The Evolution of this Project

When I began to assemble this project into its final form, I had the intention of juxtaposing writing generated entirely by me with writing generated by an LLM tuned on a large body of my essays and papers. With over 300 essays and scores of academic papers to feed into a model, I had no doubt that this LLM would be able to convincingly impersonate my writing. Some early tests I ran on friends and peers confirmed that people who knew between very well and somewhat could hardly distinguish my writing from that of the AI version of me: AI-me. I wondered if, as some recent papers suggest, the AI might even seem more like me than me.¹³ Thus my challenge to myself was to try to write in a way that seemed more me than AI-me. I even considered turning this piece into an interactive website that would randomize the ordering of the AI-me generated content versus me generated content and let readers vote on which was human versus AI, with my goal being for own writing to be more likely to convince readers that it was written by me than my AI writing. Ultimately, I outdid myself.

By the time the first few components of this project were drafted, my original plan was to feed prompts into AI-me to write very similar pieces. The problem was, as I tilted my style towards creative writing, I could no longer generate reasonable prompts because my writings interwove so many personal stories. AI-me might understand my style better than I do, but it cannot access the full body of experiences and, yes, stories, that have constituted my life thus far. I beat myself at my strange game before it started.