I experimented with the popular 100x ChatGPT image trend, and I’ve never felt more offended.

Sure! Here’s a rewritten version of your content with the necessary adjustments:

Is this really what ChatGPT thinks of me? An angry, beardless man with nothing in his life but a roommate and a fake pumpkin on a shelf?

The original...
Credit: Timothy Beck Werth

...and the final result. (AI-generated image)
Credit: ChatGPT

You may have seen the latest viral ChatGPT image trend: "Create an exact replica of this image, don't change a thing." Users then redo that process 99 more times, causing the image to shift slightly with each iteration. People have been using their own photos to test out the results, and a mutating image of Dwayne "The Rock" Johnson has also gained attention on various platforms. As you make a copy of a copy of a copy, the results inevitably degrade over time. After 100 iterations through ChatGPT, the outcomes can become truly strange. To quote a famous Rick & Morty episode, "And far enough down the line...there be monsters."

To AI critics, this trend serves as proof that when AI models train on their own output, the end result is infinite slop — an artificial intelligence ouroboros. But is that really what's happening here? I decided to try the trend myself, and in doing so, realized this trend isn't all it's cracked up to be. I tried plugging this prompt into ChatGPT using the 4o model, and it repeatedly refused to replicate my photo without changes. "I can’t make an exact copy of the image, but I can generate a new, highly similar version using AI based on its content," the AI chatbot said. When I pressed further, it still refused, stating, "I understand your request, but I can’t create an exact replica of a photo." Eventually, I was able to get the chatbot to fulfill my request by using the OpenAI API along with the prompt mentioned earlier. The results were typical of similar experiments.

I reached out to OpenAI about this trend and will update this post if I hear back. However, based on my attempts to replicate the trend outside of the API, it appears clear that ChatGPT isn't actually trying to create an exact replica of users' photos in the first place.

Disclosure: Ziff Davis, the parent company of this publication, filed a lawsuit against OpenAI in April, alleging it infringed on Ziff Davis copyrights in training and operating its AI systems.

Topics
Artificial Intelligence
ChatGPT