About

This site publishes automated summaries of AI industry news. Every digest is written by a language model from sources fetched that morning, and goes live without a person reading it first.

How a digest is made #

The pipeline runs unattended on a schedule. It pulls a configured list of RSS and Atom feeds, then runs a handful of targeted web searches over the previous 48 hours to catch developments the feeds missed. ArXiv listings go through a keyword filter and then a relevance pass, which drops any paper whose actual result cannot be stated in concrete terms — a paper that can only be described in the abstract is one the digest has nothing to say about.

Everything that survives is deduplicated against a record of previously ingested URLs, so an item that has already appeared does not come back the next day. What remains is synthesized into the digest you read, and committed straight to the site.

Each digest records the model that wrote it, alongside the date and reading time.

What it covers #

Model and product releases, research findings, security incidents, funding, infrastructure, and policy — weighted toward what someone building with AI would act on, rather than everything that happened.

Digests come in three cadences, linked from the header: a daily covering the previous day, a weekly published on Saturdays, and a monthly published on the first of the month. The weekly and monthly are not concatenations of the dailies; they are written fresh to pick out what turned out to matter once the noise settled.

Tags are applied by the same model that writes the digest, drawn from the vocabulary already in use so that a topic stays findable under one name.

Sources #

Every source cited in the last 30 days is listed on the sources page, split by whether the pipeline tracks it through a feed or found it through search. Publishers that keep appearing without a configured feed surface there as candidates for one.

What to check before relying on a digest #

A digest is a summary of reporting, not reporting. Two limits are worth keeping in mind:

• No human reviews a digest before it publishes. Errors of emphasis, attribution, and detail are possible, and a summary can be confidently wrong about something its sources were vague on.
• A summary is not a source. Every item links out to what it was drawn from. For anything you intend to act on, quote, or repeat, follow the link and read the original.

Following along #

The site-wide feed carries every cadence. Narrower feeds come free: append feed.xml to any cadence or tag page and you get a feed of just that slice — only the weeklies, say, or only what is tagged for one company.