What We Built It Into

Anyone who thinks most of what these systems produce is garbage is not making a mistake. They are describing what is in front of them accurately. Reviews that describe nothing, articles that circle a question without answering it, books nobody wrote, and songs that sound like songs and go nowhere.

The interesting part is that this has almost nothing to do with what the technology can do.

Generation became nearly free, and copying was free already. Anything that cheap to make and that cheap to move will arrive in enormous volume, and it will get there before the careful version, which still takes a person a week. Platforms rank by engagement, because attention is what they sell, and the flood is aimed squarely at that measurement. The loop closes with nobody in it intending the outcome.

Underneath the technical achievement there is also work. The filter that makes a model safe to put in front of the public was built by people paid very little to read the worst material on the internet and label it, so the system would learn to recognise what it refuses.

And the buildings running all of this draw real power in specific places, which is where a national statistic turns into a local argument.

Then there is the part that closes back on itself. These models learned language from a web written by people. That web now fills with model output, which becomes the next training set, and the rare things thin out with every pass.

None of this requires a machine that wants anything. It is what happens when the price of filling a channel drops to nothing and every system pointed at that channel keeps doing exactly what it was built to do.