Why does it feel so wrong to design something in Figma these days? It feels like I’m being stubborn, ineffective, and wasting time. It also feels like I’m cheating on my LLM, and should return home to Astra as soon as possible, so we can work on the next important thing, quickly. I kind of miss swirling around in Figma… is that wrong?
Polarization is starting to happen among creatives who sort themselves into the bin of pro or anti AI for creative work. Here are some thoughts.
I’ve been working with AI to do creative work around animation, web design, motion design, graphic design, brand tool building, and writing. Here’s what I observed about myself and the work for the domains in which I’ve been experimenting, which I will flag are very visual domains, except for arguably writing.
Most of the time when I ask AI to produce something for me, the output will be just slightly worse than what I would come up with, by at least 1%. My reaction is something like “yooo that’s pretty good” but almost never “this is a certified banger”. 1% worse, but at least 5—100x faster than if I made it myself. I should note that throughout this piece, I’m only referring to tasks which I have the skills to do myself. If I accept the output and move forward, I can move on to the next step. It also means that I’m accepting the x-1% output. But that’s fine, right? It’s just 1%, and the forward motion gained from doing something quickly outweighs perfecting it by merely 1%. However, the issue doesn’t lie in the fact that the output is 1% worse; the issue lies in me accepting x-1% over time, where the 1% degradation compounds. This 1% lenience tends to happen whenever we send a prompt off and switch tabs to let the LLM cook. It’s also worth thinking about whether this work is truly x-1%, or if there are biases that make it feel better than it is.
Firstly, I suspect that the degradation is greater than 1% in many cases caused by the dopaminergic nature of a task being completed by a non-human. It seems like there is a Pavlovian conditioning effect when a prompt leads to near-instantaneous results. This process resembles a slot machine, and certainly induces a dopamine surge that I suspect creates feelings of goodness as permutations of results flash on the screen. You must feel it too, the little excitement in your chest when you’re switching back to the LLM tab, thinking “let’s see what it cooked up”. That dopamine goodness feeling likely produces a positive bias towards the output, and acts as a slightly addictive, extraneous variable.
Secondly, there is an implicit positive bias whenever the model achieves anything above or matching our expectations. Many of us have been teachers watching a young child named LLM grow and learn and become smarter over the last half decade. The child started off generating simple diffusion pictures and using predictive text to write short stories, to now replacing entire jobs. Each time the models get better and learn a milestone task, our reactions are similar to that of the auntie who hasn’t seen us in a while telling us how much taller we’ve gotten. We are “pleasantly surprised” by an output that matches or exceeds our expectations, which is likely lower than what we expect ourselves to produce. If I expect myself to make a poster that’s 10/10, I expect the model to get 8/10; so when it gets 8.5/10, it feels like 9.5/10. I was talking to a friend about this concept, and she described it as a bad coworker who all of a sudden produced good work. We kind of make that face and whisper to ourselves “oh wow ok, that’s not bad…”.
I spent a lot of time generating animation character sheets in Higgsfield for a project. You feed the model a reference image, it generates something, and there’s a button to reuse that output as a reference. It’s crucial to avoid that button for modification purposes on the character sheet, because every time you rerun, any blemishes in the output compound. The pure white background at the beginning of sessions ends with all sorts of low resolution icky textures by the 15th rerun. It’s hard to spot as you’re progressing from generation #1 to generation #2, and it isn’t until you get to generation #15 that the artifacting becomes obvious. We become desensitized to the 1% degradation, and THAT is the biggest risk we run: atrophy of the mind causing our bar for excellence to slowly decrease over time—a process that runs passively and stealthily in the background until a friend sees our work and points out that it’s “lowkey kinda slop” which wakes us up.
I noticed this quite clearly with writing. I was writing a simple blurb for a sublet ad, and let the LLM write one in tandem. Reading what it generated, it was pretty good, but also pretty bad compared to mine. If I didn’t write my own, I would’ve leaned closer to the “pretty good” and riffed off an output that had one round of artifacting.
I’ve also noticed that it’s no longer okay to sit with a hard problem. We used to play with a hard concept in our heads for days, digesting it, even “sleeping on it”. Although no one is holding a gun to our heads saying we need answers now, prime-style, immediately, why does it feel like it? Why does it feel not okay to spend time on a problem? There is no excuse to waste time on a problem, since there is the less than semi-illusory truth that we have an abundance of answers, resources, and tools at the tips of our fingers. My friend describes how she sees learning as a rope that floats through space, with a camera following its unidirectional travel. As we learn and bump into new concepts, the rope bends. As we come across a complex, hard problem, the rope knots up. Now, we must untangle it. She describes how it seems as though only her highly academic/research-leaning friends will uncomfortably sit on the rope as it flies forward through space to untangle it, not expecting the camera to stop filming and time to stop moving. When the rope gets tangled for many of us, we feel the need for the rope to stop moving, camera to stop moving, for us to untangle it, IF we even feel compelled enough to untangle it. How often do we stay on the interface and research a topic deeper rather than sending a prompt off and changing windows to “do the next important thing”? My other friend describes this concept as calorically cheap vs. calorically expensive brain movements; it’s cheap to keep hitting the prompt button and much more expensive to just sit with the brain and think. Sure, think less, action more, but if the action is just hitting that slot machine over and over hoping for a lucky strike, the action here is just a copium illusion of real work being done. This is dangerous, because the low levels of copium “hard work” with high volumes of output mask what’s actually important to be working on.
In design, Josh Puckett talks about the importance of achieving conceptual range in ideation, and then refining one choice with conceptual depth. I’ve heard friends talk about using models for achieving great conceptual range, especially since they are exceptional at copying references. If the references fed are good (the human layer), then the model’s ability is elevated with expanded conceptual range. I’m noticing that the hard part is knowing when to bring the idea back to Figma to make it the best version possible. When standing at the fork of “that’s pretty good, let’s ship it” vs. “that’s pretty good, let’s now perfect it”, it’s a proactive and perhaps calorically expensive thing to take over and devote to conceptual depth. I’m not saying we need to always walk down the “that’s pretty good, let’s now perfect it” path, but we need to be cognizant of the fact that we’re standing at this fork more frequently simply because the fork approaches quicker with Godmode tools at our hands.
So how do we preserve our judgement, discernment, critical thinking, and clarity of thought while leveraging the powerful tools in front of us? Not using these tools is unrealistic and silly; they do indeed unlock creative range that would otherwise be unreasonably effortful. I definitely wouldn't have been able to code up the degradation effects of this page on my own, which is arguably cooler than the writing itself. My friend speaks about how this was likely the exact problem space during the industrial revolution, so history repeats itself and it’s nothing to be concerned about. That’s probably pretty accurate, and means it’s important to stay vigilant about the health and robustness of our minds and craft. Thoughtfulness about our craft and our relationship to the tools seems like an antidote to atrophy of mind. In an era where we can make literally whatever we want, what parts of the output or process are non-negotiables? Without getting into the Theory of Alienation, for us creatives, that means reflecting on the parts of our craft that are ends in themselves rather than means, and protecting their sanctity. For me, it means making physical arts like clay figures, where AI is temporarily uninvolved. Or, it’s redesigning a stylized y2k button from scratch. In this case, it means going full screen on my laptop, turning on DND, and refusing to hit “cmd” and “tab” until I finish the essay part of this project.
- Clo