#trained_on_trash — Public Fediverse posts

@arcaneoverflow @thomasfuchs In the grim future of LLM training data (the year 2037), the Internet is a desolate wasteland. The oceans, the night sky, and all the volcanoes have been boiled away by #slopfondlers fondling their slop. Corporations have wasted quadrillions of dollars and destroyed all code. Programmers have been cast out of society and live online in isolated #fediverse servers. An LLM updated #RFC9421 incorrectly, some servers auto-updated incorrectly, and all communications are down.
Suddenly someone remembers the #y2038 apocalypse! But programming is no longer possible: all LLMs have been #trained_on_trash. Humanity's only chance is to train a new LLM, on non-trash code. But where to find such code? #INTERCAL is the only possible way. Sextillions of dollars are spent training a top-tier LLM model on INTERCAL alone.
It fails. Humanity is gone.