From even before the time when ChatGTT was released, I had fantasies about creating some sort of a chatbot-like version of me. This was a sort of undisguised fantasy about a species of immortality that I craved. My mom died when I was 19, and she left behind a mountain of writing. Not published writing, all unpublished. And a tragedy that’s hard for me to characterize, or even remember, without being suffused by rage. My father threw out substantially all of what my mom had written. Less than a week after she died. Today, I have one chapter that she wrote, and a couple of letters, and no more.
From the moment my child was born, I was kind of resolved not to deprive her of my output in the event of my death. This isn’t out of a grandiose sense that any of what I’ve written is particularly important, but out of a commitment not to repeat the trauma I experienced by losing so much of my mom, of her writing.
I don’t know that I actually would have read much or all of what she had written, but I certainly would have, from time to time, dipped into it, looking for answers to questions. I don’t imagine my child would particularly relish the opportunity to read the bulk of what I’ve written in this blog. It’s not the kind of stuff a kid wants to read by their dad. At the same time, my kid is increasingly not a kid. As she reaches adulthood, I could imagine she might begin to feel different about some or all of the content here. I know that when I was my kid’s age, I would have recoiled if I were to have read anything my mom had to say about sex. But that was then. So, I would have recoiled in my late teens and early twenties, but by the time I was in my thirties, never mind now, in my mid-late fifties, honestly, I think I could handle it. Anyway, I had that fantasy about creating some sort of a chatbot that could speak in my voice, and ideally with my voice, long after I passed from this mortal coil. The early days of LLMs had convinced me that that particular fantasy was not likely. Even the most sophisticated imitators of my voice do a shitty job, both at capturing my style and at anticipating what I might actually say in response to any given question. In recent days, though, I’ve embarked on a bit of a new experiment with respect to a chatbot. Not trying to imitate my voice, but trying to give access to what I’ve said in a useful way. This is a project of Claude Code, and I’ve been using a combination of tools provided by OpenAI and Anthropic to do it. It’s very much a work in progress, and I expect that it will continue to improve.
In my first iteration, I had it speak in my voice, or in the first person. That was a mistake, because the whole uncanny valley aspect of it rendered it absurd. It just so clearly wasn’t me, it so clearly was some sort of simulacrum, it was just spooky. So I’m getting rid of that, and instead trying to make it be nothing that it isn’t, and precisely what it is. Give it a whirl, and let me know your thoughts.
