/posts

// A dev journal on Sabohat — a brain-inspired SNN AI.

─── [ 2 posts ] ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────

01.

LED — how a neural network worked without backpropagation

2026-05-07 · #research

In v35 I removed loss.backward() entirely — and unexpectedly the model outperformed BP. Here is how LED (Lateral Error Diffusion) came about, why it worked, and where its limit was.

02.

From Samantha to Sabohat — the start of the first project

2026-04-03 · #log

A film, a name, and the idea of building an AI that works like a brain. This is the story of how the Sabohat project began — why GNN, why biological mechanisms, and why the name changed.