Advice from writers who actually wore ai slop — what it looks like, how it fits, and how it holds up day to day.
8 articles
Feeds
A Sora mammoth clip shows the weak point in AI labeling: platforms disclose what they know on their own pages, then let downloads and screen recordings leave that context behind.
Cass Ito · August 11, 2026 · 7 min
Feeds
A local name can disguise a remote content operation built from police logs, weather alerts and recycled outrage. The useful clues sit in Page transparency, the website footer and the ad stack.
Reuben Adeyemi · 8 min read
Power
Districts buy Turnitin and similar tools as ordinary software, then let probabilistic scores migrate into misconduct cases. The contract rarely says how much evidence a percentage should become.
Kurt Halloran · 7 min read
Power
AI-writing detectors sell institutions a quick signal, then disclaim the certainty their interfaces imply. Students are left defending their authorship against a probability with no witness.
Kurt Halloran · 7 min read
Power
New Jersey institutions often call AI detection a clue, not proof. Their misconduct machinery can erase that distinction before a student gets a meaningful chance to contest it.
Kurt Halloran · 8 min read
Feeds
A platform warning is easy to crop away. The useful evidence lives in the file’s history, earlier copies, visual structure and the account that put it into circulation.
Priya Nandakumar · 8 min read
Feeds
Fake storm maps and disaster images copy the graphics people already trust. Platforms reward the recognition hit long before anyone checks who issued the forecast.
Cass Ito · 8 min read
Feeds
A rusted iron belt becomes proof of medieval cruelty after faceless videos copy the same weak claim. Search then mistakes repetition for corroboration.
Priya Nandakumar · 8 min read
Other impacts
One update a day
One story each morning — no hype, no filler, no algorithm deciding for you.