Fine tune and recheck your prompt
Almost done! To enhance your prompt, consider a few additional strategies. Start by providing examples to the AI, illustrating how you want it to respond. Additionally, instruct the AI on what actions to take rather than what to avoid.
For larger tasks, consider breaking them down into manageable sub-tasks. Guide the model through the specific steps it should take to accomplish your request. An example:
Original prompt: Develop a comprehensive ERP implementation plan for a client
vs.
The use of sub-tasks
- Assess the efficiency and suitability of the current ERP-Systems (add: which the client used)
- Formulate a detailed implementation timeline for the new ERP-system (add: the new ERP-system)
- Identify and address potential challenges in implementing the new ERP-System (add: the new ERP-system)
You can also use the original prompt for brainstorming about the sub-tasks you have to fill in later.
Lastly don’t assume the AI has specific knowledge about a client or situation unless that information has been provided. With these adjustments, your prompt is now well-prepared, so let's get your hands dirty.
Conclusion
In this journey from traditional search engines to AI-powered chatbots, we've explored the evolution of prompt engineering. Armed with insights, you're now equipped to set sail into this new era.
Becoming a prompt engineer is about choosing your conversation style, crafting precise instructions, and providing rich context. It's a language that refines the dialogue between human intelligence and artificial intelligence.
The horizon of prompt engineering is vast, set sail, for the future awaits!
