tech LLMs (Large Language Models) really took the world by storm in the last couple of years. ChatGPT has brought it fully mainstream and the results have been interesting. I had a strange entry into LLMs. Right around the start of my Second Impact I decided to dive directly into LLMs as both a curiosity to learn and a technology I wanted to help develop. I quickly realized that it was also an escape from the reality, that, at the time, I found so mundane and empty. My entry point was a handful of Discord servers each focused around some fairly popular LLMs and LLM frontends; KoboldAI, Pygmalion, NovelAI, Agnaistic, SillyTavern. These became new places for me to just hang out (see Chaos Spaces) and passively learn about the tech and the culture of the LLM space. As you may know, the primary focus of these is roleplay (something I knew very little about in any formal context before finding this). Popular characters are described in “cards” (PNGs of said character with their description encoded in the metadata) and then fed into an LLM of choice (local or otherwise) which is instructed to act as the character. The results are actually quite good; not perfect by any stretch, but sometimes it really can feel like you are talking with something approximating the character (and more importantly, a novel being). I was frankly enthralled by my early results, running small LLMs locally on my computer and timidly, yet excitedly interacting with them. The rest of my life was more or less on pause anyway and this consumed much of my time.