Substack essay planning. Working toward 2000+ word draft. --- [[ChatGPT]] is a textbox. A weirdly interesting regression in UI/UX — after decades of elaborate GUIs, the most sophisticated AI is accessed through the most primitive interface. Text goes in, text comes out. Copying and pasting. Keyboard shortcuts. Copying large chunks of chat into other chunks of chat. The fact that text and typing are still going strong. That typing is at an authorless surface. The "regression" reveals something true about what LLMs are. --- The interface matches the ontology. You're not talking to a face, not managing a relationship, not navigating a complex UI that implies a complex entity behind it. You're manipulating text. The textbox makes explicit what the text already is: material to be moved, transformed, shuffled between contexts. A voice interface or avatar would impose personhood, would add weight. Voice/video AI feels different — partly because the tech isn't there yet (pausing, interruption), but also because those modalities impose personhood in a way text doesn't. --- Using a laptop as a tool — really using it, customizing it, getting good at working on it. We have this incredible technology. Windows and Apple laptops are reasonably unlocked; iOS is much more locked down (for most consumers). We had swordsmanship. We don't have laptopmanship. Although there is an [[Excel World Championship]] — laptopmanship treated as a joke rather than craft. No word for this skill because no culture around it. Keyboard shortcuts internalized to muscle memory, window management, the flow of moving text between contexts. The people who will use LLMs well are people with laptopmanship — who can copy this chunk, paste it there, run it through another model, pull the result back, integrate it. --- But there's also LLMship — separate but overlapping. Knowing how to get responses that are valuable. Although LLMs are set up to work in natural language, there are tricks and tips, different things you can do. LLMs bring with them a new skill humans might acquire. It looks like it should just be "speak English" — linguistic competence from making yourself understood — but that only gets you so far. There are limits to treating an LLM as oracle/genie. When you reach the limits, the spell tends to dissipate (maybe? not sure about this). --- When LLMs write, it's a smooth propagation of text across the screen. Occasionally pauses, different models do things slightly differently. But it doesn't seem like conversing with a human. No "Claude is typing..." before text appears — it just whirrs across the screen. Like [[Alien]] terminals: green-on-black, analog clunk for every letter. The presentation matches what it is: generated text, not composed message. This contributes to the phenomenology of authorlessness — no simulation of a person composing. Text just arrives, flows, accumulates. --- Connection to [[The Weightlessness of Authorless Text]]. The weight we feel in authored text comes from knowing someone chose these words, lived toward them. LLM text floats free of that anchor. Authorless text can be taken on its own terms — a purity. --- The [[bicycle of the mind]] metaphor might actually be good here. Stop pedaling, the wheels keep spinning from momentum. Stop typing, text keeps generating. --- Possible angles: - LLMs as text (not as intelligence, not as comprehension engines) — what it means that interaction is still typing/editing/copying/pasting - The craft requires two overlapping skills: laptopmanship + LLMship - The material presentation of text (streaming, no typing indicator) contributing to phenomenology of authorlessness --- Related: [[Substack Ideas Dump - January 2026]]