Prototype It To Explain Itself

Neural networks ↔ teaching analogies

Each card pairs an architecture with a brain-inspired metaphor for learning. Flip to compare sides, then follow the links into the domain folders where the demos live.

Computer vision demos
Computer vision demosMultimodal spine
LLM prototypesNLP coherence & style
Related: sampling & probability (no dedicated demo yet)
Multimodal spine demos

Analogies inspired by public lecture framing. None of these pairings claim to describe how the brain actually works.

Reference pages

Canonical Markdown from domain folders at the repo root — llm/, computer-vision/, multimodal/, and the rest.

Architecture & the Predictor Seam
The narrow waist that makes the entire collection composable and polyglot-ready.
LLM Prototypes Guide
The living reference for the entire collection.
Sampling Strategies
Why temperature alone is not enough, and how Top-k / Top-p actually work in practice.
The 9-Prototype Roadmap
Why each prototype exists, what it teaches, and the production stack it mirrors.
Diffusion: Generate by Un-noising
Noise → structure: the generative vision loop made visible.
Computer Vision: See → Structure
Pixels become measurements about the world — not pretty pictures.
Gaussian Splatting: Scene as Glowing Blobs
An explicit, editable 3D particle cloud you can walk around.
Multimodal Spine: One World, Many Tongues
See → bind → show across vision, language, diffusion, and splats.
NLP Coherence and Style
How LLMs form sentences, why labs sound different, and how to ask for a voice.
The single-file Reflect and Attempt Quizz provides active recall over the same concepts.