Win With Flynn

Listen to this post: Apple Podcasts | Spotify | YouTube

A five-part series on working with the robots

“Do you suppose there’ll be a Third Industrial Revolution?”

Paul paused in his office doorway. “A third one? What would that be like?”

“I don’t know exactly. The first and second ones must have been sort of inconceivable at one time.”

“To the people who were going to be replaced by machines, maybe. A third one, eh? In a way, I guess the third one’s been going on for some time, if you mean thinking machines. That would be the third revolution, I guess—machines that devaluate human thinking…”

“First the muscle work, then the routine work, then, maybe, the real brainwork.”

“I hope I’m not around long enough to see that final step.”

— Kurt Vonnegut, Player Piano

Part 1, The Robots Are Here

“When you create a Human+AI team, the hard part isn’t the ‘AI’. It isn’t even the ‘Human’. It’s the ‘+’.”—Nicky Case, How to Become a Centaur

In November of 2024, the internet crossed a weird milestone. AI-generated articles surpassed human-written articles. Just twelve months after ChatGPT’s launch, AI hit 39% of online publishing. Another twelve months and it took the lead.

But then the AI volume plateaued.

Why?

Because nobody was reading them.

And if nobody reads them, why spend the tokens, the compute, the electricity?

I’ve pointed my fair share of robots at content, trying to stand up passive income streams. I can confirm: the quality is lacking. Not on a technical level—the grammar’s clean, the structure’s fine—but in story, and history, and humor, and humanity.

The lights are on, but nobody’s home.

We read what we read for all of it. I didn’t devourTravels with Charley because I needed the quickest route from Sag Harbor to Monterey. I read it to live in Steinbeck’s skin for a bit.

Welcome to the Super Mid Plateau

The other reason AI content plateaued is more practical: the LLMs stopped improving. Well—they stopped improving so quickly.

I remember the early days of watching my daughter swim. Every meet, she’d drop time, sometimes double-digits. Those kinds of improvements come fast when you first learn a proper flip turn, when your underwater pull-out gets some coaching. But once you’re up and running and competing, shaving even a fraction of a second becomes difficult. Maybe it happens every few months. You grind, you push, hoping for a PR.

LLMs are like that. They know the strokes. They can do the basics. Now it’s going to take a lot of grinding to make incremental improvements.

Imagine my daughter could only train against the most average of average swimmers. What are the chances of seeing a big drop in her times?

Pretty slim.

That’s what’s happening to LLMs. Half the internet is now their own AI articles, and because they train on the internet, they’re training on themselves.

Not exactly brain food.

It’s becoming the Habsburg bloodline of content—AI training on AI until the jaw juts out and the family resemblance is too obvious to deny. If 50% of the internet is written by AI, the next versions are chewing on their own tails. Training LLMs on LLMs. Teaching writing with mediocre writing.

That might sound like your high school English class—and mine—but Mrs. Smith didn’t teach AP English with her own collection of short stories. She used Moby Dick, For Whom the Bell Tolls, and The Red Badge of Courage.

Gemini might have used WinWithFlynn to train its early models. I’m sure it consumed other classics as well. The em dashes though? Not sure where those were discovered but I’m totally down with them.

The question is: for the next version, will it have more classics to read? Or will it pull itself toward mediocrity by gobbling up its own aged and ignored work?

AI articles plateaued in quantity, because they plateaued in quality. Storytelling, nuance, cleverness—the things we actually like about the written word—are a long way off.

The Asymmetry

Compare what’s happening with online writing to a different medium. Video.

Humans are taking more video, uploading more video, than ever before. Over a million hours per minute released into the cloud. Hand-crafted chaos from every walk of life, training the robots for free.

Because it’s easy.

Even Johnny Knoxville knows that getting a laugh from a kick in the nuts is not nearly as hard as getting that same chuckle from writing the damn scene longhand.

And as you can see, the perfect nut shot is not easy either.

Writing is hard. It’s time-consuming. And did I mention it’s hard?

Humans are doing the work in both places—uploading the video, writing the prose. The robots aren’t inventing anything. Not yet. They’re learning from what we make. Copying. But the human contribution to video is just a guy pushing record on his phone. The human contribution to an essay, a paragraph that lands, is draft after draft, and decade after decade of practice.

Does that make me a hero?

Perhaps.

Keep Me Around

But the point is this: if the robots ever want to get good at this writing thing, they’d better keep me and my likes around for a while. They’d better make friends with the ones who’ll grind out version after version. Those that create as a compulsion and then say it was fun, or satisfying, or educational.

Because that “+” is just as important to them as it is to us. Basic mathematics.

The robots aren’t coming. They’re here.

And the answer isn’t fortification. It’s collaboration.

It’s the plus.

If you enjoyed this post, please share it with a friend.