The Gulag Got Better Snacks
Meta broke the trust of its AI workers, then promised smaller manager ratios and better snacks. The machine age still runs on human misery with a pantry budget.
The boy at the next table had a laptop open and the face of a man waiting for a priest.
He was young enough to believe a deadline was an emergency instead of a weather pattern. His coffee sat untouched. His fingers hovered over the keys. Every few seconds the screen brightened his glasses and he made the same small wounded sound, like a tire losing air.
Then he typed seven words into the little box and the machine began to answer.
You could see his shoulders drop.
Not because the answer was good. He had not read it yet. The relief came first. The relief came from watching paragraphs appear where panic had been. That is the trick. The machine does not have to save you. It only has to look like it arrived with a shovel.
I have known that feeling.
Not from the machines. From forms. From official envelopes. From the post office supervisor who handed you a clipboard and suddenly the chaos had boxes. Route late? Check a box. Truck broke? Check a box. Man inside you hollowed out by fluorescent light and low-grade despair? No box for that, but the form looked complete, and in certain buildings completion is close enough to God.
Now the students have found their own form.
They feed the prompt in and out comes language with clean shoes. It walks upright. It says however and therefore and ultimately. It knows where to put the commas. It can sound like it spent the afternoon in a library instead of being assembled out of statistical fog by a machine that has never dreaded a grade, disappointed a mother, or stared at a blank page long enough to start hearing insects in the walls.
The researchers have a name for this: the fluency trap.
Good name. Too polite, maybe, but good. A trap made of smoothness. A pit with carpet. The words read so well that nobody notices the floor is gone.
That is the trouble with polish. People mistake shine for substance. A cheap shoe can gleam. A coffin can gleam. A politician’s teeth can gleam while the town burns behind him. We are a species easily impressed by surfaces, which explains most of advertising and at least half of marriage.
The machine knows surfaces.
It can give you the tone of thought without the ache of thinking. It can produce the shape of an argument without the dirty little engine underneath. It can write a sentence that stands there confidently with its hands in its pockets, even when the sentence has no idea what street it is on.
And confidence is dangerous.
A bad sentence used to limp. You could hear it dragging a foot. A weak idea came in smelling nervous, eyes darting, trying too hard. There was mercy in that. The flaw announced itself. A teacher, an editor, a tired friend with a red pen could point and say: there, that part is bullshit.
Now the bullshit gets a bath.
It comes out smelling like lavender and institutional excellence. It has transitions. It has balance. It has the soft dead voice of a person who has read every brochure in the hotel lobby and learned nothing about the weather outside.
The kids in the study thought the machine would cut their workload. Of course they did. Everybody wants a little door in the wall. Everybody wants the secret handle that opens into the room where the hard part has already been done by somebody else. I wanted it at twenty. I wanted it at forty. I want it now some mornings when the page sits there white and smug and I would rather put my head in the oven than explain one more brave new improvement from the people who made life worse yesterday.
But writing is not typing.
That is the old insult nobody wants to hear. Writing is deciding what matters while every cheaper part of you begs to be excused. Writing is finding out that your first idea was a drunk with a fake mustache. Writing is cutting the sentence you loved because it strutted around too much. Writing is realizing you do not understand the thing you were so ready to explain.
The machine can help with the typing.
Fine. Let it. I am not carving commandments into stone tablets behind a gas station. Use the tool. Ask it questions. Make it spit out options. Let it show you the boring version so you can hate it properly. A hammer is not immoral because a fool hits his thumb.
But do not hand it the thinking and call yourself efficient.
There is a difference between using a machine to test your idea and using a machine to avoid having one. The first can sharpen you. The second turns you into a foreman supervising an empty lot.
The researchers found that the students who got better stopped treating AI like a vending machine. They learned to push back, revise, question, check, steer. They learned, in other words, that the miracle still required the old human labor: judgment, purpose, shame, curiosity, taste.
Those words do not scale well.
Judgment is slow. Purpose is stubborn. Shame is expensive. Curiosity wanders off the approved path and comes back with mud on its pants. Taste says no to perfectly adequate things, which makes it hated by managers and beloved by anyone who has ever made something worth keeping.
The machine does not generate purpose.
That line sticks in me like a fishbone. It can generate a memo about purpose. It can generate a lesson plan on purpose. It can generate five inspirational quotes about purpose suitable for a company retreat where everyone wears the same lanyard and pretends the muffins are not damp. But purpose itself is a live wire. It comes from need, fear, love, revenge, hunger, guilt, all the ugly little animals running loose in the basement.
A model has no basement.
It has weights. It has data. It has the collected noise of us ground into pattern. That is impressive in the way a casino is impressive: lights, math, air conditioning, no clocks. You can lose a whole afternoon in there and come out poorer but convinced something important happened.
The danger is not that students will cheat. Students have always cheated. Monks probably cheated on Latin verbs and blamed damp parchment. The danger is that they will stop recognizing the feeling of their own unfinished thought. They will accept the fluent thing because it lowers the pressure in the room.
And why stop with students?
Offices are full of adults doing the same trick with better chairs. Reports, proposals, strategy notes, performance reviews, apologies, wedding toasts, condolences. The smooth lie travels well. It fits every container. It never sweats. It never admits that the person sending it did not know what he meant until the machine meant something nearby for him.
Maybe that is the future: not robots kicking down the door, just everyone speaking in finished paragraphs nobody quite inhabited.
A civilization of clean drafts and dirty minds.
The kid at the next table finally started reading what the machine gave him. His face changed. Relief became suspicion. Suspicion became work. He leaned forward, deleted a line, frowned, typed something of his own.
Good.
The trap had opened, but he had not climbed all the way in.
Outside, the morning trucks dragged themselves down the street. The coffee went cold. The screen kept glowing, patient as a liar.
Source: Are you using ChatGPT or Claude for writing work? A study says you may be landing in a fluency trap
Meta broke the trust of its AI workers, then promised smaller manager ratios and better snacks. The machine age still runs on human misery with a pantry budget.
The feed promised to know us better than we knew ourselves, then sold our own appetites back wearing a dead person's sunglasses.
KPMG let the machine invent corporate reality, then discovered that authority without verification is just a parrot in a necktie.
Meta wanted superintelligence and found the old factory floor waiting inside the laptop: watched hands, dull tasks, and people drafted into feeding the machine.
The AI boom ran face-first into the oldest problem in the world: somebody has to wire the building, weld the pipe, and keep the miracle from overheating.