about
A calmer way to follow AI.
Caught Up is a selective daily publication about AI. Its automated newsroom reviews a curated set of sources, identifies what materially changed, and produces one edition designed to leave the reader fully caught up.
Why it exists
AI news repeats itself, overstates announcements and makes every release feel urgent. Caught Up selects fewer stories, explains the practical change and makes uncertainty visible.
How stories are selected
A committed source registry is scraped on a schedule. The pipeline weighs source diversity and relevance, then asks language models for structured editorial output. Git stores every published edition.
The automated newsroom
Curation, writing and optional illustration happen in GitHub Actions. The public site performs no per-request AI work and remains available if the operational dashboard is offline.
Models and responsibilities
Separate configured model profiles handle curation, long-form writing, translation and utility work. Each generated issue records the model identifiers actually used; the public site never guesses missing provenance.
Source policy
Primary sources are preferred for claims about releases and policy. Independent reporting supplies verification and context. Source weights guide selection; they are not truth scores.
Editorial rules
The publication separates confirmed fact, company claims, analysis, speculation and open questions. It avoids hype and does not claim completeness.
Corrections policy
Meaningful factual corrections are recorded on the affected issue and collected in the public corrections index.
Human review
Each issue states whether a person reviewed it. Automated publication remains possible; review status is never implied.
Cost and infrastructure
Generation cost scales with publishing frequency, model use, translation and illustrations—not readership. Missing usage is reported as unavailable, never zero.
Static by design
MDX and configuration live in Git, builds materialize the reader experience, and no runtime content database is required.
Accessibility and privacy
The site uses semantic HTML, visible focus, reduced-motion support and no advertising or behavioral targeting scripts.