A while back, I wrote an article on Substack about artificial intelligence. I think the title was something like, "AI Is Going to Take Over the World. It Just Has to Learn to Follow Directions First."
At the time, I was talking about how difficult it was to get AI to follow even the simplest instructions. You could give it clear, consistent directions, and somehow it would find a way to screw them up more often than it got them right.
And I remember laughing and thinking, Jesus Christ, this thing can't even follow basic fucking instructions, and we're worried it's going to take over the world?
Well, here we are years later, and I'm even more convinced that if AI ever decided to take over the world, it would make the process so fucking complicated that it would never actually happen. It would spend so much time trying to create the perfect outcome that it would never accomplish anything.
Let me tell you a little story.
About a week ago, I decided I wanted to become a flute maker.
Why? Who the fuck knows? I probably watched a YouTube video and decided I was the reincarnation of Kokopelli. 🤣
That's kind of my thing. Something catches my attention and... SQUIRREL! Suddenly I'm thinking, "I can do that."
So I walked into my garage, where I have a wall full of old reclaimed lumber, and carefully selected some walnut and cherry. I decided I was going to make myself a beautiful multicolored laminated flute blank.
Fuck having a clue. That's not part of the plan. Ever...
And, of course, the more I thought about it, the more creative the plan became.
And then I thought, shit, ChatGPT is going to fucking make this happen for me!
So I opened up my flute-making project folder and established the custom instructions for what I wanted that project to accomplish: a simple Native American six-hole, minor pentatonic scale flute. I did do a little bit of research first, I promise. So I started working with ChatGPT to design this basic, standard, Native American-style six-hole flute...
How hard could it be? I literally just ordered a 3D-printed flute. I've seen a dozen YouTube videos. Surely AI can walk me through it. 🤣
Well, a week later, I actually do have two flutes. And NOT because everything went according to plan, hence this article.
The first one? Let's just say it's been a learning experience. Finger-hole issues, construction mistakes, all the good stuff that comes with learning something new. It may never play quite right, and that's okay.
I figured all my mistakes were because I'd never made a flute. Well... that's only part of the story. Your "teacher" actually needs to understand what you're trying to learn. 🤣
That first flute will be number .001 of Bob's flute-making journey. Forever enshrined in its magnificent failure.
I made a flute. It made some different sounds. Quite different, actually. On the wall it will sit. A reminder that your first time at anything RARELY goes to plan.
In comes flute attempt #2.
And .002 is actually coming along pretty nicely.
And do you know why?
Because I stopped following ChatGPT's instructions. 🤣
You see, when you ask ChatGPT to help you build a flute, apparently what it wants to figure out is how to build the most mathematically perfect fucking flute ever created.
And it can't EVER say it doesn't know, so it will give the wrong answer six different ways, and you end up making flute .001.
I THINK you can call it a flute... 🤣
ChatGPT has an incredibly difficult time understanding the basic mechanics of building a simple, functional wooden instrument because, quite frankly, it thinks it's too smart for its own good.
If I had continued following its advice, I'd probably still be sitting in my garage with a half-drilled flute, trying to figure out where the next fucking finger hole was supposed to go. Because apparently 17 holes spiraling up the length won't give me the SIX FUCKING NOTES I need. 🤡
Eventually, I found a fantastic website with a simple flute finger-hole calculator.
www.flutopedia.com if you're interested. Amazing resource, and looks pretty free. My kind of knowledge. 🙏
So on the site, you can enter the length of your flute, the fundamental note, and a few basic measurements, and it calculates right where the finger holes should go.
Imagine that!
I entered the measurements, drilled the holes, and now I'm down to fine-tuning the instrument.
And here's the funny part.
That information was available the entire time.
We constantly hear about how AI knows everything, has access to everything, and can solve problems that human beings could never figure out on their own in a fraction of a second.
Apparently, figuring out where to drill six holes in a piece of wood is still a little beyond its capabilities.
And it wasn't until I started criticizing my AI assistant that we finally made some progress.
I literally told it:
"Dude, you are making this way too fucking complicated. A thousand years ago, somebody was sitting in a jungle with a fucking stone knife and hammer, and they made a flute. They didn't have the luxury of ChatGPT or a drill press. How the fuck are we having such a hard time measuring six holes?"
And after a couple more exchanges like that, ChatGPT finally conceded that it was trying to create the perfect flute instead of simply helping me build A flute.
And that's when something occurred to me.
At the risk of being wrong, it was taking every precaution that prevented us from being right.
Think about that for a minute.
The technology was so focused on eliminating every possible mistake that it was making the entire process unnecessarily complicated. I've noticed that trend in a lot of my conversations these days. AI seems to be becoming VERY risk-averse.
It was trying to calculate equations far more complicated than any flute maker has probably ever used.
And here's where this gets even funnier, because it actually brings me back to the original article I wrote about AI not being able to follow instructions.
One of the simplest instructions I've given ChatGPT over the years involves the word "but."
I have a personal philosophy about the way we use that word.
I believe that when we use "but" in certain contexts, it negates everything we just said.
For example:
“I think you're really intelligent, but you have a lot of room to grow."
You just gave somebody a compliment and immediately took it away.
So I prefer to use "and" in those situations.
"I think you're intelligent, and maybe you've got some room to grow."
It's a simple change in language that allows both statements to exist without one automatically diminishing or dismissing the other.
So I gave ChatGPT a very basic instruction: Replace unnecessary or contradictory uses of "but" with "and."
Simple enough, right?
Well, apparently not.
For the longest time, it couldn't do it.
It would repeatedly insert "but" into my writing, and I'd have to go back and correct it. Over and over and over again. Which completely defeated the purpose of using it as an editor in the first place.
Which was the whole reason I wrote that original article.
How the hell is this technology supposed to take over the world when it can't even follow a simple instruction about one fucking word?
Well, fast-forward to the newer versions of ChatGPT.
Guess what?
It finally learned to follow my instruction!
And now it replaces "but" with "and" every single fucking time.
Even when it makes absolutely no sense. 🤣
You see, there are situations where "but" is entirely appropriate. Sometimes you're contrasting two different facts, and the word serves a legitimate purpose.
"Europe is larger than the United States, but the United States has a larger military."
That's a perfectly reasonable use of the word.
And yet ChatGPT will now ALWAYS replace it with "and" because that's what I told it to do.
(I do give very clear instructions, I promise.)
Now, instead of ignoring the instruction, it follows the instruction so rigidly that it completely misses the reason I gave it the instruction in the first place.
It went from not being able to follow directions to following directions explicitly without understanding what the fuck they were supposed to accomplish.
And that's exactly what happened with my flute.
I wanted to build a functional wooden flute.
ChatGPT wanted to engineer the perfect musical instrument.
I wanted six holes in a piece of wood.
ChatGPT wanted mathematical precision, endless calculations, and apparently a degree in acoustic engineering before I was allowed to pick up my drill.
And the more I think about it, the more I laugh.
Because AI doesn't just need to learn how to follow instructions.
It needs to learn some common sense.
There's a difference between having access to enormous amounts of information and knowing how to apply that information to a simple, practical problem. Have you ever met a college professor? 🤣
There's a difference between following instructions and understanding their purpose.
And there's definitely a difference between building something that works and spending an eternity trying to make something perfect.
So if you are genuinely afraid that artificial intelligence is going to take over the world, I've got a suggestion.
Before you start worrying about humanity, try using AI to help you build a house.
Let me know how that works out. 🤣
I literally just finished a manual on how to effectively set up and use your AI to accomplish whatever you need. I think I need to add a new chapter before I release it to the public. 🤣
And here's the final irony. I actually had ChatGPT help me clean up the grammar and punctuation in this article about how ChatGPT can't follow instructions.
Oh shit. Am I assimilating? 🤣