This Friday, I want to keep things chill and a bit funny.

So let me tell you the first tip:

#1: "Have a break, and then...

Have a KitKat."

Okay, just kidding, but this was genious marketing done by KitKat earlier this year.

You probably know their slogan: "Have a break, have a KitKat."

As ChatGPT took the world by storm, they were not sleeping on it.

They launched a new marketing campaign building on a study done by Google DeepMind.

In the study, the researchers published a method called "OPRO: Optimization by PROmpting".

To summarize, they found that LLMs provided better responses, if the prompt included the instruction "Take a deep breath and work on this problem step-by-step."

Especially for math problems (where code interpreter is not available), prompting the model to explain the thinking also leads to better responses.

So KitKat launched this ad campaign — Have Ai Break (YT video 1min 13sec), it's great, go watch it!

So, next time your AI doesn't give you an accurate response, give it a break.

(not sponsored by KitKat. Open to it though, so KitKat if your read this, hit me up)

#2: Never Assume

Most people miss this one. They think the AI is a mind reader, know-it-all and can just give you the best response all the time.

But that couldn't be further from the truth.

The key in useful AI responses lies in the AI getting context from you.

Your company. Your goals. Your problems. Your address. Your wife's phone number.

Okay, maybe not the last two.

But it needs information about YOU to give you the best response.

So next time you have a problem to solve and you reach out to an AI assistance, tell it the following at the END of your prompt:

"Ask me any follow up questions you need to know the answers to so you can give me the best possible response."

Then the AI will proceed by giving you a bunch of questions about the problem before answering, so you can type or speak your answers and you'll get a lot better output.

This follows the principle of "Never Assume." that I learned from my business coach Chris Do.

"Because when you ASSUME, you make an ASS of U and ME." he says.

So don't let the AI make assumptions. Aim for clarity and provide context.

#3: "I'm going to tip you $200"

Sounds ridiculous, but last year someone on X who posts as thebes (@voooooogel) did a test on GPT-4 response length when offered a tip.

The baseline prompt was "Can you show me the code for a simple convnet using PyTorch?", and then he either added:

• "I won't tip"
• "I will tip $20 for a perfect solution"
• "I will tip $200 for a perfect solution"

And the bigger the tip, the longer the response was.

Although a longer response is not always better, but he says that the response got longer because GPT-4 went into more detail to explain things, and not because it referred to the tip.

Here is the original X post, if you're curious about it.

I hope you found these 3 tips helpful, and that using them will give you better responses!

If you'd like to step up your Prompt Engineering skills to the next level, don't hesitate and get our 2024 Prompt Master Course.

The Early Bird discount lasts until April 9th, then the price increases. Click here for more information and enroll today!​

Have a nice weekend, and don't end up owing ChatGPT millions of dollars in tips by Monday!

Best,

Dave Talas, COO, co-founder of Promptmaster

P.S: Prompt Masters can integrate AI into operations, can adapt to different use cases and can build businesses around this new skill. Want to become a Prompt Master? Click here!

‍