How Bright Learning works
Most of Bright Learning is written by an autonomous news pipeline under the editorial direction of Sam Naji. Some of it is written by people. We think you deserve to know exactly which is which, so here is the whole machine, in plain language.
The short version
Our main output is News on AI in schools, produced by an automated editorial system: sourced, drafted, fact-checked, scored, and released without any human approving it one by one. Every control a human editor would apply has been converted into an explicit, measurable gate that the article must pass. If an article can't pass, it doesn't run. We never lower a bar to fill a slot.
Not everything is machine-written. Research pieces are written by named people (Sam and occasional outside contributors) and published under their own bylines, outside the automated pipeline. Editor's notes are written and edited by Sam himself, and when an outside expert contributes a piece, it is their work, edited to our house style and published under their own name. Every article carries a byline and a short disclosure line telling you which of these it is, so you are never left guessing whether a person or the pipeline wrote what you are reading.
Who writes what
| Kind | Who writes it | How it is labelled |
|---|---|---|
| News | The autonomous news pipeline, under Sam's editorial standards. No per-article human approval. | The Bright Learning Newsroom byline, plus a line noting the article was produced by the pipeline. |
| Research | A named person, usually Sam, working outside the automated pipeline. | The author's own name, linked to their author page. |
| Editor's notes | Sam Naji, in person, outside the automated pipeline. | His own name, with the note "written and edited by Sam Naji, outside the automated newsroom pipeline". |
| Guest contributions | An invited outside expert. We may help shape and edit it to house style; the thinking and the words are theirs. | The contributor's own name. |
The steps below describe the News pipeline, which is our main output right now. The human tracks follow ordinary editorial practice. We have also run automated Research and Opinion pipelines and may again; older Opinion pieces stay on the site, archived rather than removed.
What happens before an article reaches you
- Signals. The pipeline reads a curated list of education and AI sources several times a day and clusters what matters for schools. Forum chatter can suggest a topic; it is never allowed to serve as evidence.
- Selection. The pipeline picks one story per slot from the ripest clusters. One article, one story.
- Investigation and drafting. Before a word is drafted, the story is pinned to at least two independent, primary sources; a single-source story does not get written. The draft is then written from that pinned evidence, with up to three revision passes if it does not clear the gates below.
- Verification. Names, quotes, dates, and figures are checked against the pinned sources. Quotes must match verbatim or they are removed. A story that only clears a single source, or whose claims cannot be traced back to the record, does not run.
- Gates and the release decision. Independent checks score factuality, policy compliance, taste, and slop (formulaic AI prose). A weighted composite decides the outcome: publish, revise (at most three attempts), or spike. A factuality failure blocks publication outright, whatever the other scores say.
- After publication. A published article is not re-checked on an automatic clock today; re-verification runs when we trigger it by hand. If a load-bearing claim stops holding, whether we catch it ourselves or a reader flags it, the article is corrected visibly, or unpublished, with a note in the corrections log.
What a human still does
Sam Naji is the editor of record. He sets the editorial standards the gates encode, reviews a weekly digest of everything the pipeline did (including everything it refused to publish, and why), owns corrections, and can halt publishing with one switch. What he deliberately does not do is approve articles one by one. The system is designed so that trust comes from verifiable mechanism, not from a tired human skimming a draft at 11pm.
Mistakes
We will get things wrong. The commitment is that corrections are fast, visible, and logged: a timestamped note on the article itself and an entry on thecorrections page. If you spot an error, the corrections page tells you how to report it; a substantiated report triggers the same correction machinery.
Why publish this page at all
Districts are being asked to make policy about AI while being marketed to by AI. The only honest posture for a publication that is mostly produced by AI is to show its work: what is automated, what a person wrote, what is verified, what a human owns, and what happens when something breaks. That's this page. It is part of the product, not a disclaimer on it.