I realize now I was my own LLM

My own notebooks — maybe seventy of them over the years — have always been where I build theories. But I realize now that what I was doing, in a sense, was using myself as an LLM: carrying on back-and-forth conversations about what was working, what wasn’t, and where the theory du jour might go next.

I often use actual LLMs that way now. They’re good at jogging my brain into the next step, usually by rephrasing my current state back to me in a slightly different way. They organize, compress, juxtapose, and remind. And then, after long back-and-forths in which I — never it, like… never — make some qualitative advance, they’re useful for summarizing the trail we just hacked through.

Eventually AI may get to the point where it begins actually making substantive theoretical steps itself, and even someday says, “Hush, dear, I got this.”

But we are nowhere near that yet.

For now, it is not the theorist. It is the notebook that talks back.