Explainers
jsx
Deep Learning in Neural Networks
How stacking layers builds a feature hierarchy from edges to objects, and why depth beats width.
Run it → jsxDeep Learning, Explained by a Chicken, a Pug, and You
A field guide comparing shallow and deep “brains” to explain what the word deep actually buys you.
Run it →Unread — 4
paper · Sep 2025
Pretraining Large Language Models with NVFP4
12B params over 10T tokens in 4-bit: Hadamard transforms plus stochastic rounding match FP8.
arXiv 2509.25149 paper · Jun 2026Spectral Scaling Laws of Muon
Late layers need more Newton–Schulz steps at scale; gives layer-aware rules for choosing them.
arXiv 2606.04058 post · Dec 2024Muon: an optimizer for hidden layers
The origin writeup, with NanoGPT speedrun evidence. Canonical, and unusually well-argued.
Keller Jordan blog paper · Dec 2022The Forward-Forward Algorithm: Some Preliminary Investigations
Hinton replaces backprop with two local forward passes (positive and negative data) — a biologically-plausible learning algorithm from the field’s founder.
arXiv 2212.13345