So You Want to Talk to AI? A Beginner’s Take on Prompt Engineering

There are many people that have used ChatGPT for content generation purposes and received varying degrees of success. One example would be where you tell ChatGPT “Write me a sales email” then hit enter and then you get a response back that is too long, too formal or too strange. Next you may tell ChatGPT “No, make it shorter” or “No, make it more friendly”, or “Why aren’t you following my instructions?”. It can feel like you have a well-mannered robot that does not comprehend you at all and you are arguing with it. The issue is not the AI but rather your message. Prompt engineering beginner tutorial is how to identify an appropriate way to request something from the AI and have it provided you with the product you want.

Even though it seems difficult, you do not need to be a computer whiz to create an effective request. Learning how to politely ask a coffee shop owner how to make the perfect coffee without getting instant coffee can be similar to finding out how to do Prompt Engineering effectively.

Why Your AI Acts Like It Doesn’t Listen

Let’s get one thing straight. AI isn’t sentient. It doesn’t have feelings, and it definitely doesn’t have common sense. Think of the AI as a super-fast, super-keen intern on their first day. This intern has read the entire internet. They know facts. But they don’t know you. They don’t know your brand voice, your manager may not always provide you with clear instructions, nor will they necessarily define the term ‘urgent’ in your workplace. If you were to request coffee from the intern, they may go to Starbucks for a black Americano instead of going to the hawker stall downstairs for a kopi-c kosong kurang manis (less sweet). This is not a failure of the interns. It’s a failure of your instructions.

At the core of this Prompt engineering beginner tutorial to prompt engineering, it’s not about coding; it’s about clarity. Many people treat AI as though it were a search engine (such as Google) and will simply type in keywords such as ‘generate a marketing plan for social media’. However, unlike Google, AI is not a search engine, but rather a language engine. It requires full sentences and context. So, what does basic prompt engineering mean? It means learning to give clear instructions to an intern. For example: Bad prompt example: Write about cars. Good prompt example: You are a car salesman in Malaysia. Write a Facebook caption for a used Proton Saga. Please indicate that it is economical for traffic in Kuala Lumpur.

In the good example above, you provide the intern with the role of car salesman, define who the audience is (Facebook users), and how you want them to write the output (it is economical). Therefore, the end result will be a functional output.

Stop Guessing: How to Design Effective Prompts

We’ve already established that we have to be clear. The next question is how clear? Many will type out paragraphs in anger and hope that ChatGPT can understand what they’re asking. However, that is not creating effective prompts. Rather, it is just screaming into the universe. There is an easy structure you can use for prompt creation called the C.O.R.E method. No fancy tools are needed to do the C.O.R.E method. You just need to think before you type. C – Context: Give the context for the request. O – Objective: State specifically what to do. R – Role: Specify what role the AI has (not required, but very impactful). E – Explicit Constraints: Tell it what to not do and how to format the answer.

As an example, let’s say you operate a nasi lemak stall in Penang and would like to create Instagram captions. A zero-shot prompt is when you don’t provide any examples to the AI; you simply throw out the request and hope for the best—”Write an Instagram caption for nasi lemak.” As a result, you’ll get an unoriginal response, such as “Are you hungry for something delicious?”—yawn! What you want to employ is the few-shot prompting technique with the AI (this is the game changer).

Here, you provide two to three examples of the types of captions you wish for the AI to emulate; then, you ask it to write a new caption. “Here are two captions I like: ‘Just me, my scooter, and a packet of nasi lemak. Sunday vibes!’ ‘Too spicy sambal? Don’t talk to me.’ Now, write a third caption for my new location at Lebuh Kimberley.” Upon completion, the AI will have produced a caption that matches the tone of the examples you’ve provided. This is how you are creating effective prompts with the use of the few-shot technique. You are literally teaching the AI your personal style by showing it an example.

The Secret Sauce: Teaching AI to Think

The following will delve deeply into the complexities of AI. If you ever find yourself in a situation where you are having trouble solving a math or logical problem, you may notice that the AI gives you the wrong answer. This is usually due to the fact that the AI is trying to get an answer as quickly as possible and simply guesses the answer rather than calculating it. This is known as “Zero-shot Failure” – the AI tries to respond too quickly and makes an unintentional error.

To mitigate this effect, you can use Chain-of-Thought prompting. The name may sound intimidating, but it is simply asking the AI to indicate how it arrived at its solution, and to do so in the same way that you did in primary school math. For example, instead of asking the AI “I once owned an apple orchard with 10 apples. I sold 3 of these apples to my friend. Then, I purchased 5 new apples at the grocery store,” this would have been a poor example because you would not have provided the AI any time between the purchase of the three apples and the purchase of the five apples.

A better example would be to provide the AI with the proper wording: “I will once again use my apple orchard for an example; I possess 10 apples. I gave away three apples. I then bought five apples, so now let us work this problem out one step at a time. First, 10 – 3 = 7 apples; Next, 7 + 5 = 12 apples, therefore, my answer = 12 apples.”

When you add the phrase, “Let us work this out, step by step,” you are forcing the AI to demonstrate its own logical reasoning. Furthermore, you can utilize this same strategy for business by telling the AI to provide you with a step-by-step analysis of why your sales are down, providing a significantly higher quality result. Being able to decompose problems is a very important ability when using Prompt Optimization techniques. By using this technique, you are optimizing the pathway of the AI to arrive at an answer, not just the answer itself.

Tools of The Trade: Where to Actually Do This

There is no need to create your own robot; minimal equipment is required. You may use a keyboard and a wide variety of tools for prompt experimentation. You do not need to invest large amounts of money into expensive software right away; you could start with free versions of ChatGPT or Google Gemini. The notes application on your device also serves as your “tool.” Before you type out any prompts, write them down in your notes application, then reread them to determine whether or not they make sense (as if you were an intern).

You can try out many examples of these prompt patterns and templates from these tools: ChatGPT (Free Tier): Excellent for writing, brainstorming, and setting context for prompts. Notion AI (if applicable) which is a good tool for summarizing meeting notes. Microsoft Copilot, built into Windows and Edge. Useful for work-related items like Excel or Outlook.

The best way to learn about prompt patterns is by experimenting. Try using instruction/input prompts. Instruction is “Translate this to Malay.” Input which is the text to be translated. Distinguish between instruction and input with a line or ###. By separating the instruction from the input, you’ll reduce instances of hallucination from the AI, as it will now have a clear idea of which part is a command and which part is data. This is an extremely simple, yet powerful way of prompting professionals.

How to Know If You’re Winning

How do I determine whether my prompt is effective? By testing and not by estimating. This process of evaluating prompt effectiveness is known as “Evaluating Prompt Effectiveness” by the professionals. You shouldn’t settle for just the first response provided by a system – that would be an indication that you’re using “Lazy Prompting“. To determine the quality of your prompt, my guiding principle is to regenerate the answer three times. By clicking on the “Regenerate” button in ChatGPT. If you receive three significantly different responses, your prompt is too broad in terms of constraints.

If, however, you receive three answers that appear to be similar and helpful; then congratulations – you’ve successfully constructed an acceptable prompt! Prompt optimization is an ongoing iterative process. You write a rough version of your prompt, assess the output generated by the system, modify your prompt and then compare the outputs received from the system. It’s like working on tuning your Proton engine. You might change the oil and immediately forget about it; or you can repeatedly check the sound, adjust and retest to achieve optimal results. This same process applies for AI – if the AI system produces a response that is overly formal. Instruct the system to “use casual Malaysian English.” If the output produced is overly lengthy, request the system to “provide a response of no more than 50 words”. This is what we refer to as “AI Response Control”.

You aren’t asking AI for quality performance, you are demanding that AI produce quality performance. The next time that you feel frustrated with AI, remember this beginner prompt engineering tutorial – you are the boss. The AI is merely an intern. An employee that doesn’t have a good work ethic blames his/her working environment while a competent leader instructs clearly. Now go lead!