The sweet spot of models and humans
Sweet Spot A couple of months ago, my task was to test the model and find its sweet spot. The sweet spot of the model lies somewhere between hallucination (for example, something overly distorted and visibly wrong) and a reasonable answer, the model will fail, but not fail too badly. What’s interesting is being human in this process, getting to the point where you can predict the sweet spot of the machine. It is during those moments that I begin to reflect on our own sweet spots. For example, I knew the model would have an easy time generating an organized kitchen or a messy kitchen. But if I asked it to generate a chaotic kitchen, it would struggle. There are very human concepts that live between experience and mystery. A chaotic kitchen is the result of so many nuances—was there a party the night before? What kinds of dishes or food would be left behind to make it feel real? Or was it a rushed morning before school, a mother trying to prepare breakfast and pack lunches for her kids before leaving for work? So much has to happen beforehand to create a scene that is chaotic,but not literally chaotic, because literal chaos would be something else entirely. As a human, I like to think that the sweet spot is like a bridge between levels. It’s when you wake up feeling slightly off, look in the mirror, and ask yourself: who am I, and what am I doing? We go through many small shifts to reach that sweet spot. It’s that moment when your predictable thoughts go off track, but not completely. It might resemble imagination or intuition, but it’s not quite that. The sweet spot has a glitchy quality. You know when you’re speaking to your partner and something glitches? Something shifts, and you can’t name it. You’re human, there aren’t words for everything happening inside you. Humans, glitchy as machines, have their own sweet spots. I feel that when I am glitching, I am close to one. It bothers me, and it needs to bother me. Of course, the sweet spot is not all sweet. Then, suddenly, eureka, it unfolds. It wasn’t an accident. You triggered it. Now you need to adjust, but there is no one training you the way I patiently train the machine. I tell the machine where its sweet spot is. I correct it. But who corrects me? I’ve had moments like yesterday, watching The Amazing Digital Circus and paying attention to the characters and their behaviors. They all live on the edge of being human or being abstracted. Abstraction feels like a collapse of pixels, an exaggerated sweet spot, almost feels like hallucination but its not! All of the characters exist in this anguish of knowing they are neither one nor the other, not merely human, not merely machine. Just… something of these new times. I felt displaced, an intense glitch inside, not knowing exactly where I belong. And it’s okay. Sweet Spot.
