| Assignment Type | Assignment — Creative Exercise |
|---|---|
| Topic | Understanding Generative Adversarial Networks through storytelling |
| Author | Cesar Zaldivar |
Generative Adversarial Networks (GANs) are an AI method where two models learn by competing with each other: a Generator and a Discriminator. The generator’s job is to create outputs that look real. The discriminator’s job is to look at a mix of real and generated examples and decide which ones are real and which ones are fake. Each round of feedback helps both sides improve: the generator learns how to make its output more convincing, and the discriminator learns how to spot what seems unrealistic.
For this assignment, I used personal anecdotes and wrote them in a way that could sound slightly unrealistic to someone who does not know me well. This is a useful test because it highlights how hard it can be to judge authenticity from text alone — sometimes reality beats fiction.
To analyze the stories, I read each one as if I were the “discriminator” and made a real-or-fake judgment based on four writing signals:
| # | Signal | Description |
|---|---|---|
| 1 | Sequence Realism | Whether the sequence of events felt realistic |
| 2 | Natural Dialogue | Whether the dialogue sounded natural and unscripted |
| 3 | Specific Details | Whether details were specific, consistent, and grounded |
| 4 | Earned Outcomes | Whether the outcome felt earned rather than too convenient |
One of the stories described meeting my wife at a coffee shop outside my old apartment — a place that always smelled like espresso and cinnamon, even before you opened the door. The story included specific details like a protein smoothie routine, afternoon work calls, and noticing small habits (sticky notes, a pen twirl, a quick smile to the barista) before any conversation began. These kinds of grounded, specific details are precisely what a GAN discriminator — or a careful reader — uses to evaluate whether a story feels real or generated.
This assignment made the abstract concept of GANs genuinely tangible. By playing the role of discriminator myself, I understood firsthand what kinds of signals separate convincing output from artificial-feeling content: specificity, natural flow, consistency, and earned outcomes.
The most surprising insight was that real stories can sometimes seem more unbelievable than fabricated ones. This mirrors exactly how GAN-generated content can become indistinguishable from real data — and why deepfake detection is such a hard problem. The iterative feedback loop between generator and discriminator is a powerful principle that shows up everywhere: in creative writing, in scientific peer review, and in how humans refine ideas through debate.