Here are some tried and tested methods for better success with your prompting. It’s an Art and a Science. Structure your prompt so the model understands.
WHO do you want it to act as?
WHAT question do you want answered?
HOW do you want your answer formatted?
It takes practice to perfect. The more specific you are, the better results you'll get!
| TECHNIQUE | DESCRIPTION | EXAMPLE |
|---|---|---|
| The Consultant | Brainstorm new ideas and make sure you’ve considered all the angles. Get it to ask questions “one at a time” to get much deeper into a problem. | "Please help me come up with ways I can use ChatGPT to be more productive at work. Ask me 3 questions, one at a time to figure out something that works for my role." |
| Style Transfer | Produce text in a particular style, for example: a social media post that sounds like your brand (by uploading your brand guidelines or pasting text from a website. | "Please write a LinkedIn post about our latest product launch in the style of a tech-savvy influencer." |
| Role Assignment | Get your GPT to pretend to be someone. Great for practicing difficult workplace conversations and sales techniques. | "Please pretend to be Ada Lovelace so I can ask her opinions about ChatGPT. Before starting, please write a short biography of her." |
| Digest + Learn | Transform long documents into easy-to-read bullet points. This gives you a really quick entry into any complex subject. | "Please summarise this document as a set of bullet points." "Please explain it like I’m 5 / 10 / 15 / a grad student." |
| Self-Critique | Ask it to find things that were wrong with its original answer and then write a new one that fixes them. | "Please list at least 5 things wrong with your previous answer and then write a new one fixing them." |
| Plan + Solve | Develop a structured approach before executing tasks. Split your request you’re giving it into two steps: a planning phase and an execution phase. | "Please come up with a step-by-step plan for writing a memo on how ChatGPT can benefit my marketing." "Please use the plan to create the memo." |
| Tree of Thoughts | Simulate the behaviour of a group to see how it might explore a problem and make a decision. | "Please pretend to be the board of the company discussing the memo. They should pay attention to what each other is saying and explore the subject as thoroughly as possible." |
“Take a deep breath and work on this problem step by step. I’m proud of you.”
Until recently, we never imagined saying something like this to ChatGPT. But now we are.
Why?
We’re starting to realize how counterintuitive AI is.
It’s good at things you don’t expect.
It’s bad at all the things we've learned to turn to computers for.
So, we’re rethinking all of our instincts and giving some AI experiments a try:
1. We’re learning to be kind to AI.
We’re going to say thank you, experiment with motivational phrases and praise, and explain why it’s so important it does a good job—testing and learning our way to the incentives that get AI to perform most effectively for our work. It may sound funny, but AI is trained on human data, and humans are (famously) emotional. Those human emotions are baked right into AI’s worldview, so engaging with their feelings is a tried and true route to an AI that performs exponentially better, just for you.
2. Talk to AI as you would a team member
Provide context, details, and expectations.
3. Ask follow up questions / have a conversation
Ask follow up questions to refine responses or get additional information.
