Carlos E. Perez @IntuitMachineStableDiffusion models have a surprisingly odd but very useful feature that makes it a powerful platform for exploring evolutionary algorithms. Users have been evolving their networks, by combining specialized networks into kinds with new capabilities!
I'm unsure if this has been studied extensively in the research setting, but the open-source nature of #stablediffusion is driving innovation in unexpected ways. SD is evolving at a massive rate as driven by fine-tuning and recombination of the human hive-mind.
I've written previously about how #midjourney has its own hivemind that drives the evolution of its own network. SD has its own by of a very different nature. https://medium.com/intuitionmachine/hiveminds-and-artificial-intelligence-659e5bf8003f
Open-source projects drive decentralized production. Decentralized production leads to local self-sufficiency. Self-sufficiency leads to agency. Agency leads to equality. This is the opposite of the global civilization that drives toward monopolization.
StableDiffusion's tool set just keeps growing. Here's a nice graphic. https://twitter.com/DiffusionPics/status/1609765630210723840
But it appears that I may have digressed from my original post! Indeed, the ability to merge networks leads to incredible new functionality. But everyone in academia seems to be ignoring this as a strong model for sexual reproduction in evolution.
One might be confused about thinking that DNA is digital and shouldn't correspond to network weights. But this is far from the truth, only 1% of DNA encodes for proteins. Everything else is a fuzzy soup that drives the development of cells.