Are AI Prompts Fake?

Are AI prompts fake?

Seriously, how do they work?

Artificial intelligence models are now used to generate everything from articles to images, and more lately, video with a fidelity that is almost indistinguishable from actual footage. Many people have now integrated it into their daily lives. Some people use it because they have not developed their skills in whatever creative pursuits they enjoy. Meanwhile, others use it as a blatant attempt to make money with minimal effort.

We have already seen an innumerable amount of articles decrying artificial intelligence as a plagiarism engine. While this rings true for prompts that contain wording like “in the style of” or “write me a blank story about blank”. And any other similar inputs. However, this does not explain the results of other prompts.

While I have a general aversion to using artificial intelligence, unfortunately, I have to admit that I do have to use image generators on occasion. For me, this is a resource of last resort when publishing articles on Vocal. I only use it when I can’t find an image in Unsplash that I like, or I don’t have one of my own. A good example of this is when I had to use an AI image for my poem, A Venetian in Constantinople, because no image of Renaissance Venice or Constantinople exists.

But the more I have used these, the more they have raised questions that remain unanswered. Mainly, where do the results from these prompts come from?

For some, the answer is obvious. The software is skimming the internet for images or text to generate its results. However, I got bored one day and started to throw random things into the prompts to see what happened.

First, I started with some of my poems. Then I started to type random characters in. This then led me to start using a random password generator to produce a larger body of text I could use. This has led me to switch to using random text generators to make longer prompts.

At first, this started as fun, to see what the image generator would come up with. Then, knowing that most of the results are regurgitated based on the software skimming, I started a solo campaign to put in as many nonsensical prompts as possible. I started doing this, knowing that generators are increasingly working off of a closed loop where they are stealing from each other more than pulling from existing art. And well, the results got strange.

I quickly found out that it did not seem to matter what I put in; I would almost always get something back. Further, the results were not consistent using the same prompt.

For instance, if I use the prompt of “A lone figure stands silhouetted against a crimson sunset, the sky ablaze with fiery hues that reflect in the still waters of a vast ocean.” One expects to get relatively similar images across several iterations of this prompt. However, this does not happen using randomized prompts. The randomized prompts generate some of the most random images possible for the same prompt. And the strangest part is that these generate more photorealistic images than prompts with that in them.

Therefore, what is the cause of this?

My personal theory is that the creators of these AI programs don’t actually know where the results come from. The prompt results are being pulled from somewhere unknown. Whether that’s from within the worldwide machine mind or elsewhere is yet to be known. And that’s a problem.

Thank you for reading my work. If you enjoyed this story, there’s more below. Please hit the like and subscribe button, you can follow me on Facebook, Twitter, and Instagram @AtomicHistorian. To help me create more content, leave a tip or become a pledged subscriber. I also make stickers, t-shirts, etc here.

More from this author: