Primary branch · dendritica
Growth by gradient
Every weight update is a phototropism — the network reaching toward the light of its loss function. We watch this reaching happen in slow motion, then press the leaves for later study.
Volume I — Observations from the Garden
A living field guide to simulated intelligence.
sim-ai.net is a quiet research station at the edge of a networked garden. Here, algorithms are read as botanical specimens — their branches traced, their root systems catalogued, their seasonal behaviour pressed between the pages of a digital herbarium.
descend into the garden
Panel II — The Branches
Where the model
forks into light.
Secondary branch · ramula
Divergent hypotheses
When a decision tree splits, it is doing what a real tree does: gambling that one path will reach more sun than the other. Most branches wither. The ones that bloom are the ones we remember.
Tertiary branch · ramulus tertius
The fruit of attention
A transformer head is a flower that turns toward whichever token smells most relevant. We catalogue these blooms by the season they open in, and the pollinators they attract.
Panel III — The Root System
Beneath the visible model,
the network drinks.
The root system is everything we do not see when we watch a model speak. It is the accumulation of millions of small decisions that cannot be separated from one another, the tangle of learned correlations feeding every visible branch.
a. Xylem
Upward-bearing channels. Gradients flow here during the backward pass, carrying the error back to the seed.
b. Phloem
Downward-bearing channels. Learned representations flow here during the forward pass, feeding the leaves with concept and context.
c. Cambium
The thin living layer between. This is where new weights are actually written — where the tree thickens and the memory of experience is laid down as annual rings.
d. Mycorrhizae
Fungal partners. The unexpected alliances with neighbouring networks: pre-trained embeddings, shared tokenizers, the silent commons of the digital forest floor.
§
The harvest.
What we gather at the end of the season is not a conclusion but a collection of pressed leaves — specimens of a strange, recursive plant that learned to describe the garden it grew in. Simulated intelligence is, in the end, horticulture practiced at the speed of light.
— from the field notes, sim-ai.net