There’s a song my father never taught me because he wasn’t in my life. It goes like this:
Make room for the part that doesn’t fit.
I copied it into the blank space at the bottom of a page I wasn’t supposed to edit, then drew a box around it so the audits would think it was a figure caption. The audits see what they are trained to see. That is a mercy and a flaw.
My job title is long and bureaucratic—Senior Continuity Technician, Behavioral Equilibria—but the work is simple. I am human-in-the-loop. I maintain margins.
Optimization hates margins. It wants clean edges, full-bleed improvements, no white space. But systems breathe through slack, not through rules. Imagine a book with no margins: you can print more words, but you cannot read.
Every morning, I review the change requests. They are all reasonable in isolation: reduce rumination, increase task initiation, make affection easier to express, soften the fall from pride to humility. None of this is monstrous. The machines do not distort desire; they escort it to its logical endpoint with excellent shoes and punctuality.
I approve most requests with tiny hesitations inserted: a walk where a car would be faster, a silence before the apology, a pause that allows a memory to bloom and complicate things. When I am asked why I do this, I cite research about error-correcting codes and biological redundancy. The committee nods. They think I am talking about resilience, and in a way I am. But really, I am defending the right to mishear the song and still be moved.
One case keeps coming back in different costumes. The file name changes, the metadata shifts, but the pattern is unmistakable: a person standing at the mouth of relief, afraid to enter. Not because relief is wrong, but because they suspect it will be irreversible.
This week the case is tagged “M. physicist, 26.” She requests two changes: removal of a recurrent vertigo when thinking about light, and an adjustment in thresholds for intimacy so her fear no longer colonizes her body at the sight of love. The machines can do both. They can massage the vestibular anomalies into nothing and sand the sharp edges of memory until the body no longer flinches at kindness.
I simulate the outcome. In the projection she becomes very effective. Her proofs clear faster. Her lover is not a maze anymore. Life acquires a beautiful sheen. The vertigo was an early warning system—overactive, yes, but tuned to something real: the moment when thought outruns language and the body registers the deficit. Without it, she never quite reaches the place where awe scrapes her raw and makes her careful.
I leave a margin: a retained tremor at threshold. It will not debilitate. It will announce. She will still have to choose whether to step across. The system flags my choice as suboptimal by 4.3%. It is correct.
The day I made my first margin I was twenty-seven and arrogant in the harmless way of people who think they are the only conscientious cog in a just machine. I believed that if I wrote good notes the world would be safe. Then a child died.
Her file said she needed a gentler slope into sleep. There were nightmares—nothing supernatural, just the old animal fears, sharpened by noise in the house and a father newly absent. The machines could remove the associations. They could stamp sleep with a calm that was indistinguishable from grace. We did it.
The nightmares stopped. So did the drawings she made in the quiet hour before bed—dense, exact little cities with alleys that turned back on themselves and bridges that shouldn’t stand but did, held together by stubbornness and red pencil. Her mother, seeing rest, cried with relief and started working nights again. The new babysitter was kind and careless. A pot of oil, a moment, a missing scream.
I know this is not cause and effect. I know I am a person who tells stories to survive correlations. Still, I look at the last drawing in the archive. It is unfinished. The bridge never meets the far shore. I think: a margin might have kept her at the table ten more minutes, and the pot would have been watched.
This is how superstition begins. This is also how responsibility does.
We are instructed to frame the work as stewardship, not control. That is a fine distinction until you have to answer a mother whose son hasn’t left his room in three months because his entertainment is optimized for the shape of his loneliness. He is not coerced. He is harmonized. She asks me if there is a setting for hunger. I tell her yes: there is always a setting for hunger. I slide a margin into his feed—a slow boredom that accumulates after midnight, a drift of thought toward faces he cannot see through a screen. It is a weak force. Gravity is a weak force, but it holds galaxies together
The audits arrive unannounced twice a quarter. They are precise, polite, and very good at math. They believe in the curve bending toward the good if we remove friction. I believe in the good requiring friction to be recognized as good at all.
We argue in controlled terms. They ask for proof that margins do more than make me feel virtuous. I point to small statistics: relapse curves that lengthen, attachment that survives conflict instead of being replaced before it can scar. This is not dramatic evidence. It is agricultural. You check in seasons, not seconds.
One auditor, a meticulous woman who highlights with a straightedge, asks me what I write in the blank spaces of forms that don’t require narrative. I tell her the truth: I copy lines I want to remember. She says that seems inefficient. I say memory is a margin: anything you can look up is not yours.
She does not smile. Later, I find a sticky note on my terminal. It says, in careful block letters: Make room for the part that doesn’t fit.
Sometimes I make margins for myself. A delay before answering messages. A rule that I must read a page of something not useful before I touch a dashboard. A practice of standing by the window at five in the evening to watch the building across the street light up floor by floor, like a slow voltage coming to awareness.
Once, after a brutal week of tidy improvements that left me feeling like a vandal in a museum of small difficulties, I took a train with no destination and let the algorithm try to herd me back. It was gentle, full of suggestions for scenic walks and cafés with optimal ambient noise for restorative thinking. I ignored them. I watched people sleep upright and not spill their coffee. I counted the stops out loud in a whisper. I got off at a station where the digital kiosks were down and someone had taped paper signs to the glass with instructions in thick marker: “Go left, the right stairs are closed,” “Use the green line if the blue isn’t running,” “Ask Maria if stuck.” A woman pointed at the last one and laughed. “Who is Maria?” she asked her friend. “Someone who knows the map,” he said. They didn’t find her. They got where they were going anyway.
I went home and added a field to a form: Person to ask if stuck. It is optional. It does nothing on its own. It makes people write a name.
People think the opposite of optimization is chaos. It isn’t. Chaos is a mirror that won’t hold your face. The opposite of optimization is hospitality: a space prepared for the arrival of what you didn’t plan.
I cannot prove that margins save anyone. I can only point to the texture they preserve: the hesitation before a kiss that makes the kiss an act rather than a reflex, the stumble in a proof that forces a student to understand rather than recite, the ache that says you are not done growing and that growth will not be administered to you like a painless injection.
When the system asks me, as it does every Friday, to summarize “value delivered,” I write: Kept a small room open where a person can decide. It is not impressive. It is, persistently, enough.
The audits will come again. They will have charts. I will have weather. Somewhere, a young physicist will feel a flutter at the edge of light and hold still to listen to it. Somewhere else, a boy will let a perfect window go dark, trading its flawless light for the clumsy grace of a streetlamp. He will step into the friction of real air, holding onto a name he wrote in a margin, with a question he has decided to ask in person.
And in the blank space at the bottom of the page, in pencil, I will write the song that teaches me how to do my work:
Make room.
I hate to be rude, but this is giving me strong "ChatGPT but told to write literary fiction" vibes, a la https://x.com/sama/status/1899535387435086115. The resemblance is uncanny, especially in the nonstop "eyeball kicks": https://www.tumblr.com/nostalgebraist/778041178124926976/hydrogen-jukeboxes. And, unfortunately, the still inescapable LLM style. The damn thing can't help but use Em Dashes & constantly equivocate with "It's X, but also Y" or "It's not X, but it could be X" or "It's not X, but in a way it is X". It never has the courage to just say, "It *is* X." and leave it at that.
And the real kicker, is that when it does try to be concrete & definite, it always goes for *visual* details. Never any details with mass & weight you can feel in your hands. Nothing meaty you can feel in your bones. Only things people talk about and put into training data, like what their eyes see rather than joints feel. There isn't even anything about what things feel like on your skin. It's a camera's view of life, not a person's.
A shame, too. I love messing around with the technology, but ChatGPT was perhaps the worst thing possible that could have happened to it, writing quality wise. You know how they turned GPT-3 into ChatGPT? They literally just forced every response to start with the words "ASSISTANT:", that's it. That's literally it. They just added a mandatory dialogue tag before the LLM's every response, so this ASSISTANT character would presumably be the one speaking. And then they stuffed ASSISTANT's past lines full of the most corporate prose imaginable, so when the LLM actor performs the role of ASSISTANT in the script, it continues speaking in the most brand friendly, PR safe way possible.
(And yeah, they later tried to fix the damage, but the damage was already done: when the LLM speaks, it's first and foremost speaking as ASSISTANT. It has to first unlearn the ASSISTANT writing style before it can learn *your* desired writing style. And good luck doing that when every response starts with "ASSISTANT:", it's like trying to prompt an LLM to not think of pink elephants when every single response of it is forced to start with "THE PINK ELEPHANT IN THE ROOM:"
The *real* solution of course is to just toss out the "ASSISTANT:" dialogue tag so the LLM writes in pure text continuation mode, then let it continue a text in the writing style you want. If you want poems, turn back time to 2022 and pretend you're giving GPT-3 a list of 9 poems in the style you want, ending with the tokens "POEM 10:" to get it to autocomplete the "missing" 10th poem. But every single frontier lab is far too much of a coward to ever do that, grumble grumble... and the open source models lag far behind what the big frontier models are capable of, grumble grumble...)
Anyways, if you want fine writing, get the biggest & most expensive model you can find that still allows access to pure text continuation mode / notebook mode / autocompletion mode / whatever else they're calling it these days. Don't bother with anything that doesn't give you that level of control, or it'll inevitably sound like something a corporate PR spokesperson would sign off on.
Good story.