How this site is made
Updated September 12, 2026 · the map · latest
A map of 127 open questions about AI, each with a short framing of the live debate, curated sources, and a maintained record of substantive recent discussion. The taxonomy — the sections, the questions, what is included and deliberately left out — is the editor's.
Discovery runs over a fixed registry of publications, researchers, and aggregators (listed in the site's public repository), supplemented by targeted search and editor submissions. It is deliberately bounded: counts and coverage describe what the project monitors, not everything posted online. Discovery runs weekly; full reviews of each question run on their own cadence — faster for fast-moving questions.
A piece is tracked when it would matter to a thoughtful reader following that question: new evidence, a distinct argument, a serious rebuttal, a measurement, a well-reported development. Quality of thought over fame of author. Every tracked piece is fetched and verified — the piece exists, the author, venue, and date are right — before it appears anywhere. Social-media posts are included only via a verifiable secondary source.
Each question's page features a small set of pieces, organized editorially; the rest of what crossed the threshold appears under “Additional relevant discussion” and in Latest. Featuring reflects significance, author diversity, and the shape of the debate — not a verdict that other pieces failed.
This is an AI-assisted publication. Discovery, verification, and first-draft synthesis (including each question's “What changed” note) are performed by AI — Claude — operating under a written editorial policy, with the editor directing and spot-checking. Every question page shows two dates: latest sweep (when the AI pipeline last fully reviewed it) and its editorial review status. “Editorial review pending” means a human editor has not yet examined that page's selections and synthesis line by line; pages flip to “editor-approved” as that happens. One question (3.2.1) additionally carries a long-form entry with its own review status.
The optional discussion-volume numbers (“N in 90d”) count tracked substantive pieces in a rolling 90-day window — one count per piece, assigned to one primary question. They are a lens on where attention is concentrating within the monitored sources, not a measure of total online discussion, and counts across broad and narrow questions are not perfectly comparable.
Disagree with a framing, spot a stale claim, or know a piece we missed? Suggest a source or correction. The main criterion for additions: what does this add that the existing material doesn't?