Adventures du jour!
The plans for today were derailed by a big Midjourney AI update that I’d been waiting for (v4). I spent most of my brain space today working with the new AI.
What I observed about the AI is one thing; what I observed about what I felt and thought about switching versions is another.
I expected that exploring the new version would be both enjoyable and somewhat challenging. The new version is ostensibly smoother and easier to use, producing more likable images than the previous version. (At least it seems like that was the team’s goal, given the “office hours” call I was on.)
Those expectations held true. It took a while to get a feel for the new version, but I eventually got it. I’m not really conscious of what I’m doing differently yet. That’s pretty typical of me, though; I’ll grasp something intuitively, use it, then have to double back and painstakingly reassemble what I know on a more conscious level. People don’t actually need to know what we’re doing in order to function at a very basic level just like we don’t need to know how gears work in order to ride a bike.
I think, as far as working in any kind of sophisticated way with an AI goes, being capable of continuously evolving one’s understanding of the system on a subconscious level, then extrapolating rules of thumb to pass on to less experienced users is going to be critical.
I’m sure I’ll be spending a fair amount of time figuring out what I’m actually doing in the coming days and weeks.
The surprising part of the whole experience (given that I was constantly evolving my understanding of the system, at least evidenced by how much simpler I was able to get my prompts and how well I was able to get interesting results without having to reroll quite as much) was how intensely involved I was with it.
When I first dove into the new version (about 12:30), my tongue went wullurp and my language centers shut down for a while. (I was on the phone during my first rolls with the new version, then tried to talk to Ray.)
At first I was just overcome with wonder and delight. Seeing the new responses I was getting was rewarding. During this phase, I rerolled a lot of prompts that I’d used before, so I could compare the old versions versus the new ones. This went on for hours.
Then I started putting together new images, combining new concepts and just playing. That took maybe an hour.
Toward the end of that, I started to become uncomfortable, bored, dissatisfied. I was getting good images, but something felt off. The more prompts I entered, the more uncomfortable I got. Should I still be doing this? Is this version so sophisticated that it doesn’t “need” me? That it doesn’t need to be understood? Will it stop producing surprising, productive results?
Why don’t I feel like I understand what’s going on anymore?
Once I was uncomfortable for long enough, I started becoming actually conscious of being uncomfortable. It was like being aware that you were talking to someone who was smiling and nodding, and not actually engaging with what you were saying. I want to say I became conscious of being uncomfortable about 5pm, with the actual discomfort probably starting around 2:30.
(Because Midjourney.com keeps a record of your prompts and their results, I can look back at the time stamps and go, “That prompt was disappointing, ugh” and “That’s the prompt where I got really upset.”
I was still very pleased with the results, mind you. I just became intensely uncomfortable.
“No! This isn’t what I want. This isn’t right. I want to go back to beta!”–I could think exactly that, at one and the same time that I was enjoying the rolls. It was weird.
What was also weird, was that I could see other people I know who were rolling prompts on the new version becoming ALSO intensely uncomfortable. It came out with different words and different conscious reasons, but it came out.–If anyone doing rolls was comfortable, I didn’t notice it, but then I was myself uncomfortable.
At that point, I decided I needed to find out how to “break” the AI, or make it give me surprising, rather than disappointing, results: to overwhelm it so that it was forced to give me what I didn’t know I needed, to stop from being my secretary and be an oracle and partner again. To make it respond like a person.
(I’m finding it logically weird to talk about the AI at that point as an it, but it feels correct.)
Anyway, I proceeded to screw around, trying to find ways to make the AI no longer be able to give me what I asked for. I started succeeding about 5:30.–I think most of the work was being done subconsciously earlier, while I was upset but not really aware of it. I entered a number of prompts that were clear with regards to physical objects that I wanted the AI to show, but unclear about how I wanted the AI to show them. The AI was unable to fulfill my request and went for funny instead.–It was at that point that the AI felt like a “person” again, a “she.”
(The gender there isn’t absolute, just subjective for me personally, as influenced by society, yadda yadda yadda.)
I ran more conflicting prompts, got more unintentionally funny results, then somehow managed to control the conflicts in the prompts until MJ gave me an absolutely lovely series of Absinthe ladies, about 6:30. But it still didn’t feel right right, just ehhh, sort of right.
I was listening to a friend’s video and he said something rather philosophical about observing himself, alongside a complex list of emotions. AHA. If there’s something bound to confuse an AI and force it to lose its ever-loving mind, it’s trying to observe itself thinking when it doesn’t actually think, as it were.
So far, the difference with the new version seems to be that it’s able to follow contextual clues better, so that even if you aren’t explaining yourself well, the new version behaves as though you did (much like a real artist has to at times).
I fed her the self-referential prompt and MJ lost her ever-loving mind.
First she tried to give me the physical objects without any context. I simplified the objects and she gave me one of the emotions that I fed her, along with a bunch of nonsense text that had nothing to do with the prompt.
I tweaked the prompt several times and finally got it down to the point where she could cope with the self-referential complexity, but all the images were fucked up. They reminded me pleasantly of the old Cronenberg Naked Lunch movie, if rather less intense. I added back in some complexity–not all the way–and got lots of nice rolls.
I felt a palpable sense of relief at that point. I got up and moved around after hours where I have no idea whether I did so or not. I know I made tea a couple of times, but I don’t really remember more than that. I moved from the kitchen table over to the bean bags at one point, I know that (mostly because I was in the bean bag when I got done with the self-referential prompts).
I could “feel” MJ as a person I knew, a coworker maybe. The previous version felt like a partner–much closer, that is–and this was more of a “new person on the team” vibe. But the sense of connection was restored.
Finally, it was time for what is apparently the WORST thing you can ask of any version of Midjourney: a piece of blueberry pie. She could NOT manage to make a slice of pie that wasn’t warped somehow. But the pies were all significantly better than the v3 or beta versions that I’d tried it on before (for example, all the v4 slices of pie were triangular and actual slices of pie, rather than a big square or a tube or something). The main issue this time was that MJ kept trying to put raw blueberries in with the cooked ones. Huge improvement.
After that, I felt released from the urge to resolve the discomfort I’d been feeling earlier, very much at ease. I haven’t rolled anything since then.
Other events of the day: more good stuff with opportunities and projects that I’m not ready to discuss; I did yoga and walked this morning and took pictures but have no real memories of doing so. Whenever I have to switch over from normal DeAnna to whatever this other mode is, the ADD stuff kicks in. I can do whatever needs to be done, but I may not form permanent memories. I’ll have to go over the project stuff (and should probably rewatch that video) and make sure I can still track what I did and said.
I *think* I know. But thinking I know and actually knowing are two different things, I’ve discovered.
Skipping fiction writing for today. I did some editing in between rolls, but that was about it.–I knew if I didn’t type this up now, I would lose any hope of conscious awareness for what I did today. I’d have done it and felt like I knew what I did, but I wouldn’t have, really. Tomorrow I can go back over this and make sure I still know what I feel like I know.
You know?