AutoDataBench tests whether agents can write training tasks
A research benchmark evaluates agent-written executable tasks for validity, difficulty, and whether they elicit intended behavior.
Issue 2026-09-10Search the complete card catalogue
Across every issue
Titles, domains, summaries, categories, signals, dates, and editorial-interest scores across the entire published catalogue.
Results
A research benchmark evaluates agent-written executable tasks for validity, difficulty, and whether they elicit intended behavior.
Berea is a desktop Bible study app for exploring Scripture, keeping annotations attached to the words, and presenting passages live.
Breez is a web reader that displays text one word at a time, at a pace the reader controls.
A carrier lookup and email alert service that tracks FMCSA records for insurance, authority, out-of-service, and registration changes.
An interactive Korean fortune-telling forest where visitors find an offering, then consult the site’s divination stations.
A computer-vision system turns chess clubs’ existing cameras into live positions, move lists and PGNs for games played on physical boards.
A searchable directory of Christian TV channels, streaming services, and regular church broadcasts, with ways to watch each.
An evidence-focused site investigates viral claims, conspiracy theories and pseudoscience, with searchable claim pages, topic guides and a library of sources.
A co-curated spiritual formation community for people seeking faith and gathering beyond institutional religion.
A Chinese-language site for browsing and searching Daoist classics, organized by traditional collections and subject themes.
Artist Davey Jose explores spinal cord injury and bodily difference through painting, digital work, and a VR film.
A prelaunch handheld concept paired with nine browser-playable pixel games and a pitch built around one community-picked game each day.
A Spanish-language tool for issuing electronic transport-control documents (DeCA) and collecting signed delivery slips on a recipient’s phone.
A free tracker for several trading card games that connects your collection to deck lists and shopping lists.
Florida Intervault presents a Fort Lauderdale document-storage service for prenups, wills, deeds, premarital records, and other papers that may matter in a divorce.
An iPhone focus timer pairs adjustable Pomodoro sessions with a dog companion, collectible finds, room décor, and optional app blocking.
A custom service that turns a child’s drawing and game ideas into a playable game.
A page posing as India’s government eChallan portal prompts visitors to install an 18 MB verification app on Android.
The page presents itself as an Indian government eChallan portal, then says a vehicle has been flagged for internal review and directs visitors to download an Android app.
A Chinese-language EJU study site combining syllabus cards, explained past questions, Japanese-language practice, review, and mock exams.
A downloadable desktop suite for editing, signing, redacting, OCR, and comparing PDFs on Windows, macOS, and Linux.
A French-language tracker organizes vulnerabilities reported as actively exploited by affected products, drawing on CISA’s catalog and CERT-FR alerts.
A rapper’s playful promo site combines social clips and self-reported stats with a rich-or-broke photo album, a flip-style trading card, and a tip jar.
A browser-based utility that writes dates, locations and captions from Google Takeout’s separate JSON files into photos and videos.