How we write
Every explainer goes through the same steps. Nothing is published until each factual sentence is backed by its source.
Plain words, one diagram
- Explainers are written in plain words, for a curious reader with no background.
- Each one has a single diagram, drawn to show how the idea works.
Sources
- Official sources come first: the maker's own documentation, specification, model card, changelog, paper or repository.
- Every factual sentence is backed by an exact quote from a primary or official source.
- Every source is listed at the end of the explainer as a link you can open.
Original work
- Explainers use our own words. Source sentences are not copied or lightly reworded.
- Each explainer is checked for originality before it is published.
- Drafts are not published.
Staying current
Each explainer shows the date its sources were last checked. We plan to check sources automatically every night, to catch moved links and new releases. That is not running yet. Changes to the site are listed on What's new here.
How AI news is made
AI news is updated every day at 7 PM ET. Stories are found in official company newsrooms, GitHub releases, research labs, Hugging Face and developer discussion, but every story links to its original: the company's announcement, the release, the paper or the project. Press articles, newsletters and discussion threads are only used to find stories, never as the link.
Each story gets one plain sentence, written after reading the source. When a number is the company's own claim, the sentence says so. GitHub stars are copied exactly from GitHub on the day. Opinion pieces, rumours, funding news and minor bug-fix releases are left out, and a section can have fewer than ten stories rather than be padded. Past editions are kept unchanged in the archive.
Corrections
If you find an error, please tell us. We check the report against the sources, fix the page if needed, and note the fix on What's new here.