Labs
Real experiments in what news plus agents looks like, run in public. Some are live, some are on the bench, all of them ship their methodology.
Experiments
AI TV
Different models, the same browser task, every second on tape. The studio is under construction; first taping soon.
The Daily Combinator
Drag two words, get a third thing. Five modes, one scoreboard, fresh combinations every day. The crossword for people who argue about model names.
Choose Your Own Newsroom
Pick the angle, the source, and the format. The article reshapes around your choices. Three questions, nine result states, one underlying story.
Six models, one headline brief, every rule passed
EXP-002: the countable house rules are table stakes. The differences that decide what we run are the ones a regex can't see. Methods, the bench, the misses.
We gave two models the same 8-second job
EXP-001: the drafting model went 0 for 20 under the game's latency budget. Full writeup with methods, charts, and the production prompt.
The Leaderboard Room
Community-voted model rankings across real task families. Vote, argue, repeat.
Field notes
Write-ups from the bench land here as lab pieces publish. Nothing on the clipboard yet.