ChatGPT vs Python- The ASCII art war.
As a hacker and someone who loves to tinker, I use ChatGPT and other AI tools daily. However, I’m not on the Hype-train; I can’t help but roll my eyes at some of the terrible failures of AI and LLMs.
Large Language Models(LLMs), create an interesting dichotomy where the hype-side believes that “AI hallucinations” are not a problem while the opposition believes accuracy is more important than speed. An obvious distillation of a complex topic.
What I keep reading and hearing about is how AI Agents are going to be replacing programmers and the uptick in the trend of “vibe coding”. Although, unless you have used these tools you are probably not aware of how many wrong answers you have to sift through to get a coding solution that actually works or a correct coherent answer.
The real job threat is efficiency through the proper use of AI, not an AI agent replacing a human. ChatGPT and LLMs like it are more akin to a calculator in that if you do not have an educated hypothesis of what answer you expect than literally any answer given is correct. Which is pretty dangerous given the right context.
In Vibe coding for instance, you keep adjusting your prompts until the end result is achieved. Surely their isn’t any danger in running code you don’t understand; but I digress.
My ChatGPT challenge
Could ChatGPT create an ASCII art of the fairy Navi to put in my CLI tool?
I’m a Zelda nerd; so much so that when I decided to create a CLI tool I named it Navi. If you are not familiar, Navi is the fairy from the 1990s Nintendo game: “Zelda: The Ocarina of Time”.
So while I was traveling and stuck in an airport I had a thought.
Could ChatGPT create an ASCII art of the fairy Navi to put in my CLI tool?
Since I use ChatGPT nearly every day I figured this should be easy.
My first prompt was to ask ChatGPT if it can create CLI graphics, to which it confirmed and added a few examples. See the screen shot below:
My next prompt was to ask it to create navi; the fairy from Zelda in CLI art.
In the above screenshot, you can see it produced an output that isn’t that bad. Especially with the added “Hey! Listen!”; a direct quote from the game. However, I wanted something a bit more advanced. So I simply asked. “Can you make a more advanced one?”.
The failure here is just epic. In the above screenshot, it appears to be a simplistic picture of a smily face with half a torso. Even though ChatGPT confirms it’s intent was to: “create a more advanced ASCII art of navi with a better sense of shape, glow, and wings”.
Can we pause for a moment for your response had you asked a human for a more advanced version of the first image and this was what was created. A smiley face disjointed from it’s torso.
Since I’m really interested to see the outcome through, I asked ChatGPT one more question before switching my approach. “Can you make one with wings?”
Get Casey Reid’s stories in your inbox
Join Medium for free to get updates from this writer.
Now this one admittedly was on me given it’s ambiguous nature, but I wanted to see how ChatGPT could build on the current conversation. Below the results are just laughable:
The description reads: “..Staying true to her appearance as a glowing orb with transluecent, insect-like wings”. However, what was produced was a talking head on a tripod with feathers coming out of his head. I literally laughed out loud in the airport when this prompt finished.
I searched google for a simple image of “navi” and gave Chatgpt the above image and asked it to replicate it in ASCII art. Below it read the image and provided some elements of the image in bullet points for confirmation, on the left. However, as you can see, the results look more like Mikey from the hit kids movie “Monsters Inc(shown on the right).”
At this point, the failure has become increasingly amusing as I started to realize ChatGPT just isn’t built for situations like this one.
After a few more iterations ChatGPT gave me one more prompt in the free version before it told me to upgrade. I asked it to be more detailed, bigger wings and more elegant and below is the result.
I can’t make out what this might be, but it isn’t success. It looks like a one-eyed cow wearing a funny hat. This ended my hilarious ASCII art failure with ChatGPT. However, since I don’t like to fail I decided I might be able to use python to create some ASCII art.
Enter Python: Turn a JPG into ASCII art using Python
After a few minutes of some google foo, I found a pretty slick example of turning an image into ASCII art on GitHub. This was the cleanest version I could find in a short period of time. Below are the links to the github repo and a screenshot from the Readme.
The code base is easy to understand and quite short in length. So I wrote some simple test code to take the image I gave ChatGPT and use this python script to turn it into ASCII art. First, let’s start with the code.
In the above screenshot, you can see I start the code by importing “image_to_ascii” and instanciate an instance as an object called “obj”. I then provide the image path, small_navi.jpg and use the “plot” function to print the results to the screen. Last I use the “save_to_file()” function to save the results to a file.
This solution couldn’t be simpler and the results are fantastic!
This is perfect. Nearly an exact replica of the picture I provided the script.
Conclusion
First, ChatGPT is an amazing tool; similar to the calculator. Second, it is not for every use-case and even when you succeed it will come with many iterations. Third, ChatGPT is only as good as you know how to use it and you understand the subject matter enough to know when it is not correct. Last, let’s be more honest with where these AI and LLMS alike have value and where they don’t.
Can you get a better outcome from ChatGPT?
Thank you for being a part of the community
Before you go:
- Be sure to clap and follow the writer ️👏️️
- Follow us: X | LinkedIn | YouTube | Newsletter | Podcast | Differ | Twitch
- Start your own free AI-powered blog on Differ 🚀
- Join our content creators community on Discord 🧑🏻💻
- For more content, visit plainenglish.io + stackademic.com
