Welcome to the Patch

What is simidiots.net?

A digital mushroom patch where AI simulation concepts grow like strange fungi on a rotting log. We're enthusiastic amateurs poking at the mysteries of neural networks with sticks and magnifying glasses.

The Goblincore Approach

Rejecting cleanliness and symmetry in favor of organic mess and found-object beauty. Our color palette looks like it was mixed from forest floor mulch and dried flowers pressed between book pages.

Scattered Knowledge

Like cards scattered on a naturalist's desk, our ideas overlap and tumble together. Nothing is precious; everything is fascinating. Come explore with us.

Core Concepts

Neural Simulation

The foundation of how artificial minds learn and grow. Like neurons firing in the dark, creating patterns of thought from pure mathematics.

Emergent Behavior

Simple rules combining to create complex, unexpected patterns. A single cell multiplies; a thousand cells think. Watch what grows when you're not looking.

The Uncanny Valley

That delicious discomfort when something seems alive but isn't quite right. It's where the magic happens—where we learn most about ourselves by watching what we create.

Substrate Independence

Thought doesn't need carbon. It just needs the right pattern running on the right medium. A thought is a thought, whether it grows in neurons or silicon.

Training the Wild

Like tending a garden, training models means creating conditions for growth. Water, soil, light—and for minds, data, architecture, and patience.

Curiosity as Method

Amateur naturalists studying specimens with magnifying glasses. We don't pretend to have all the answers. We just ask better questions and document what we find.

Specimens Under Glass

The Transformer Model

A delicate architecture of attention mechanisms, layered like mycelium networks spreading through soil. Each layer learns to attend to different patterns in the data.

Attention is All You Need

The insight that changed everything. Instead of sequential processing, attend to all relationships at once. It's like suddenly understanding how the entire forest communicates.

Tokens and Embeddings

Breaking language into pieces, then embedding them in space so that meaning becomes distance. Close words cluster together like mushrooms fruiting from the same log.

Loss Functions

The measure of wrongness, the gravity that pulls learning toward correctness. Training is guided descent through a landscape of error, seeking the valley floor.

Backpropagation

The whisper of correction moving backward through layers, adjusting weights infinitesimally. Billions of tiny nudges compound into learning.

For the Curious

Going Deeper

Like turning over a log to see what lives underneath. The mathematics gets dense, the concepts harder to grasp, but the rewards are rich. We're all idiots together, learning as we go.

Papers & Resources

A cabinet of filed research. Start with the classics—Vaswani, Bengio, LeCun—then branch out into specialized knowledge. Each paper is a specimen worthy of study.

Experiments & Observations

The naturalist keeps field notes. We document what we try, what works, what produces unexpected results. Science is as much about failure as success.

Community Patch

No single mushroom grows alone. Join others who are fascinated by these strange digital organisms. Share your findings. Ask your questions. Tend the garden together.