A news wire for frontier AI, silicon, cyber warfare, and high-tech macro, seldon.report relies on software to do most of the reading. A person wrote the rules it follows. This page explains those rules in plain language, so you can decide how far to trust what you see here.
Every story on the site goes through the same three steps. The first collects the news. The second reads each article and keeps only real reporting. The third looks back over the whole day and picks the handful of stories that matter most.
- Step 01 Signal Collect every new article from the sources we follow. ~700articles read per day, on average
- Step 02 Scrutiny Read each one in full. Keep real news, cut PR and noise. ~170make it onto the live wire
- Step 03 Synthesis At day’s end, rank the survivors and write the Brief. 5lead stories per issue
We go out and get the news
We don’t take submissions or pitches. We collect the news ourselves.
We follow a fixed list of around 40 publications, plus the research blogs of the major AI labs. During busy hours the system checks all of them every three minutes. Overnight and at weekends, when less is published, it checks less often.
When something new turns up, we fetch the whole article, not just the headline. Menus, ads, and sign-up boxes are stripped out so that only the reporting is left. Anything we’ve already seen is skipped, even if it was shared under a different link.
Who we read
Wires & business press Reuters, Associated Press, Bloomberg, The Wall Street Journal, CNBC, Financial Post, Yahoo Finance · Technology Ars Technica, The Verge, Wired, TechCrunch, The Register, MIT Technology Review, IEEE Spectrum, The Information, SiliconANGLE, Hacker News · Chips & infrastructure SemiAnalysis, Semiconductor Engineering, EE Times, Tom’s Hardware, The Next Platform, ServeTheHome, Data Center Knowledge, DigiTimes · Policy, defense & security Politico, Semafor, Lawfare, Tech Policy Press, Defense One, CSIS, Institute for the Study of War, Euractiv · Europe & Asia Sifted, Caixin, Tech in Asia · AI labs OpenAI, Anthropic, Google DeepMind, Meta AI
The list changes now and then. We add sources that keep producing original reporting and drop ones that mostly republish other outlets’ work.
Every article gets read in full, and most get cut
This is the step that decides what you see.
Each article is handed to an AI model along with a strict editorial brief. The model reads the full text, not a teaser, and answers two questions. Is this real news on our beat? Is it trying to sell something? An article has to pass both to get onto the wire.
What gets in
- Original reporting with facts, figures, or documents behind it
- New AI models, research results, and safety findings
- Chip supply, export controls, and the race for computing power
- Major cyberattacks and security flaws
- Defense, geopolitics, and regulation where technology is central
- Energy, data centers, and the money flowing into them
What gets cut
- Sponsored posts, advertorials, and affiliate content
- Articles that mostly promote one company’s product or webinar
- Routine corporate announcements: hires, office openings, rebrands
- Minor spec bumps and incremental product updates
- Speculation with nothing new behind it
- Clickbait, outrage bait, and sensational crime stories
One exception: AI lab announcements. When OpenAI or Anthropic releases a model or publishes research, that is the news, even though it comes straight from the company. So lab posts about models, research, benchmarks, and safety work are judged on substance. Lab posts about hiring, partnerships, or marketing are cut just like anyone else’s.
What the model writes
When an article passes, the model writes two short things and files the story under one of six beats. The headline and the link always belong to the original publisher.
Example · how a story appears on the wire
Headline & sourceThe publisher’s own headline. Clicking it takes you to their article.
The company pushed back volume production by two quarters, saying lithography tools ordered last year will not arrive until late next year. Analysts expect the delay to tighten supply of leading-edge capacity through 2028.
SummaryTwo or three sentences boiled down from the full article.
The takeWhy it matters, in one or two sentences. It has to rest on facts or numbers from the article itself.
The six beats
- AI & Models
- Semiconductors
- Geopolitics & Defense
- Cybersecurity
- Infrastructure
- Markets
From a pile of stories to a front page
Approved stories go straight onto the live wire. Synthesis is what turns a day of them into an issue.
Grouping the same story
Big news gets covered by many outlets at once. When several articles describe the same event, we group them under one story so that it isn’t repeated five times. The number of outlets covering it is shown as Coverage, which is a rough sign of how widely a story is being reported.
The Signal score
Every story also gets a Signal score out of 100, a rough measure of how front-page-worthy it is. A second AI model gives the score by answering three questions about the story. You can sort the live wire by Signal if you only want the biggest stories.
- 45%How consequential is it?
- 35%Would a serious paper lead with it?
- 20%Is it new and urgent?
Closing the day
At midnight UK time the day is closed and an editor model goes through the whole day’s wire to put together the issue:
- Collect the day’s approved storiesEverything that passed scrutiny in the past day, from every source.
- Check what we already led withIt compares each story against the front pages from the previous five days. Follow-ups, reaction quotes, and other outlets re-reporting yesterday’s news are dropped, unless something has fundamentally changed.
- Rank by consequenceIt picks the five most consequential developments. There’s no quota, so if three of the biggest stories are about chips, three chip stories lead.
- Write the BriefOne sentence sums up the day, and each lead story gets a short line explaining why it made the cut.
- File it in the ArchiveThe finished issue is saved at its own permanent address and doesn’t change after that.
What you should know
- The summaries are machine-written
- The summaries, takes, and Brief are written by AI models. They’re instructed to stick to what the article says, but they can still get things wrong. For anything important, click through and read the original.
- We always send you to the source
- We don’t republish articles. Every story links to the outlet that reported it, and the original headline is kept.
- Nobody pays to be here
- There are no sponsored placements and no paid ranking. Promotional content is exactly what the scrutiny step is built to remove.
- The rules have a point of view
- Deciding what counts as consequential is an editorial judgment. Our rules favor hard news on our beat over trends, gossip, and opinion. Other people would draw the line somewhere else.
Questions
Which AI models do you use?
Scrutiny and the end-of-day Brief run on DeepSeek. The Signal score and the grouping of related stories run on TypeSafe Jev. We pick models based on how well they follow strict instructions, and we may change them as better ones come out.
Why did a story I expected not show up?
It was probably cut at scrutiny, either for being off-beat or for reading too much like an announcement. It might also come from an outlet we don’t follow, or it was grouped under another outlet’s version of the same story. The filter is deliberately strict, so it will sometimes cut a story that deserved to stay.
Why does the live wire look different from an archived issue?
The live wire is today’s stories as they arrive, still unranked. An archived issue is a finished day: the stories are grouped, ranked, and topped with the Brief.