Editorial standards
These are the rules AIAIAI.FYI publishes by. Most of them are enforced in software as well as by editors, so they apply at 3 a.m. the same as at noon. If we break one, tell us at corrections@aiaiai.fyi.
Source tiers
Every source we monitor is assigned a tier before it ever reaches the newsroom.
- Tier 1: primary sources. The organization that did the thing, speaking for itself on the record: company blogs and newsrooms, product changelogs and release notes, government and regulator publications, court records, and research papers and lab sites.
- Tier 2: credible secondary sources. Established news organizations and specialist journalists or newsletters with a track record of accurate reporting and corrections.
- Tier 3: community sources. Forums, link aggregators and some commentary newsletters. Useful for spotting what's happening early. Never enough on their own to publish.
We review tiers when a source's track record changes. A source that publishes fabrications, or fails to correct errors, is demoted or removed.
Confidence labels: Confirmed, Reported, Rumor
Every news story carries exactly one label. The rules are mechanical, so you can predict the label from the sources listed under the story.
- Confirmed. At least one Tier 1 source, or at least two Tier 2 sources from different organizations (different web domains) reporting on the record. A claim that rests on unnamed sources stays a Rumor no matter how many outlets repeat it, until a primary source confirms it. Research preprints (such as arXiv papers) count as Reported until peer review or independent coverage.
- Reported. Not Confirmed, and based on a single credible (Tier 2) outlet that reports the claim on the record. It upgrades to Confirmed automatically when an independent credible source or a primary source is added.
- Rumor. Not Confirmed, and the core claim rests on unnamed sources, leaks or speculation, as reported by a credible outlet. We say so in the story's first lines.
What the labels mean, and don't:
- Labels describe how well sourced the event is, which is different from whether a claim inside it is true. If a company announces that its model is "the best in the world," the announcement is Confirmed; the claim is still the company grading its own work, and we say so.
- A research paper from a Tier 1 source confirms that the paper exists and what it claims. It does not mean the findings have been replicated.
- Labels can change as sources arrive. The story page shows when it was last updated and lists every source.
How we use AI, and how we disclose it
We are an AI news site that uses AI, so we owe you precision about it.
- What AI does here. A small model sorts incoming items by relevance, newsworthiness and importance. A larger model drafts our short news stories in both reading levels (red pill and blue pill). An embedding model groups coverage of the same event. We currently use Anthropic's Claude models through Amazon Bedrock and Amazon's Titan embeddings.
- What constrains it. The drafting model may use only facts stated in the source text. It must supply verbatim quotes from the source as evidence for its key facts, and our software rejects any draft whose quotes can't be found in the source. Drafts are also rejected for banned phrasing, copied headlines, close copying of the source text, or the absence of a concrete fact. A draft gets two attempts; after that it is held.
- What people do. Editors set the rules, choose and tier the sources, review held stories, handle corrections, and write the hand-curated guides (AI for Humans, The Operator, The Build), checking every figure against its source.
- Disclosure. Every story on the live wire and in the Terminal is produced by the automated newsroom described here, under these rules. This page is that disclosure, and each story links to its sources so you can check our work. Hand-written guides carry a "last checked" date.
- What AI never does here. It never invents quotes, never publishes a number that isn't in a source, and never decides on its own to publish a story that our hold rules say must wait.
The no-slop rules
Every story, automated or hand-written, must:
- Give the reader a specific fact (a number, version, date, amount or named result) or something they can use.
- Name and link its sources.
- Attribute claims ("the company says," "Reuters reports") when the source attributes them.
- Label estimates as estimates and never present one as a fact.
- Avoid hype vocabulary. Our banned list includes delve, game-changer, revolutionize, unleash, harness the power, seamless, cutting-edge, robust, transformative, paradigm, groundbreaking, mind-blowing and shocking, among others.
- Skip clickbait: no question headlines, no teased reveals, no exclamation marks, no emoji.
- Write in our own words. Summaries that copy long runs of the source are rejected.
We do not publish listicles without news, webinar or event announcements, product promotion without news, recaps of old news, or minor patch notes.
Sensitive-story holds
Some stories can hurt real people if we get them wrong. A story is marked sensitive if it involves allegations against named people, deaths or injuries, military targeting, claims about medical treatment, elections, or minors.
- A sensitive story is held until it has a Tier 1 source or a second independent Tier 1 or Tier 2 source.
- A story whose only sources are Tier 3 community sources is held until a credible outlet reports it.
- A story whose draft fails the quality gate is held for an editor.
Held stories are not visible on the site. An editor can release, rewrite or discard them. Holds expire after 14 days.
Corrections
We correct errors of fact promptly and visibly. See our corrections policy for what counts, how fast we act and how corrections are displayed. Report an error at corrections@aiaiai.fyi.
Conflicts of interest
- AIAIAI.FYI is published by IDFire Technologies Inc.
- We pay Anthropic and Amazon Web Services for the AI models and infrastructure that run the newsroom. That is a customer relationship; neither company has any say in what we cover or how. We cover both, and we apply the same labels and rules to them as to everyone else.
- Editors and contributors disclose to the editor in chief any financial interest in a company they cover and do not write or edit stories about companies in which they hold a material position. Nobody at AIAIAI.FYI trades on information before we publish it.
- We do not accept payment, gifts or free products in exchange for coverage. Review units, if any, are disclosed in the story.
- Sponsorships and advertising, if we sell them, are clearly labeled and kept separate from editorial decisions.
Affiliate independence
Some links on the site may be affiliate links, which can earn us a commission at no extra cost to you. Affiliate relationships never influence which stories we run, which tools we list, how we rank them or what we say about them. Tool lists in The Build and the chatbot comparison in AI for Humans contain no affiliate links. Details: affiliate disclosure.
Contact
- Corrections: corrections@aiaiai.fyi
- Everything else: hello@aiaiai.fyi