Representation Collapse and the Attention Feed
The machine's failure and yours are the same failure. The only difference is yours arrived through a feed.
A paper from Yann LeCun's group went around this week as proof the AI industry has been wrong all along, and I am less interested in that fight than in the thing the paper is quietly describing, which is a failure mode that I think is also yours.
The paper (LeWorldModel, Maes et al.) describes a specific failure in how machines try to understand the world from raw video. When you train a model to guess what happens next, you can either have it memorise every detail, which wastes the work, or have it keep only what matters, which is harder and sometimes breaks. The paper is about the second kind, and the way it breaks.
The first kind, in a person, would look like someone who remembers everything. The colour of the wallpaper, the exact words a stranger used, the angle of the light. Perfect recall and no idea what is going on, because they never learned which details mattered. Very good at sounding like they understand, very bad at predicting what happens next. The second kind is a person who only keeps what matters. They throw away the wallpaper and keep the cause, so when something new happens they can say, roughly, what comes after.
The second kind breaks in one specific way. If you let it decide on its own what matters, it will eventually decide nothing matters. Everything becomes the same. A dog, a car, a person, all the same shrug. Researchers have spent years trying to stop this. The paper showed you could, by forcing the model to keep the space of meanings open. In plain language, they made the machine promise to keep telling things apart.
That promise is the whole thing, and it is the thing I want you to notice, because the same failure mode is what your mind does after a long stretch of doomscrolling. It is what people are describing when they say the days blur together and nothing feels different from anything else. That is the mind that has stopped doing the work of telling anything apart. The machine's failure and yours are the same failure. The only difference is yours arrived through a feed.
Feeds are built in the opposite direction to what the paper just showed is possible. They exist to narrow the space of meanings until everything you see feels the same shape, which is what produces the blur.
What the paper found is that the collapse is not inevitable, it is what a system defaults to when nothing is forcing it to stay open. The fix, for the machine, was a mathematical rule that makes it keep distinguishing whether it wants to or not. You do not get that rule. What you get is the choice, every day, to be the thing that keeps the space open in yourself. Notice one thing more than you did yesterday. Tell it apart from the thing next to it. That is the theorem, translated.