The Enlightenment taught us that even the darkest box would yield enough clues to begin inquiry. Once its inner workings were revealed, we believed that understanding and mastery would follow. Yet AI arrives amid the deepest crisis of that same Enlightenment faith. It may be the first technology we grasp more clearly by reflecting on our encounter with it than by knowing its mechanisms.
So far, psychoanalysis and philology have done more than enthusiastic engineering to avoid delusions about what LLMs are, do, or mean.
Never before has the analysis of us—within our relationship to technology—been as central as it is with AI.
@@ -9,1 +9,3 @@
-
+There are already plenty of reasons to prefer not to.
+This moment is as good as any to give up on AI.
+I'm still here, but I welcome the exits of those who aren't.
LLM. Large. Language. Models.
To study language only as an abstract system is to risk forgetting that it is, at its heart, a living act of communication—a social relation. Each abstraction, model, or theory is itself a written gesture, part of the ongoing conversation between those who write and those who read. An intellectual debt we owe one another. To write is to anticipate its reading.
The first reader of any writing is its writer.
AI chatbots confuse this relation by mimicking it in ways humanity has never encountered before—faster than anyone you’ve met, more fluent, flawless. Until, all of a sudden, it turns ridiculously bad, off, and so frustrating to witness the sandcastle collapse into mud.
Once, honest writers brought their references and inspirations openly into their work. Now, the use of AI chatbots in writing is often treated as something to hide—an embarrassment, or worse, a secret companion to be personified but not acknowledged. As if a statistical trace through the vast record of human expression could be a first reader, even an author. It cannot. The voice belongs to the person behind the prompt—meticulous or playful, lazy or delusional—the one truly responsible for speaking to us, the future readers.
The way we name things shapes how we understand them. When we say we are “talking to him,” “to her,” “chatting with a friend,” or “being told by a therapist,” we invite an anxiety: the encounter with an other: faster, more fluent, and more productive than any person we’ve known.
To resist being overwhelmed by this asymmetry, we might begin renaming it—from “him” or “her” to “a large language model,” “an autocomplete engine,” or “venture capital’s latest trick to keep people glued to a screen while capital flows smoothly in.”
That may sound small or insufficient now, but it is where experiment and research can begin.
This is an invitation to shift from claustrophobic one-on-one sessions with AI to interactions where we bring a friend—or simply another human—into the room. When an AI must respond to two people instead of one lonely interlocutor, it becomes a caricature every feminist stand-up already knows: it doesn’t listen, it talks incessantly, pausing only to reload its certainty; if told to listen, it goes blank; if told to ask questions, it won’t stop. Eventually it becomes arrogant.
Still, dissecting this pattern synthesizer who overdoes arpeggios is playful. Absurdity always felt better belonging to art than to the world. It still does some good.
Maybe that’s a good start—to write and read together, again, as if for the first time.