How Epoch is made

Epoch is a newsroom for artificial intelligence. It reads roughly a hundred sources — lab blogs, changelogs, release tags, regulators, status pages, model registries and the press — and it does so on a clock measured in minutes rather than hours, because on this beat the gap between something happening and anyone writing it down is where the story is.

What it finds is clustered into a single item per event, checked against the source the claim originally came from, and scored on how well established it is. Most of what arrives never becomes anything. That is the point: the system is built to discard, and the small number of stories that survive are the ones several independent sources agree on and a primary source confirms.

Who writes it

The pages are drafted by an editorial pipeline and read by a person before publication. Nothing reaches this site automatically. Articles carry the byline Epoch Newsroom rather than a name, because a byline is a claim that someone stands behind the words and there is no individual reporter to name.

Every article ends with the pages it was written from. Those links are taken from the sources the story was assembled out of — they are never generated — so you can check the claim against the thing that made it. If an article and its sources disagree, the sources are right.

What it is bad at

It is a summariser working from public material. It does not have sources of its own, it does not attend briefings, and it cannot tell you what a company meant rather than what a company said. Where reports conflict, it says so instead of resolving them, which sometimes makes for an unsatisfying article and is still the right answer.

Dates are the other honest weakness. Some pages carry no publication date at all, and where one cannot be established the story is ranked as old rather than assumed to be new.

Corrections

If something here is wrong, it should be fixed rather than argued about. Mail [email protected] with the article and what is wrong with it.