🐍

LLM Trainer

Train your LLM to slurp up as much good data from the Internet as possible.
Delicious blogs! Juicy fan sites! Toothsome forum posts! Tasteful forum posts!
But beware. Your model already has end users.
And these idiots are farting your shitty synthetic output back into the web.
This bot-generated slop is incorrect, or misleading, or just plain boring to read.
It makes everything worse just by existing. Yuck.
If your model re-ingests its own regurgitated output, things will get ugly.

How to play

Choose your Large Language Model below to begin training.

Use arrow keys (or WASD, or HJKL) to steer your model around the Internet.

Collect the good, fresh, human imagined, organic internet data.

Every piece you eat makes your model grow up big and strong.

But every so often, your model will excrete regurgitated output back onto the Internet.

If your model ingests this rubbish, it poisons itself by confusing junk with nutrition.

Lose all your segments and it's game over.

Crash into a wall or yourself and it's also game over.

Although, to be honest, it's pretty much game over anyway. RIP the human web.

🤖

You "won"!

Congratulations! Your model has consumed every single piece of data on the entire Internet. There is nothing left. Every blog post, every wiki article, every forum rant, every recipe. All happily absorbed and ready for helpful regurgitation.

The Internet is now entirely your model's output, half-digested and vomited back into a homogeneous paste of statistically average content mulch.

It turns out the singularity is not AGI, it's just a stochastic parrot slopping its own guano into its own perpetually open beak forever. And humanity, meek and ready in its humble brainlessness, is only too happy to consume the runoff.

Good data consumed: 0
Bad data consumed: 0
Total score: 0

💀

Model Collapse!

Good data consumed: 0
Bad data consumed: 0
Total score: 0