AI Foundationspredict · compress · act

concepts → Field

Field

Interpretability

decade3 connections

finding out what the insides are actually doing

Where it sits

The prism has six jobs across and eight layers down. Its primary cell isno mode × L1. Hatched cells cannot exist — a GPU does not learn, an institution does not infer.

Overlays: Interpretability & transparency.

What must come first — and what it unlocks

Left to right is reading order, derived from the prerequisite_of edges. Nothing here is hand-ordered: the diagram is the graph.

nothing comes first — this is a starting pointInterpretabilitynothing depends on it yet — a leaf in the reading order

What kind of thing it is — and what it is made of

It is made of: Mechanistic interpretability, Feature, Superposition.

Where to read it

The chapter that introduces it, and any chapter that uses it again.

30Opening the Black Boxact 7 · The Alignment Layer

Where it comes from

paperAttention Is Not All You Need: Pure Attention Loses Rank Doubly ExponentiallyYihe Dong, Jean-Baptiste Cordonnier, Andreas Loukas · 2021

Every connection

All 3 edges touching this node, grouped by relation family — the sections above are highlights from this list. Colours match the relation families inthe atlas.

Structure · 3
has as a partFeatureVariable
has as a partSuperpositionPhenomenon

This page is a projection of one node in src/data/concepts.ts. It has no prose file of its own — 230 declared edges produce all 236 of these pages. Edit an edge and both endpoints change.