GPTZero, a Princeton student's app that uses perplexity and burstiness metrics to help detect whether a human or ChatGPT wrote an academic essay, goes viral
“Humans deserve to know when the writing isn't human.” — Deputy Editor, Innovation & Tech … Edward Tian was fast asleep when his bot broke a website.
The Daily BeastTony Ho Tran
Context & Ripple Effects
GPTZero’s viral moment arrives as text generators were already moving beyond novelty use cases: early adopters were using them for business emails and creative work in everyday writing workflows. Its focus on academic essays turns that broader adoption into a provenance problem for institutions deciding when writing can still be treated as evidence of individual work.
Later coverage reinforces both sides of the emerging contest: researchers reported high claimed accuracy for detecting ChatGPT-written academic text, while tools designed to make AI output appear human and evade detectors create pressure on any detector-led enforcement model.
First-order effects
GPTZero and co-founder Edward Tian gain immediate visibility as a provider of a screening tool for educators and other academic evaluators confronting ChatGPT-generated submissions.
Students and academic institutions now have a prominent, accessible mechanism for challenging the authorship of an essay, rather than relying only on stylistic judgment.
Second-order effects
As generative writing spreads into routine work, institutions evaluating written output face a new verification step that can slow or complicate assessment even when AI assistance is permitted.
Detection providers must substantiate reliability claims as researchers publish competing detection approaches and evasion-oriented writing tools target the same academic market.
Third-order effects
If generation and detection continue to advance together, authorship verification is likely to become a standing layer of education and editorial workflows rather than a one-off response to ChatGPT.
The durable constraint is not text production but credible attribution: systems that cannot explain how a determination was reached may struggle to serve as the sole basis for high-stakes academic decisions.
The trend:Generative AI is creating a verification bottleneck in which cheap synthetic writing drives demand for provenance and detection tools, alongside incentives to evade them.
EXCLUSIVE: @edward_the6 built an app that can help detect if an essay was written by a human or an AI like ChatGPT. While he's already getting attention from big name VCs, he just wants to refine the app—not to mention graduate college. https://t.co/sXipGJ5Qw9
A new app — built by @edward_the6 '23 — can detect whether your essay was written by ChatGPT, as researchers aim to combat AI plagiarism. ⬇️ (via @TechInsider) https://www.businessinsider.com/ ...
Interesting...this is a very tough problem, right? Nearly impossible to detect all such cases? “Princeton Student Builds ChatGPT Detection App to Fight AI Plagiarism Edward Tian says he was inspired to create GPTZero by increased AI plagiarism.” https://www.entrepreneur.com/ ...
Okay, found something real. The above method isn't a fair attack, since no teacher would accept emojis, but this one is: 1) Generate a text using ChatGPT 2) Insert a zero-width space before all instances of “e” 3) The text will now pass the GPTZero detector https://twitter.com/..…
Doing some tests generating product descriptions and they seem pretty good. They've all passed the GPTZero thing that tries to detect AI content (who knows how good that is). https://twitter.com/...
GPTZero is surprisingly good! My first attempts were not able to defeat it. However, I found a way to defeat it. The technique does require a human sample to copy from, however. I fed it presumably human written text and instructed it to transform it. Passed. #GPTZero #ChatGPT ht…
GPTZero is a proposed anti-plagiarism tool that claims to be able to detect ChatGPT-generated text. Here's how it did on the first prompt I tried. https://twitter.com/... https://twitter.com/...
I used GPTZero to test if ChatGPT-aided *editing* can be detected by @edward_the6's GPTZero The answer is...no! But it's designed for detecting base text written by ChatGPT, not editing efforts I do recommend using ChatGPT as @Grammarly or ProWritingAid on steroids https://twitte…
Amazing enterprise by Princeton senior building a #chatgpt essay detector. Seen via @TonyHoWasHere story: https://www.thedailybeast.com/ ... https://twitter.com/...