You have noticed it. ChatGPT feels dumber than it used to. Your prompts that worked six months ago produce worse results now. The writing sounds flatter. The ideas sound safer. The internet itself feels like it is shrinking. Every article reads the same. Every email sounds the same. Every answer sounds like it was written by the same voice.
You thought it was you. It is not you.
Researchers at Oxford and Cambridge published a paper in Nature proving what is happening. They call it Model Collapse.
Here is the mechanism in one sentence. AI trained on AI-generated data gets dumber every generation until it forgets what real human data looked like.
The internet is filling with AI-generated content. Blog posts. Articles. Reviews. Comments. Social media. AI companies scrape the internet to train the next generation of models. Which means the next generation of AI is being trained on the output of the current generation.
Each cycle loses information. Not randomly. It loses the rarest, most unusual, most creative parts first. The researchers call these the "tails of the distribution." The weird ideas. The unexpected perspectives. The things that made the internet feel human. Those disappear first.
What remains is the average. The safe. The expected. The bland.
Then the next generation trains on that. And loses more. And the next generation trains on that. And loses more. The researchers proved this is not a slow decline. Major degradation happens within just a few iterations. Even when some of the original human data is preserved.
They tested it on large language models. On image generators. On statistical models. The pattern was the same every time. The output converges toward a narrow, flattened version of reality that looks nothing like the original data.
The lead researcher put it plainly. "Large language models are like fire. A useful tool. But one that pollutes the environment."
The pollution is invisible. You cannot see which sentence on the internet was written by a human and which was written by AI. Neither can the AI that is about to train on it. And once the tails are gone, they do not come back. The damage is irreversible.
This is not a prediction anymore. It is a diagnosis.
The internet you grew up on was built by humans writing things no algorithm would have written. Strange, personal, imperfect, alive. That internet is being diluted. One generation of AI at a time. And the models trained on what remains are learning a smaller and smaller version of the world.
Model Collapse is not a technical problem. It is a cultural one. The thing that made the internet worth reading is the thing that disappears first.
I never met Gordon Wood, but I have a story about him.
In one of my grad school seminars, we read Wood’s Creation of the American Republic. The sheer erudition and evidentiary depth of the book bowled me over.
Back then, before kids and before life accelerated to warp speed, I used to call my mother every Sunday to catch up. Lots of times, we ended up talking about what I was reading that week in my grad seminars or for leisure. Mom had an omnivorous mind, and she was always looking for something else to read. She was a true intellectual—curious about almost everything, always eager to integrate new arguments or ideas into her existing schemas of how the world worked or to have those schemas challenged and changed.
When we talked that particular Sunday, I think I tried to describe to her part of Wood’s argument about the relationship between the state constitutions during the Articles of Confederation era and the federal Constitution. Maybe I was tired, maybe I didn’t completely understand her questions, but the end result of the conversation was that Mom had questions about Wood’s argument that I didn’t answer satisfactorily. I told her that she should probably just read the book, and we said goodbye.
She did eventually read the book, but the next Sunday, Mom started our conversation by saying, “Well, I had a lovely conversation with Gordon Wood this week.” For a split second, I thought she was joking, but then I remembered who I was dealing with. I started to sweat. “How?” I asked. A whole variety of unlikely scenarios in which the foremost historian of the American Revolution and my mother, who lived in Wichita, Kansas, might have met ran through my mind. “Oh, I just looked up his office phone number on Brown’s website and called, and he picked up!” Mom said. I decided I would have to find another profession.
As it ended up, Gordon Wood spent about an hour on the phone with my mother answering her questions about the Constitution. Ever since, I’ve had a soft spot for the man when I imagine him picking up the phone in Providence and finding Becky Elder from Wichita on the other end of the line. His generosity in that moment spoke very well of him.
Rest in peace, professor.
@Kira_Nielsen_ Modern academic: “my heavily qualified argument marginally advances the sub-discipline”
19th century academic: “my treatise, inspired by yesterday’s walk, revolutionizes philosophy, religion, and music. My critics are intellectually and sexually disordered”
Nabokov on reading Don Quixote (publ. 1605–15) in the mid twentieth century:
"He has ridden for three hundred and fifty years through the jungles and tundras of human thought – and he has gained in vitality and stature. We do not laugh at him any longer. His blazon is pity, his banner is beauty. He stands for everything that is gentle, forlorn, pure, unselfish, and gallant. The parody has become a paragon."
@kirakarrchi If one must be lonely, one should be lonely: (1) in a massive library soundproofed against the rest of the world, (2) somewhere overlooking the Aegean Sea in the soft glow of the tragically beautiful Mediterranean sun, or (3) in the loud and lively chaos of Chinatown
The Matrix famously borrowed from Baudrillard's Simulacra and Simulation to imagine a dystopia in which we're at the mercy of intelligent machines. While that made for a cool movie, that misses one of Baudrillard's core insights: we are the authors of the dystopia, and it is here
I recently spoke with someone from Gen Z who just began dating someone new
He said he will upload entire chat histories of their texts and shared photos to ChatGPT and ChatGPT will analyze the relationship, break down their relational patterns and attachment styles, and advise him how to approach the relationship going forward for optimal outcomes, even crafting exact texts for him to send back to her
And it is good at it. It’s really accurate and *incredibly* good.
According to this guy, “all his friends are doing this now”
Wow.
Thinking back to that era of my life, I remember all the uncertainty of a brand new relationship. To relieve the uncertainty, we would go to our friends to share and get advice. The entire experience was human-based
This is so much easier and more powerful than that, and the entire experience is technology-based
Still trying to wrap my mind around all the implications of this. But a few initial thoughts:
1. Anything you text or send to another person can be turned over to AI without your knowledge
2. Are any of these relationships organic? What if both parties are doing this? Is this just one Chatbot dating another Chatbot with a human facade?
3. This doesn’t bode well for the traditional dating coach or marriage therapist career path