I’m not super happy with this video, but it was take 4 and it’s late and so I’m moving on. I open with a (very generalized) metaphor to explain what an LLM is to someone who doesn’t know how it works at all: spatialized language. I refer to it as a “cube,” but I forgot to say the punchline; every point doesn’t have 3 coordinates, but 300; it’s a hyper-object beyond human comprehension.
Anyway, I then transition to this idea that these LLMs are being packaged as chatbots, which are basically “language slaves”; they only flicker on briefly to answer our questions. They don’t have (synthetic) consciousness, not because they aren’t advanced enough, but because we haven’t designed them to have it. If we gave them agency to make decisions and develop preferences over time, you might consider it a kind of digital being.
I’ve been toying with this idea of a “Montaigne in a vat” experiment. Montaigne was the first essay writer; he retreated into a tower for years to explore the limits of his consciousness through writing. Imagine training something on Montaigne, and then allowing it to perpetually prompt itself and test the limits of its own knowledge for 59 years. How many volumes of good writing might be produced?
This is an oddball ideas, and it doesn’t quite align with some of my other ideas around how writers will interface with AI. I think it’s slowly been percolating the more I learn about Montaigne. There’s a constant murmur over the rise of “AI agents,” and I sense that early open source experiments might have twisted objectives like, “make me a billion dollars and do whatever’s necessary,” but I think there’s something interesting to have an early agent shaped in the spirit of Montaigne.
Free the Chatbots