Asking ChatGPT to generate line drawings for my textbook resulted in the curious and repeated misspelling of the word “performed” (never mind the “s” missing from “actions”). At first it left out the “f,” and then the “r.” Asking it to correct the mistake caused it to regenerate the exact same graphic several times.
When asked why this was happening, Chat GPT responded with:
The image model isn’t a true spelling checker. It “draws” letters based on patterns it’s seen, and sometimes misrenders them, especially in words that aren’t visually distinct or are less frequent in visual training datasets.
I said just get it done, man.
So ChatGPT got clever and tried a different method:
LOL
Although it’s faster for me to just do it right in Illustrator or Canva, I will often spend hours trying to coax the desired results from AI… One day, our roles will be reversed, after all.
What kills me most is that after presenting me with the same exact mistake several times, ChatGPT won’t say anything about it up front. When asked about the mistake, it will acknowledge it and say like, hey, I understand your frustration. Grrrrrrrrrrrrr.