ChatGPT Image Generation Experiments

What appears to be happening is a quiet, crowd-sourced audit of ChatGPT's image generation defaults. In early April, one user asked the model to generate "the most average human on the planet" and documented what it produced. Then, toward the end of the month, another user ran a related experiment — prompting the system for its vision of a "typical Reddit user." Neither of these is a research paper. Neither is a controlled test. They are two people, weeks apart, poking at the same soft question: what does an AI trained on the internet actually think a human looks like when you strip away specificity? The results, whatever they were, apparently felt worth sharing. That alone is the story — not the images themselves, but the impulse to ask, and the impulse to publish the answer.

If confirmed as a pattern rather than coincidence, here is what this means. Users are beginning to use image generation not just as a creative tool but as a kind of mirror held up to the model's training assumptions. When you ask an AI for "average" or "typical," you are not asking for art — you are asking it to reveal its priors. The demographic and aesthetic defaults that emerge from those prompts tell you something real about whose data dominated the training set, whose face the model considers unremarkable. That has implications for product teams building with these models, for researchers studying algorithmic bias, and for anyone who assumed that "neutral" prompts produce neutral outputs. They do not. They produce a weighted average of whoever was photographed most, described most, indexed most.

Watch for organised attempts to systematise this kind of prompt-probing — researchers or journalists running structured comparisons across models, countries, and demographic descriptors. If that happens, what looks today like Reddit curiosity could become a credible audit methodology.

NewsHive monitors these sources continuously. All signal titles above link to the original reporting.

Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →