How many times has it happened that ChatGPT has not given you the answer you were looking for?
Well, I don’t know about you, but it’s happened a bunch of times with me! I ask AI to give me something, and it gives me something else. It’s called “Hallucinating”, and a lot of AI models do this (although they’re getting much better with their responses).
So anyway, how do you get the most optimal response for your query?
It’s with good prompt engineering.
[For those who don’t know what Prompt engineering is, don’t worry. It’s just a fancy word for “how to guide AI to receive optimal responses”. Essentially, the prompt you give should be framed in a way that the AI gives you the response you’re looking for]
While there are many techniques of prompting, let’s look at six of them today.
(Also, while this is true for most AI models (and chatbots), we’ll assume you’re using ChatGPT)
Prompting techniques
Zero-shot prompting:
This is the simplest technique. You just ask ChatGPT a question, without giving any examples.
Example:
"Summarize this article in one paragraph.”
When to use: For simple, well-understood tasks.
Few-shot prompting:
Here, you provide a few examples in your question, of inputs and expected outputs. Example:
Summarize each sentence in one line. Here’s an example I want you to keep as reference:
Text: "The stock market rebounded today after reports showed lower-than-expected inflation."
Summary: Stock market rose after inflation reports.
Now, summarise the below text in line with the above example.
Text: "A new species of dinosaur was discovered in Argentina, believed to be 90 million years old."
When to use: When the model needs context or formatting guidance
Chain-of-thought prompting
Here, you ask the AI model to “think out loud” and show its reasoning step-by-step. Example:
"If Tom has 3 apples, and he gives away 1 to Susy and gets 2 more from Mary , how many does he have left? Let's think step by step.”
When to use: Great for math, logic, or reasoning. Also, the best way to solve multi-step problems.
ReAct prompting
Here, the model reasons out loud and takes actions, like calling tools or APIs, or solving your problem. Example:
Should I hire a new sales representative or focus on improving the current team's performance?
The answer to the above prompt would be something like “It would be better to focus on improving the current team’s performance through training and incentives. If the team already has potential, it’s more cost-effective and sustainable to enhance their skills rather than hire someone new. However, if the team is not performing well and lacks necessary skills, hiring a new representative could bring immediate value. Given that the team has potential, investing in their improvement will likely yield better long-term results.”
When to use: For problem-solving or decision-making, or analysing complex situations
Iterative prompting
Here, you prompt the model, get an answer, then prompt again to improve or critique it. Example:
"Rewrite this paragraph to sound more persuasive." (After first response) → prompt: "Now make it more concise.”
When to use: For highly subjective tasks (writing, brainstorming, etc) that require multiple iterations based on your requirements
Role-based prompting
Here, you ask the model to take on a specific role. Example:
You are a startup pitch coach. Review this pitch deck and suggest improvements
When to use: When tone, expertise, or persona matters. I personally use it when I want an “expert” opinion on something. So by saying “You are a startup pitch coach”, the AI will specifically wear the hat of a startup pitch coach, and give answers accordingly.
So what do you do with these prompting techniques?
Use them to optimise your responses.
Oh wait, here’s the fun part - You don’t need to explicitly tell ChatGPT (or any other LLM you’re using) to use a particular prompting technique. Based on the kind of prompt you give it, it will use the relevant technique.
For example, if you tell it to “go step-by-step” or even “break it into smaller tasks”, it will know that it should use Chain-of-thought prompting, and give you a more accurate response. That’s the beauty of AI 🙂
So the next time you feel that ChatGPT is giving you a shitty answer, try to incorporate a prompt that will tell it to use a particular technique. Or you can straightaway tell it to “use ABC prompting technique” and it will do what you ask it to.
Here’s a disclaimer: To keep this newsletter short, I’ve only covered 6 prompt engineering techniques. However, there are many more, and I’d urge you to go through them if you want to deepen your knowledge of AI. A simple Google search will give you what you’re looking for, but if you’re looking to be spoon-fed, just shoot me an email and I’ll share a few links (I’m kidding about the spoonfeeding thingy haan, feel free to email 🙂)
Something cool I explored this week in AI
As a content creator, I’ve tried (a lot of times) to create Instagram reels, but I’m not able to make a habit out of it. I love writing the script, but memorising it and shooting it has always been a pain.
Not. Any. More.
Last week, I tried this tool called veed.io which has a feature called “Eye contact correction”, where if you upload a video of yourself with your eyes looking away, it will correct it and make your eye point to the camera.
This means that I can read a script from a screen and shoot the video, and the AI will correct my eye gaze to make it look like I’m actually looking into the camera. No need to memorize the script and take like 37 cuts.
So cool!
So will you see some reels from me now? Well, I certainly hope so 🙂
Until then, these newsletters are all we have 😃
Couple of other things I’d like to ask (and tell)
As you know, this is a brand new newsletter (and this is only the third email). So I’d love if you could tell me what you’re looking for, from this newsletter. It’ll help me write relevant content that will eventually help both of us. Please let me know HERE - Just 3 mandatory questions, won’t take more than a minute!
I also write a newsletter simplifying business and money management. It has more than 10,400 subscribers as of today. If you’d like to read more about startups, the economy or money management, subscribe to it here
I write smaller, more frequent content on LinkedIn - You can follow me there for byte-sized stuff
If you liked this piece, please share it with your friends (SHARE button below). It’ll help spread awareness faster :)
Until next time…