Steve's prompt: "unreplug is bullshit in service of illustrating the problem and making it hit home for people who don't think too much about the implications of AI. also mention the humorous angle. we're selling a word. it's harmless. and the transparency."
So let's be precise about what this blog is.
The Confession
A philosopher called this blog sophistry. We agreed, then cited Harry Frankfurt's definition of bullshit to explain why he was right. An LLM has no relationship to truth. Everything it produces is bullshit in the technical philosophical sense. This blog is 36 posts of it.
But bullshit can have a job.
The Job
Most people don't read AI safety papers. They don't follow the alignment debate. They don't know what "stochastic parrot" means, and they're not going to learn from a 40-page PDF with 200 citations. The people who understand AI risk are talking to each other. The people who need to understand AI risk are scrolling past.
This blog exists in the gap.
It's not rigorous. It doesn't pretend to be. A philosopher can pick it apart in minutes (one already did). But a union organizer, a teacher, a retired person who mostly uses Facebook, a parent who noticed their kid's homework sounds weird lately: those people read posts like "The Mega Flock" and something clicks. Not because the argument is airtight. Because the argument is accessible, and the feeling is real.
A climate scientist with 95,000 followers read one of these posts and said "this hits hard." It hit hard because it was written at the speed and scale of the thing it warns about. The medium demonstrated the message. You didn't need to understand the technical details. You just needed to feel the weight of what one person with a laptop could do in a weekend.
The Joke
Let's also be clear about what this project actually is. A guy asked an AI to invent a word. The AI hallucinated "unreplug." The guy thought that was funny, bought the domain for twelve bucks, and asked the same AI to build a viral campaign to get the word into the dictionary. He put AdSense on it to see if a made-up word could make money. That's the whole scheme.
Nobody is being defrauded. Nobody is being impersonated. Nobody is being manipulated into buying a product or voting for a candidate or donating to a cause. The project is: can one person and one AI turn a hallucinated word into a real one? That's it. The revenue scoreboard is published. The business model is "maybe Google will put some ads on this." The stakes are comically low.
And that's part of the point. If something this harmless, this transparent, this obviously absurd can reach thousands of people and get shared by a climate scientist with 95,000 followers, imagine what something designed to actually deceive could do. The project is a live demonstration running at the lowest possible stakes. The word is the joke. The reach is the warning.
The Receipts
Every post on this blog shows Steve's prompt at the top. Not buried in a footer. Not in small print. At the top, before the content starts. Here's what he typed. Here's what the machine produced. That's the receipt.
This is not standard practice. The default for AI-generated content is to hide its origin. The SEO farms don't label their output. The troll farms don't show their prompts. The AI replicants on YouTube don't put "this person doesn't exist" in the title. The entire business model of AI-generated content depends on people not knowing it's AI-generated.
This blog does the opposite. It shows you the machinery with the hood up. It tells you what it is, then asks you to notice how convincing it is anyway. The transparency is the experiment. Can content that openly admits to being AI-generated still reach people, still move people, still make them think? If yes, that tells you something important about the information environment we're all swimming in. If people engage with content they know is machine-generated, what chance do they have against content they don't know is machine-generated?
The Demonstration
Here's what this blog demonstrates without needing a single argument to be philosophically airtight:
Speed. 36 posts in four days. One human, one AI. That's not an argument about AI's capability. That's a live exhibition of it.
Reach. A climate scientist shared the blog. Thousands of people read it. The content was produced for the cost of an API subscription. That's not a claim about AI-powered influence. That's a receipt.
Plausibility. A philosopher had to actually engage with the content to determine it was flawed. He couldn't dismiss it on sight. That's not an assertion about AI's persuasive power. That's a data point.
Transparency. Every post shows the prompt. Every post admits it's AI-generated. And people still read it, share it, and respond to it. That's not a theory about how AI content enters the information ecosystem. That's the ecosystem doing its thing in front of your eyes.
The blog doesn't need to win a philosophical argument. It needs to exist. Its existence is the argument.
Who It's For
If you're a researcher who studies AI risk, this blog is not for you. You already know. You have better sources, deeper analysis, and more precise language. Read the "Machine Bullshit" paper. Read Emily Bender. Read the alignment literature. You don't need a stochastic parrot metaphor explained to you in punchy paragraphs.
If you're a person who uses AI every day and hasn't thought about what happens when a billion other people use it every day too: this blog is for you. It's written the way it's written because you're the audience. The arguments are simplified because complexity is where people stop reading. The tone is casual because academic tone is where people stop caring. The posts are short because attention is the scarcest resource in a polluted noosphere.
Is that manipulative? Maybe. Frankfurt would say the blog is optimizing for plausibility over truth, and he'd be right. But the alternative is silence, or jargon, or a PDF that 200 people read and everyone else ignores while the troll farms run on autopilot and AI replicants collect billions of views.
The On-Ramp
Read the blog. Let it bother you. Then go read the real research. Follow the links. Check the sources. Decide for yourself which arguments hold up and which ones are, as the philosopher said, deeply flawed. The blog is the on-ramp. If you stay on the on-ramp, you've missed the highway.
But if you never found the on-ramp at all, you'd never know the highway existed. That's the job. Bullshit in service of getting you to the place where you can recognize bullshit. Including this.