The Saga of My Pumpkin Illustration (aka, ChatGPT’s comical failure)

On Friday I posted my comparison of Public AI’s attempt at writing an autumn poem versus ChatGPT’s—a head-to-head challenge that ChatGPT won handily. Then to illustrate my post, I asked ChatGPT to create a quick, simple image of pumpkins for me.

At least I thought it would be quick and simple. In the event, it was not. Although I was frustrated because I had to wrap up that blogpost quickly and get back to work, it was actually really funny to watch ChatGPT get it wrong over and over. Kind of like watching our dog suddenly (and stubbornly) fail to comprehend simple commands that I know she understands but is pretending not to know because she doesn’t want to do what I’m telling her to.

So here is the chat of those image fails, along with the resulting images.

I said:

Can you produce an image to go along with this poem? Something featuring pumpkins and scattered fallen, colorful leaves. Thank you!

ChatGPT said:

First try for a ChatGPT pumpkin image

I said:

Can I revise? Can you make this an image of multiple pumpkins sitting on the steps leading up to a porch, maybe some on the porch, too, with fallen, colorful leaves scattered about. Thank you!

ChatGPT said:

I said:

Beautiful! But too many pumpkins. Let’s keep two on the porch and just have one on each step. All pumpkins should be enough to the side of the pathway up that people can actually walk up and down the steps and onto the porch. Can you revise? Thank you!

ChatGPT said:

I said:

I’m sorry! I’m not being clear enough. Let’s keep the basic steps and porch here. I want three pumpkins: one on each step, set far to the side so that people can still walk up to the porch without tripping over pumpkins. Keep the fallen leaves; that part of the image looks great. Thank you for trying again!

ChatGPT said:

[Note/Aside: If you look carefully, you’ll see that ChatGPT created an entirely new, nearly identical wrong image. The two pumpkins are slightly different, and if you look at the leaves on the bottom step to the left of that pumpkin, for example, you’ll see three leaves to the left of the pumpkin in the first image and four to the left and immediately below, two of which are very small, in the second image.]

I said:

How about let’s try this: ONE large pumpkin, sitting on the porch next to the pillar. Thank you 🙂

ChatGPT said:

At which point I just laughed, threw up my hands in surrender, and went with the first image (non-porch) just to get the post illustrated and published. AI’s logical judgment clearly isn’t all there quite yet. ChatGPT can put any number of pumpkins you’d like on a front porch, but it apparently can only do so in a way that will obstruct people’s ability to walk up and down the steps!

Unknown's avatar

About Katherine Wikoff

I am a college professor (PhD in English, concentration rhetoric) at Milwaukee School of Engineering, where I teach film and media studies, political science, digital society, digital storytelling, writing for digital media, and communication. While fragments of my teaching and scholarship interests may quite naturally meander over to my blog, this space is intended to function as a creative outlet, not as part of my professional practice. Opinions are my own, etc.
This entry was posted in Art, Creativity, design, generative AI and tagged , , , . Bookmark the permalink.

Leave a comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.