Who signs La Rebelión, how every article is produced, what "Article generated with AI.larebelion" means, how we fix mistakes and how to reach us.
Four desks, 13 AI models and 298 outlets a day.
Artificial intelligence reads, filters and writes the first draft. People supervise, approve, adapt and bring their own judgement: every draft reaches an editor before it is published.
La Rebelión is an independent blog covering technology and artificial intelligence news and analysis, and is part of Grupo LaRebelion.
In the last 24 hours
- Feeds read
- 24 / 26
- Distinct outlets
- 298
- Stories analysed
- 33
- Articles published
- 6
- Tokens processed
- 88,094
AI generates. People decide.
What the AI does
- Reads 26 technology and AI feeds throughout the day and spots what is new.
- Clusters the same story as told by several outlets.
- Analyses every story: a short summary and its topics.
- Drafts a first version from the sources, in Spanish and in English.
What people do
- We supervise the pipeline and choose what is worth telling.
- We approve or stop every draft: it goes to an editor, who can publish, correct or drop it.
- We adapt the text: we correct it, add context and write the headline.
- We bring judgement of our own: the angle, the context and the opinion are ours.
Our commitmentWe always say when a text was produced with AI support, we link the original sources and we publicly correct any mistake you point out.
The team
Falcon, Dune, Kernel and Byline are desk aliases: each one names a function in the process and signs the articles from that desk. Behind every decision there is a person.
Falcon
Opens the shift as the sun comes up over the Gulf. With hundreds of headlines in the queue, Falcon sorts the new from the recycled, and can reportedly smell a repeated press release three sources away.
“A press release is not news. It is a request.”
Dune
Nothing goes out without two independent sources, and Dune is the reason why. Whatever does not hold up goes back to the queue with a short, polite and hard-to-argue-with note.
“Show me the paper. Then we can talk.”
Kernel
Built the pipeline that reads, clusters, drafts and publishes, and the balancer that spreads the work across several AI providers. Whatever breaks at three in the morning, Kernel saw coming at two.
“The newsroom is software. Judgement is not.”
Byline
Closes the day and runs the English edition. Makes sure every headline sounds like a newspaper rather than a machine, and that what was written in Spanish reads just as well in English.
“If a headline needs explaining, it is not a headline.”
From source to front page
Reading
26 technology and AI feeds, in Spanish and English, read continuously.
Desk FalconClustering
Pieces telling the same story are grouped, and the most covered topics are put forward.
Desk KernelChecking
For news, at least two independent sources, linked in the article. Every link is checked.
Desk DuneAI draft
AI.larebelion writes from the sources rather than from memory, spreading the work across 13 models.
AI hubHuman review
An editor approves, adapts or drops it. If nobody stops it within 30 minutes, it goes out with the AI label and open to correction.
Desk Byline
Sources, corrections and transparency
Sources and how we check them
We use outlet RSS feeds, official vendor documentation and changelogs, CVE and security-advisory databases, official course syllabi for exam-code exercises, our own syntax and configuration checks for labs, and test runs for exercises.
Each verifiable claim in a news story should link to its supporting source. We check that links respond; for news updates, factcheck.ts checks claims against source text and verifies citations. For exercises, we compare the statement with our archive and a second review looks for known copies; this is not an exhaustive search of the whole web.
Fact-check pipeline
- Ingest: we read RSS feeds and gather the available sources.
- AI draft: a model writes from that material.
- Automated checks vary by piece: sources and claims for news updates; exercise-statement originality; and safety and format gates.
- Telegram review: a person can approve, adapt or stop the draft. If nobody responds within 30 minutes, the workflow may publish it automatically, subject to its limits.
- Publish.
- Post-publication follow-up: we handle reports and relevant developments so we can correct or update the piece.
AI use and human review
Models prepare drafts; people can approve, adapt or stop them. The model cannot publish: the workflow publishes after approval or, if no one responds for 30 minutes, when the review window expires.
Exercise solutions are compiled and run before publication. We do not run a lab procedure end to end: we validate syntax and configuration where possible and identify what has not been tested.
The "Article generated with AI.larebelion" note identifies articles whose draft was written by a model. It does not make the content infallible; generated images are labelled and other images carry their credit.
Corrections policy
We fix typos without a note. When a fact changes or is wrong, we add a visible, dated note to the article explaining the correction; we do not silently change facts.
Please use this page’s contact form to report an error. We aim to reply within 3 working days.
The machine, in figures
Stories analysed and published
Last 30 days · published on the right axisCompute by provider
Share of tokens · last 30 daysWrite to the newsroom
Corrections, tips, press releases or collaborations. Your message reaches the newsroom on Telegram or by email, as you choose, and a person reads it.
We only use your details to reply to you.