How this works

Every 30 minutes this page reads new posts from the 57 accounts listed below and runs five steps.

1. Group. Posts that share a link, quote the same post or sit in one reply thread are grouped together. Replies to and quotes of posts that are already part of a story join that story directly, with no model involved.
2. Sort. A language model (z-ai/glm-5.3-flash) reads each remaining group next to the list of tracked stories and decides whether it continues one of them, starts a new one, or is noise (personal posts, jokes, evergreen debates, job ads).
3. Rank. A story's heat counts people, not posts. Each person counts once, fading with a 48-hour half-life after their latest post about it, and stories get a bonus when several communities (researchers, policy people, press, lab staff) are talking. A story stays on the front page until its heat falls below the bar; the order follows the last few hours (8-hour half-life), so what's most active now is on top.
4. Write. When a story gains enough new people or posts, or a lab or reporter posts in it, its headline and description are rewritten from the posts and the most-shared linked pages, folding in anything important that happened since.
5. Publish. The page is rebuilt as static HTML.

Reading the page

• Front page: heat of at least 4 and at least 5 different people.
• Early signals: under 4 hours old with at least 2 people. Many fade; a few become the day's story.
• Earlier: everything that has been on this page, by the day it started. Click the heading to fold it away; the page remembers.
• Show: pick a kind of story (release, policy, research...) to see only those; the page remembers that too.
• Rising: at least three new people in the last hour and heat up by 30%.
• New / Updated: changes since you last had the front page open in this browser.