A Theory of Dumb*

*Full title: A Theory of Dumb It’s not just screens or COVID or too-strong weed. Maybe the culprit of our cognitive decline is unfettered access to each other.

I wasn't sure if I should share this one for a couple of reasons. 1. I don't like sharing PDFs that come from behind paywalls.... 2. It can come across as quite condescending but I will admit it had me laughing out loud on a few occasions. The provocative tone/argument coupled with the dry sense of humour, is possibly what I enjoyed most of all.

Some of my favourite lines (which you might see highlighted in the PDF) I've listed below. I am interested in how you all react to it!

"Stupidity, like intelligence, is a nebulous thing, hard to define but easy to spot in the wild."

"The stigma that was once attached to ignorance has disappeared, and the loudest and least informed voices now shape the conversation, forcing everyone else to learn to speak their language."

"Not so long ago, the dolts among us were free to think their thoughts quietly to themselves with no easy way to share them. At worst, a person would usually just embarrass himself in front of his own family or bowling team. Bad ideas had a harder time scaling and reproducing, so lots of stupidity stayed local, and everyone else could happily overestimate the average person's intelligence because they saw less of it. …"

"The collective knowledge base has been overwritten with received wisdom, fourth-hand opinions, and counterintuitive pop social science that flatters our cleverness while sparing us the effort of understanding anything deeply."

"Shumailov had been thinking about a Catch-22 of artificial intelligence, which goes like this: LLMs require massive amounts of high-quality human-created text to train on, which AI companies collect by scraping the internet, often indiscriminately. Now, though, a growing share of the web's data is itself generated by AI; by one recent count, about half of the articles and listicles posted online in 2025 were written by chatbots. This means future LLMs will learn by ingesting the compressed and degraded knowledge produced by their predecessors and their own regurgitations will, in turn, train the models that follow. Over time, this feedback loop could rot the foundation of the original training data as traces of human thinking are diluted by machine-spun drivel."