Videos are everywhere and the offer is always the same. Let the AI do your homework and you won’t get caught.
According to a New York Times investigation, TikTok and YouTube are now full of classes that sell two types of tools to students. Humanists rewrite AI-generated text so it no longer reads like a chatbot. Automaterials do a trickier job: they drip words into a document for hours, making fake entries, deletions, and edits so that the finished essay looks like a real writing session.
Both are designed to defeat the software teachers use to catch the AI.
The same companies sell the disease and the drug
That’s the worrying part. Some companies that sell detection tools also sell tools that beat them.
Grammarly, now owned by Superhuman, offers teachers an “authorship” checker that scans a document’s history for signs of AI. The same program will also create text from scratch, “humanize” it, and rewrite phrases that can trigger the detector. A detector born as a Princeton thesis, GPTZero can write an entire article complete with citations in seconds.
The NYT found that a marketer set up a fake teacher’s assistant on TikTok to pitch to students.
Superhuman understudy Jenny Maxwell has been open about where this is leading. The race between detection and evasion, he said, “is ultimately a dead end.” His summary: “Bigger cat, bigger mouse.”
The detectors are almost working
He has a point, because cats aren’t very good.
University of Florida researchers tested five of the most popular AI text detectors and found false-negative rates as high as 99.6 percent, with a single word being enough to beat most of them, Digital Trends reports. The tools also disproportionately flag non-native English speakers and produce false positives.
Therefore, schools that discipline students with detectors are standing on very thin ice. The technology they rely on is, by the manufacturers’ own admission, losing.
From oral exams to internet outages
Faced with this, institutions improvise, and responses range from sensitive to extreme. On the quiet end, Harvard professors rely more on oral and paper-and-pencil exams, which a chatbot can’t do for you.
On the other hand sits compulsion.
India has ordered Telegram to be blocked for several days to stop cheating in a national medical school entrance exam, The Register reported, after the test was canceled and rescheduled following a suspected leak. More than two million people appear for this exam for about 100,000 seats.
Digital rights groups have called the shutdown disproportionate and part of a broader pattern by governments combating the misuse of artificial intelligence with very blunt instruments.
The number was always a problem
Step back and the delusional panic seems like a sign of something old. School long ago turned learning into a single number, a grade.
Philosopher C. Thi Nguyen calls this “value capture”: you accept an external metric, then quietly let it replace what it’s meant to measure. In his book “The Score,” reviewed this week by the MIT Technology Review, he cites GPA as a classic case in point. Students stop understanding and start chasing the grade. It’s Goodhart’s law in the backpack: when a measure becomes a target, it ceases to be a good measure.
Artificial intelligence is the most efficient optimizer yet invented for this goal. If the purpose of the essay is not to think, but to score, then offloading thinking is a rational move, even though research warns that offloading such a cognitive load allows for the exhaustion of real skills.
No gas pedal, no brakes
The people who built this technology are also worried.
Anthropic co-founder Jack Clarke told the BBC that the industry “has a gas pedal but no brake pedal,” noting that Anthropic’s own model now writes most of its code. He has company coordinated braking called AI at the border. Maxwell, on the other hand, argues that denying students AI is an “educational mistake” because they will use it at work regardless.
Both things can be true.
The detection arms race is unwinnable, and detection was never a real issue. What’s more difficult, which schools have avoided for a century, is what the price is actually for. AI didn’t create this problem. It just did cannot be ignored. Until someone answers it, the bigger cat will continue to chase the bigger mouse and the mouse will win.
