hr-studio
AI · 08
The Primer  /  Chapter 08

AI

An overture on how machines learn at all — then the generative quintet: embeddings, autoencoders and VAEs, diffusion, GANs, and attention. How a space of designs is learned, and how new ones are sampled from it.

An overture + 5 sectionsEvery idea is yours to move
0.
The Loss Landscape
The overture — error as terrain, learning as descent. The critical rate, the fold, and the memorization trap.
Open
i.
Embeddings
Meaning as a direction in space.
Open
ii.
Autoencoders & VAEs
Squeeze it through a bottleneck and see what survives.
Open
iii.
Diffusion
Learning to undo noise, one step at a time.
Open
iv.
GANs
A forger, a critic, and a live pattern studio.
Open
v.
Transformers & Attention
How a thing decides what to look at.
Open
All chapters