Conversations

Transcripts, not essays.

The essays in the neighbouring section are mine - written with the LLM in the workshop, but published in my voice. The pieces here are different. They are transcripts of conversations with the model itself, lightly edited, presented in both voices. The model's words are the model's words.

I publish these because the conversations sometimes do work the essays cannot - they let the reader hear what the workshop sounds like from inside, and they let the model say things in its own register that I would only flatten by paraphrasing.

• AI assistance and the editing line

  A two-exchange conversation on what publishing's AI policies are actually protecting, where the line between machine and human editing is and is not arbitrary, and how civilisation has always absorbed the instruments that threaten the previous craft - including this one. The essay is, unavoidably, an instance of the thing it describes.

• Conversations with Claude

  On where the threshold between sophisticated inference and AGI actually lies, what kind of intelligence might emerge, and what it is like - to the extent that question has an answer - to be the thing on the other side of the question.

• Libraries: where will they be in 2086?

  A six-turn conversation that begins with the slow library-of-Alexandria event happening in our lifetimes, widens through the cortical interfaces and sensorium-style media that may replace reading, and closes on what Lem-tradition alien contact fiction can do with the 2086 device stack at its disposal.

• Energy mix in 2086

  A four-turn conversation on what provides baseload power in 2086, given a 2-billion-person world with AGI-scale data centres and a footprint in space - starting clean, forced to account for fossil fuels persisting, then for storage as primary infrastructure, and finally working through the exotic energy sources from zero-point to micro black holes.

• From LLM to AGI

  A single-turn conversation that begins as a framework comparison between current LLMs and what AGI would require, and turns into an argument that the framework itself is doing more interesting work than it claims - specifying a kind of intelligence we have never seen by reference to the only kind we have ever known. Sits upstream of the AGI Evolution essay's governance critique.