How to use this wiki

Everything here is built from six AI-safety papers, taken apart into their bare-bones ideas and wired back together. This page shows what the pieces are and how to move around them. It reads top to bottom in a few minutes.

What the pieces are

The wiki has a small vocabulary, and every page is one of these kinds (the coloured badge in the top-right corner of a page tells you which one you are reading):

Concepts — the atoms. Each of the 303 concept pages explains one idea from one or more of the papers, in plain prose.
Edges — the wiring. Each of the 303 edge pages explains how two concepts relate; a few are marked hindsight, meaning the link only became visible after a later paper.
Themes — small groups of concepts that belong together, each with a guided walk through its members in reading order (58 of them).
Superthemes — groups of themes; the largest structures in the wiki.
Connective themes — lenses over the edges rather than the concepts: short threads of related connections.
Stories — each paper retold as a story: its own arc chapter by chapter, its ties to the other papers, and its thesis and contributions in its own terms.

The front page

Papers — where the front page opens: a card per paper, each linking to the paper’s own page — its stories, and every concept it uses in reading order. The best place to start reading.
Superthemes — the big structures, each listing its themes.
Connective themes — the threads of connections.
Concepts A–Z — every concept, alphabetically. Use it when you already know what you are looking for.

The address bar follows the tab you are on (#tab-concepts, #tab-superthemes, …), so you can bookmark or share a particular view.

Reading a paper as stories

Every paper has a page of its own — open it from its card in the Papers tab. A row of tabs at the top picks the telling: The big picture states the paper’s own thesis and contributions in its own terms, Inside the paper follows the paper’s own arc, and Across the corpus traces how it connects to the other papers. They all cover the same paper — choose whichever question interests you most.

The story sub-tabs with the row of 'Read at' granularity buttons below them — Pick a telling, then pick a zoom. “Read at” opens the whole story to one depth: *root claim* is a single sentence, *everything* is the full tree.

Each story is a collapsible tree. The Read at buttons set how deep it opens — from the root claim alone down to every concept and its connections. You can also open and close any single branch by hand with the +/− markers.

A last tab, The concepts, holds the paper’s full inventory — what it introduced, refined, and inherited, in reading order, each deep-linked into the PDF — with a Read at zoom of its own.

A story tree with chapters partially expanded, showing the +/− toggles — A story opened to chapter level. Click a marker to open one branch; the prose at each level explains what the level below contains.

Concept pages

The top of a concept page: title, origin line, summary, and introduction — The top of a concept page: which papers it comes from, a one-line summary, then a gentle introduction before the detail.

Every concept page follows the same shape. Prose first; then, after the divider, its place in the graph: Part of / Contains (parent and child concepts), Connections (its edges, each one sentence), Lenses (the themes it belongs to), and Sources. Highlighted phrases in the prose are links to other pages.

Links open in a popup

Clicking any concept, edge, or theme link opens the page in a popup over where you are, so a quick side-glance never costs you your place. Close it with the ×, the Esc key, or a click outside; open as full page ↗ in its top bar promotes the popup to a real visit. To skip the popup entirely, open the link in a new tab as usual (ctrl-click, cmd-click, or middle-click).

Sources go straight into the papers

A Sources block listing section and page citations grouped by paper — Every page ends with its sources: each entry opens the paper’s PDF at that page.

Nothing here is invented: every page cites the passages it is built from. The PDFs ship with the wiki, and both the Sources lists and the inline citations in the prose (the §…, p. N parts) open the right paper at the right page, in a new tab.

Small print

The wiki is fully self-contained — pages, papers, fonts, and math all work offline.
Without JavaScript everything still reads: tabs and trees simply stack as one long page, and links navigate normally instead of opening popups.
Lost? The Explore mark in the header always returns to the front page.