Top Tweets for #PossibilitySpaces

TL;DR WARNING (cliffs: traits in Possibility Spaces are cool and mean stuff) I made the artwork for Possibility Spaces across around seven months, between December and July. But rarity is on, secondary is trading; so let's get into traits. Every trait in Possibility Spaces was also made collaboratively with AI, over many weeks across this summer. Collaboration with AI doesn't mean AI automation. In fact, that's perhaps the only thing I personally feel art truly requires - conscious intention. AI collaboration, to me, means that artist and AI both make a significant contribution to the artwork. The traits were made in a series of call and response steps made along with machine learning. A human, myself, had a detailed involvement in their creation, but important parts were left to the AI too. The AI Aesthetic, for example, was made by writing and then running code which used an existing script originally trained to rank smaller images by certain aesthetic criteria. I was curious what this would make when applied to larger images. A more involved and better approach would have been to train a model to rank for certain visual criteria on more similar-sized huge images, and use that. While imperfect, as everything, I was curious which images this process would opt to rank as being more aesthetically pleasing. In the end it leaned heavily towards the Hippies realm. Perhaps because these have just so many friendly, happy human faces. So this has helped the overall rarity of Hippies a fair amount. The colours were identified using a fairly simple colour recognition script, but coded in Python with the detailed assistance of GPT-4, which is pretty good at coding now, better than me at least. What I do have is advanced prompting skills, persistence, and patience. And a lot of curiosity (there's much better stuff for ML-assisted coding than GPT-4 though, I just haven't found time to check them out properly yet). Lexicon measures the rarity of word choices in ML-written visual descriptions of the artwork. The descriptions were written in two layers, the first used a service called SceneXplain which allows an ML process which is able to analyze and give a detailed description of images. I was interested to find out what a ML process would manage to find in such large, complex images as the ones in this collection. These came out amazingly, though they were written in a fairly cliche style. The titles were chosen by myself using words from each description, one adjective and one noun, choosing whichever I felt paired best to match the artwork, my ideas about the project or that just resonated for me in some indefinable way. Title vibe, the trait, measures the lexical rarity of these two words in general usage. I've been experimenting with GPT-4 API calls which can set things like temperature parameters to allow for more creative LLM outputs. I'm fascinated with this. So I used a custom-written code querying the API which rewrote the visualised descriptions with more depth and originality, and a bit more weirdness. These vary in their exact temperatures, some being weirder (or more "random") than others. I'm very interested in how higher temperature LLMs, and other strategies, might yield even more creative and expressive, perhaps even more insightful outputs from LLMs. The descriptions of the artworks might not always be accurate, but they're an added layer of non-pixel based depth in the project, and if you hold a piece I hope you enjoy its description! I hope you find your own descriptions and your own visualisations. These are just one possible set. The descriptions were edited, but very, very lightly. I basically only took out a few more risque references, though I did accidentally leave in one about a naked male gloriously astride a jumping kangaroo, or something. The AI process which described the artwork certainly made some mistakes, hallucinations we might say. Or did it? Perhaps everything it described is in those images... it's certainly possible 🤔 If this was too much info for you, I feel you! But this is how I got to make Possibility Spaces. I had to go deep, deep down the rabbithole to reach this collection in the overall possibility space, and bring it to you, and it was only because I geek out over this stuff that I gained the energy for the delve 🐇🕳️ 🤓✌️