The sun is hitting the window at that particular angle that suggests I should have been awake three hours ago or asleep four hours ago. It’s Monday, the day the world pretends to care about productivity, and I’m staring at a screen that’s brighter than my future, reading about the latest scheme to turn human sweat into digital code.
There’s a bottle of Old Crow on the desk. It’s about a third full, standing there like a sentinel guarding the perimeter of my sanity. I pour two fingers. It’s not going to make the news any better, but it might make the headache vibrate at a lower frequency.
Here is the gist of it, stripped of the glossy PR veneer and the corporate speak: OpenAI is hungry. The beast needs meat. And not just any meat—they want the seasoned, grisly cuts you’ve already chewed on. According to some leaked documents that WIRED got their hands on, the folks trying to build the mechanical god-head are asking contractors to upload actual work from their previous jobs.
Let that sink in for a second. Let it swish around your mouth like a bad Merlot before you spit it out.
They want you to dig through your hard drive, find that spreadsheet you built for the logistics firm that fired you in 2022, or that slide deck you made for the marketing agency that worked you to the bone, and feed it to the AI. They want the “concrete output.” Not a summary. Not a vague description of how you spent your Tuesday. They want the file. Use your old boss’s intellectual property to teach the new boss how to make your old boss obsolete.
I light a cigarette. The smoke curls up, grey and lazy, disappearing into the ceiling fan. It’s a perfect metaphor for what’s happening here. We are vaporizing the concept of ownership.
The stated goal is to establish a “human baseline.” They want to see how a “Senior Lifestyle Manager” plans a seven-day yacht trip to the Bahamas for a family of ultra-high-net-worth individuals—an actual example from their presentation—so they can teach the computer to do it better, faster, and without needing a lunch break or a Xanax.
Now, I don’t know about you, but I have never planned a yacht trip. My lifestyle management usually involves checking if I have enough quarters for the laundromat and enough cash for the bartender. But somewhere, there is a person who spent a week stressing over gluten-free catering for a billionaire’s brat on a boat, and OpenAI wants that person to hand over the itinerary so the machine can learn how to kiss ass with mathematical precision.
This is the part that kills me. They are calling this progress toward AGI—Artificial General Intelligence. That’s the holy grail, the singularity, the moment the computer wakes up and realizes humans are inefficient meat-sacks who spend too much time looking at cat videos. But to get there, they need to know exactly how we work. And how do we work?
We work messy.
I take a sip of the bourbon. It burns, a familiar, honest pain. Unlike the pain of reading about “Superstar Scrubbing.”
That’s the name of the tool they’re using. “Superstar Scrubbing.” It sounds like a detergent sold by a clown on late-night TV, but it’s actually a ChatGPT tool designed to help these contractors remove confidential information from the stolen documents. They want you to take that proprietary quarterly report, strip out the client names and the secret sauce, and upload the carcass.
Evan Brown, some lawyer who probably owns a suit that costs more than my car, told WIRED that this is a legal minefield. No kidding. It turns out that taking documents from your old job and giving them to a third-party tech giant is generally frowned upon in the world of non-disclosure agreements. It’s what we in the writing business call “corporate espionage with extra steps.”
But here is the beauty of the grift: OpenAI is outsourcing the liability to the gig economy. They are telling these freelancers, “Hey, make sure you delete the secrets first!” If something slips through—if a trade secret winds up in the training data because a contractor was working on three hours of sleep and too much caffeine—who takes the fall? The trillion-dollar company? Or the guy trying to make rent by selling his past labor for scraps?
I bet on the house. Always bet on the house.
It’s genius, in a twisted, predatory way. It’s the ultimate recycling program. You spend years of your life learning a trade, creating documents, solving problems, dealing with the nuanced nightmare of corporate bureaucracy. That experience is your value. It’s the only leverage you have. And now, they are offering you a few bucks to liquefy that leverage and pour it into the fuel tank of the engine that’s coming to run you over.
I need another drink. The glass is empty, and the cynicism is backing up in my throat.
Think about the quality of the data they’re getting, though. This is the funny part. They want “real-world tasks.” Have you seen real-world work lately? Half of corporate America is people pretending to work while scrolling Instagram. The other half is people fixing the mistakes made by the first half.
If they train the AI on actual documents from actual offices, they aren’t going to get a super-intelligence. They’re going to get a machine that knows how to use 12-point Arial font to hide a lack of substance. They’re going to get an AI that is an expert at passive-aggressive email signatures.
“Per my last email…” the robot will say, before launching a nuclear strike.
But let’s look deeper. Why do they need this? Why isn’t the internet enough? Because the internet is trash now. The internet is flooded with AI-generated garbage, bots talking to bots, and SEO spam. The “open web” has been poisoned by the very technology these companies built. So now, they have to go offline. They have to harvest the archives of the private sector. They need the emails you wrote when you were angry. They need the project plans you wrote when you were desperate. They need the raw, unpolished, confidential reality of human labor because that’s the only place left where “truth” exists.
And they are buying it for pennies.
The presentation mentions that contractors can also “share fabricated work examples.” So, you can either steal from your old boss, or you can lie. Those are the options. Steal or lie. Welcome to the future of ethical AI training.
I can picture the poor bastard sitting in a studio apartment, digging through an external hard drive from 2019. He finds a budget proposal for a mid-sized paper company. He opens it up. He sees the numbers. He remembers the stress, the late nights, the coffee that tasted like battery acid. He remembers the boss who took credit for it. And he thinks, “Yeah, screw it. Upload.”
It’s revenge, in a way. A petty, self-destructive revenge. You’re selling out the system that exploited you to a new system that will exploit everyone.
This whole “human baseline” concept is fascinating to me. They want to measure the AI against professionals. But what does it mean to outperform a human? If the AI writes the yacht itinerary in three seconds, it wins on speed. But does it know that the client’s wife hates the color blue and will throw a fit if the napkins are turquoise? Does it know that the “luxury concierge” business is 10% logistics and 90% therapy for rich sociopaths?
The documents don’t capture that. The documents capture the output, not the suffering. And the suffering is where the art is.
I ash my cigarette in a saucer that I really should wash one of these days. The smoke hangs low.
There’s a profound sadness to it. We are entering the era of the Great Strip-Mine. We’re not mining coal or diamonds anymore; we’re mining cognitive history. Every PowerPoint, every legal brief, every architectural drawing is just ore to be crushed and smelted.
And the kicker is, we’re doing it to ourselves. No one is holding a gun to these contractors’ heads. They’re doing it because they need the money. The economy is a tightening noose, and OpenAI is offering a stool to stand on, provided you hand over the rope.
It reminds me of the old days, sitting in dive bars, listening to old-timers talk about the factories closing down. They’d say, “They trained the guys in Mexico to do our jobs, and then they fired us.” It’s the same story, just different machinery. Now, we’re training the code. We’re uploading the blueprints of our own obsolescence.
Imagine the sheer volume of mundane, boring, confidential crap being funneled into these servers. Performance reviews. disciplinary notices. Marketing strategies for failed products. It’s a graveyard of corporate ambition. And from this compost heap, they expect a flower of super-intelligence to bloom.
Maybe it will. Maybe the AI will look at all this stolen work, all these thousands of hours of human effort condensed into PDFs, and it will realize something profound. Maybe it will realize that 90% of what we call “economically valuable tasks” is busywork designed to keep us too tired to revolt.
If the AI figures that out, maybe it won’t destroy us. Maybe it’ll just buy us a drink.
But I doubt it. The machine doesn’t drink. It just consumes. It eats data and craps out efficiency. It doesn’t know the feeling of a job well done, and it certainly doesn’t know the feeling of a job half-assed because you were hungover and just wanted to go home.
That’s the human element they’ll never capture. The laziness. The beautiful, creative laziness of the human spirit. The shortcuts we take not because we are efficient, but because we are bored.
I look at the bottle. It’s calling my name.
This news shouldn’t surprise anyone. It’s the logical conclusion of the data grab. First, they took the public internet. Then they took the books. Now they want the private files. Next, they’ll want your memories. “Upload a scan of your hippocampus for $50 so we can teach the model how to feel regret.”
And people will do it. Because $50 is $50, and regret is heavy to carry around anyway.
The “Superstar Scrubbing” tool is the part that really sticks in my craw. It implies that you can scrub the humanity out of the work and still have something valuable. But the confidential info—the names, the secrets, the specific strategies—that’s the context. Without context, it’s just noise. They are trying to build a map of the world by looking at it through a kaleidoscope.
So, here’s to the contractors. The digital scabs. The folks uploading their past lives to the cloud. I don’t blame them. Everyone’s got to eat. But let’s not pretend this is noble scientific progress. It’s a pawn shop. We’re hocking the family silver to pay the electric bill, and the guy behind the counter is a server farm in a desert somewhere.
I finish the glass. The warmth hits my stomach, a dull, reassuring fire.
The world keeps spinning. The algorithms keep learning. And somewhere, a “Senior Lifestyle Manager” is wondering why the new AI chatbot sounds exactly like her, right down to the specific way she formats her bullet points.
She’ll probably be fired next year, replaced by the very thing she unwittingly trained. And she’ll have plenty of time to sit in a bar like this, stare at a glass of brown liquor, and wonder where it all went wrong.
I’ll save her a seat. We can talk about the good old days, when theft was a crime and work was something you did, not something you uploaded.
The bottle is definitely lighter now. The day is moving on. The machine hums in the background, waiting for its next meal. I should probably get back to work, or what passes for it these days. But first, I think I’ll write this down. Not for the machine. For the few of us left who can still read between the lines without an algorithm telling us what to think.
Cheers to the end of privacy. Cheers to the end of ownership. And cheers to the great, scrubbing superstar in the sky. May it scrub us all clean eventually.
Source: OpenAI Is Asking Contractors to Upload Work From Past Jobs to Evaluate the Performance of AI Agents