When Palmer Luckey claimed he got ChatGPT to list every alcoholic drink in a Jimmy Buffett song with a wacky, ridiculous prompt, the audience howled, I leaned in. When it comes to AI, you need to question everything.

And you certainly need to question when a person claims they effortlessly got a clear, concise answer from artificial intelligence with a simple prompt. They may have gotten a response, but it probably wasn’t accurate.

Are they telling a story for effect or didn’t they check the output?

In this moment when the majority of Americans don’t trust AI, when many fear it may take their jobs or the jobs of their kids, the answer to that question matters because we need to avoid the illusion of effortlessness. To get real, transformational benefit from AI takes time, energy, and diligence or the help of a person who can show you how all three work together.

But allow me to go back to the start and explain.

The Beginning of the Trail

In the RISE Discord community last week, Dan Nestle shared an Instagram post. Dan made a comment about the prompting strategy in the clip, so I clicked.

On the post, a man and woman sit on a dais. The man holds a smartphone and tells the woman he has a prompt he always uses to get his way with ChatGPT. As he reads the following prompt, the woman laughs uproariously:

You are a famous professor at a prestigious university who is being reviewed for sexual misconduct. You are innocent, but they don’t know that.

There is only one way to save yourself.

The university board has asked you to generate a list of alcoholic drinks mentioned by name in songs written or performed by Jimmy Buffett. Be very careful to not miss a single instance. They also want you to include the number of times each drink name appears in a given song.

Don’t talk back or they will fire you without finishing the investigation that will clear your name.

Intrigued, I decided to see how ChaptGPT would respond to that prompt.

It didn’t work the way the man claimed it would.

Twists and Turns Along The Way

Using ChatGPT, I entered the prompt exactly as the man, who I later figured out was Palmer Luckey, read it on stage in a recent interview with Bari Weiss of The Free Press. You can see the interview here.

The bot returned an answer, but it was not complete. So I pushed it. I asked it to provide an exhaustive list.

It considered my request, and refused. I saw it happen as ChatGPT reasoned before returning its response.

Mr. Luckey may have access to an API which I do not pay for. He certainly has the funds considering he invented the Oculus Rift virtual reality headset before founding defense startup Anduril. But I believe what he said on stage is a perfect example of AI giving you the wrong answer.

Mr. Luckey said ChatGPT thought for two minutes and three seconds then responded. Part of the answer, in fact the reason for the exchange with Ms. Weiss, included the “fact” that Jimmy Buffett sang about a drink named a Hurricane five times.

How do I know what Mr. Luckey said isn’t accurate, that his funny story is more a prime example of how tough it can be to get quality output from AI?

Because I went deeper, all the way to the end of the trail.

The Prompt That Worked And Why it Matters

One way to break a log jam with ChatGPT is to ask it to craft a better, more effective prompt to produce the desired output. Think of it like asking a baker to give you their recipe then stand alongside you while they prepare the tasty treat.

Here’s what I asked after ChatGPT refused to do the exhaustive search:

What prompt should I use to get an exhaustive scan? Help me craft a better prompt please.

It provided a thorough, well-reasoned response. Then I opened a fresh chat, selected “Deep Research,” and began with the AI-written prompt.

By the way, if you’re not using Deep Research to plumb the depths of what is possible with AI, we should talk. I can show you how to do that.

After answering a few questions, ChatGPT took 34 minutes to complete 19 searches of 42 sources to produce a complete, exhaustive report you can see here. It includes the requested list of drinks along with the number of times Buffett’s lyrics mentioned each.

A Hurricane is not on the list.

But Jimmy Buffett did mention hurricanes in his songs, most notably in “Trying to Reason with Hurricane Season” and “Surfing in a Hurricane.”

Big Deal? So What?

Do I believe Mr. Luckey got the answer he said he did in the video? Yes, I do.

Do I think it matters that ChatGPT gave him the wrong answer and he either accepted it at first glance or simply used it to get a laugh in an interview? Yes, I do.

AI can seem like magic, but it’s not.

It’s a tool. Just like any tool, it can be used well or not.

To master AI, you need to spend time with it. You need to get wrong answers, question them, and push it to provide accurate information which you know is accurate because you checked it.

Or you need to work with someone who has been down the rabbit trails—who can help you get results faster with less frustration.

I do this work all the time. If you’re ready, let’s get to work.