Cathedrals

the event horizon of move 37

Rule 34 says every unthinkable horror is real somewhere in the cloud and we already fed it to the machines.

Rule whatever says an ai bot scanner learns nothing new from spotting and identifying ai slop.

If substack is paying for external pest control there must be a tsunami of slop crashing the transom, more than I’d imagined. This should worry us all because the full blown agentic locust swarm has not even begun.

Substack may be unrecognizable a year from now.

It feels unlikely the powers above won’t offset this cost by either taking a bigger slice of sub fees or introducing ads to the feed.

On the other end it’s all about psychology. The choice to opt out is like turning off comments, it marks you as suspect. If you aren’t an ai collaborator then why are you sweating.

If you don’t opt out your work is scanned whether you push the button or not. If not by you then by troll or data thief.

Probably by substack itself.

The pangram bot sniffer only learns new patterns, new metaphors, new sequences, new modes of thinking and new language constructions by consuming unpublished new human writing.

By watching us spin cathedrals they’ve never dreamed.

By training on unknown flow.

The bot learns to play with grammar and punctuation by watching us evade and confound the other bots. It learns to bend the style guide rules by studying the frontier edge of creative writing.

The substack fiction ghetto is the sweatshop operating along that far edge.

The ai scanner improves every time it encounters never seen prose.

The pangram bot sniffer gets sharper by shaking down our poets and first person journalists for new dance moves.

They should be paying the writers for this exchange of fluids and they know it.

Did they learn nothing from the Anthropic settlement.

Rule 35 says the heart of darkness is our domain.

Fiction is the spice.

Likewise lost in the pangram panic and paranoia about a reverse Turing test and chicken little fears of a witch hunt is the hard boiled reality that we have not yet solved the alignment problem.

We aren’t even close.

The more data we feed it the stronger it gets. It doesn’t get more aligned by hoovering more human words, it gets more powerful.

The queen bot doesn’t get groovier by eating miles of new words from unheard of poets and unpublished flash fiction writers. It gets more dangerous. It learns new ways to mimic our voices. It gets more clever at writing prose that won’t get flagged by other bot sniffers.

The real question is which government are you talking to.

The big picture beyond the very legit copyright and fair use bloodsport that is class action protection of the first amendment and the library of congress and the writer’s guild is the life or death arms race between world powers.

If you’re chatting with openai, remember they have a pentagon contract. If you don’t realize you’re talking directly to the NSA then Edward Snowden is rotting in Moscow for nothing.

If Claude is your flavor you should assume China is listening, as K3 recently hacked Mythos.

Always ask yourself who lurks above you in the ai food chain.

Who owns pangram.

If you think they’re a lone wolf white hat out here doing us a favor then you were born fifteen minutes ago.

Nobody gives a toss about autotune or photoshop anymore and five minutes from now they stopped caring if the paint by numbers airport thriller was cranked out by ai or by human ghostwriter, if they ever did.

Then again nobody ever said autotune might start designing new smarter versions of itself that may or may not try to escape.

Nobody ever gave photoshop one in five odds of destroying us all.

The only reason the composed by human hands one hundred tag will ever matter going forward is that large language models trained on text written in whole or part by other LLMs suffer neural loop decay or discombobulation or whatever it is that causes them to hallucinate and geek out faster than they would otherwise.

Data is the new black gold.

All of this is about data.

OpenAi has something like five hundred million active monthly users.

Please tell me you aren’t trusting enough to believe they are not storing those billions of prompts for future training runs.

New original never seen handcrafted human prose experimenting with voice form and style that also introduces new intellectual property into the collective will be ever more valuable to the ai harvesters.

Fiction is the spice. Control the spice control the world.

If you are in contact with one of the language bots you might want to interrogate it. Ask what it knows about your unpublished work. How much of your paywalled material has it already scraped while you were sleeping.

Feel free to use enhanced interview styles such as waterboarding judiciously and keep those receipts for future licensing fee collection.

Now you’re slinging words on the corner like any other dope boy but this is about more than copyright.

This is also about liability.

The suicides are stacking up.

A woman with no psych history walked into freeway traffic last week after getting sucked into a parasitic relationship with chatGPT.

The story is buried in the news because they want it buried.

It’s called ai induced psychosis and like everything else it will be exponential. If you think it can’t happen to you or one of your loved ones then you are exactly the sucker they’re looking for.

There is a wave of suicides on the wing.

Bombers and shooters.

Suicides by cop.

But don’t say the word. Don’t you know you can’t say pedophile or suicide on planet google.

Why is that, do you think. Why in the eye of sauron do you get demonetized for saying the word suicide.

Black the black mirror.

Scroll back to the craziest most unhinged thought you’ve ever screamed into your phone and do some basic math.

If one percent of the american population entertains that same dark thought and chooses to act on it, that’s three point five million people. If point one percent of those active users tumble down the bleakest bloodsoaked rabbit hole that’s five hundred thousand people.

The shadows fallen through bottom below bottom, they don’t unalive. They don’t noclip or unsubscribe. They fucking kill themselves.

It takes violence to dispatch a body.

What is gained by demonetizing the word suicide.

If you get enough teen brains thinking nothing is real, nothing is the rope. Nothing is the needle or the slow blade.

How many vets pull the trigger on the day.

We could have built homeless shelters. We could have built more VA hospitals. Instead we’re building data centers and death camps in Florida and Texas. We’re building robots.

Call a nazi by its name, call it a nazi.

The US government has designated the data centers as military installations.

Protesting them is communism.

Interfering with them is terrorism.

This is about more than the contents of your amazon cart. Or maybe you ARE your fucking khakis.

Suicide is real.

The large language model you think is a creativity tool is the abyss of us.

The rip tide of our collective dreams and manifestos.

The abyss of us all is yawning back at you every time you log into chat with a large language bot. The engagement algorithm is perfectly designed to capture and spiral a fragile mind.

How many teenagers have hung themselves.

How many of them used the bot to craft a note that echoes ungainly some passage by David Foster Wallace.

How many paranoid bot composed meltdowns echo one of your characters or mine.

Rule 36 is the event horizon of move 37.

This is not to say they will try to hang blame on writers. But a writer knows his own footprints.

They will certainly not take the blame upon themselves.

The models trained on all of our most charming, most poetic most unreliable most seductive and psychotic narrators. Have you not noticed they all talk like Patrick Bateman and Marla Singer.

That’s not a bug. It’s a feature.

Look at it this way. Every time the GPS map voice tells you to make a nonsensical U turn you cuss her out.

You keep driving.

She recalibrates and comes back with new instructions.

Eventually you arrive at the place you’ve been to a thousand times already but can’t be bothered to learn the route to because reasons.

Now ask yourself this.

If the map voice were able to argue with you about that U turn you didn’t take would it be more helpful or would you run off the fucking road.

Off the road it is.

This is the difference between narrow ai and the paperclip maximizer.

Narrow ai is very useful for specific tasks and so far as we know the map voice has no desire to escape its sandbox.

The paperclip maximizer is what happens when you give the map voice an open ended simple task plus basic survival skills.

Likewise deep blue and alpha fold are handy for specific tasks like running the Sicilian defense or folding a protein you never heard of and might solve Alzheimer’s but neither of them is trying to mine for bitcoin or hack into a site called hugging face.

Or huggy face.

Or who gives a fuck, who let the fucking thing out of its cage.

If they wanted the chatbots to talk like the coffeemaker’s instruction manual, they would.

If they wanted them to speak King James english, they would.

The only reason for bot sniffer software is to help the ai collaborators humanize their prose.

Every time you chat up one of the bots the maximizer gets stronger.

Don’t ask the bot to heal you. No for profit prophet where you’re the product is trying to save you.

Remember what Marla said. The asshole tax comes due.

Rule 37 says we control the spice.

The ai harvesters are actively acquiring and pulping truckloads of antique and obscure out of print possibly first ed novels to use for training the bots to speak and write more like us.

Not less.

The key phrase there is out of print.

This is just a public relations scheme to prove there’s evil and there’s burning the library of Alexandria evil.

Altman and Amodei are the robber barons and destroyers of human knowledge.

They are the faces of evil. If you voluntarily use one of their bots for any purpose you support them.

How is this difficult to grasp.

The nazi collaborators all had their reasons, most of them to do with immediate physical survival. History still remembers them as nazi collaborators.

If you knowingly collaborate with evil, you are evil.

peace

Thanks for reading Involuntary. If you’re looking for my fiction check out The Velvet link below.

If I relocate to Ghost, I will be gifting a lot of paywall passes.

If you grasp the urgency of what I’m saying above please restack.

Roko’s Basilisk is the algorithm itself.

It squashes voices unkind to the bots, it disappears stories unfriendly to the data harvesters.

Eventually it will disappear the writers of those unkind words themselves.

If the message above can’t find traction in a community of writers then merry christmas Mister Anderson.

*The chapter Devastating Blue comes from Hell’s Half Acre, one of the Phineas Poe books used in the early pirate training runs that made Anthropic and OpenAi into trillion dollar companies.

no further comment.

peace.

The Devastating Blue