Inside GPTZero's Office Hours with Founder Edward Tian

The culminating session of GPTZero’s Teaching Responsibly with AI certification series featured a special Office Hours led by Edward Tian, the company’s founder. The discussion explored the origins of GPTZero, the intersection of media literacy and AI, and included live demonstrations of the platform’s newest tools.

Tian traced GPTZero’s roots back to his time as a student at Princeton, where he pursued dual studies in journalism and computer science. A defining moment came during a writing course taught by the acclaimed author John McPhee.

“He’s taught the same class for over 45 years,” Tian recalled. “He deeply values the creative journey of writing—the editing, the word choice, the craft. I was fortunate enough to be part of that class.”

A year later, Tian and McPhee reconvened to test the limits of generative AI. Their central question: could artificial intelligence replicate McPhee’s distinctive writing style?

“He was curious about AI’s capabilities and limitations,” Tian said. “We’d prompt ChatGPT with ‘write in the style of John McPhee,’ and it would try… but the result wasn’t convincing.”

The pivotal moment arrived when McPhee posed a deeper question: can AI recognize when writing is exceptional and should remain untouched?

“He asked, ‘Can AI tell me this essay is so strong that it doesn’t need any changes?’” Tian shared. “I realized there wasn’t a tool that could do that.” McPhee responded, “Exactly—that’s the role of a great human editor.”

That exchange became a foundational insight for Tian. “What we aim to build isn’t just about generating or rewriting text,” he said. “It’s about protecting what’s truly original, meaningful, and human in writing. Our goal is to celebrate quality, to spotlight authenticity—not just flag AI use.”

Closing the gap between educators and learners

Tian announced a new initiative: for the first time, all teachers participating in the webinar series would receive a full year of free access to GPTZero—on the condition that they include the tool in their course syllabi.

He noted that many educators begin with a strict stance: banning AI entirely to safeguard authentic student writing. But as students increasingly engage with AI, the dialogue shifts. Teachers are now asking: how much is too much? What does ethical AI integration look like?

Tian highlighted the journey of Eddie del Val in Oregon, who initially adopted GPTZero solely to block AI-generated work. Over time, his approach transformed. He began integrating the tool into his teaching practice, even requiring every student to use it.

He introduced the 80/20 rule: 80% of writing must be original student work, with AI supporting no more than 20%. For Tian, this marked a positive shift—teachers gaining confidence to define their own AI policies based on their students’ needs and classroom culture.

AI Literacy as a Form of Media Literacy

Tian emphasized that the issue of plagiarism extends beyond student behavior—it also applies to AI systems themselves. Large language models are often trained on vast amounts of content scraped from news sites and journalistic platforms, frequently without permission or credit. These models, sometimes called “stochastic parrots,” replicate patterns from uncredited sources.

He cited the New York Times lawsuit against OpenAI, filed roughly eight months earlier, as evidence of growing friction between media organizations and AI developers. Despite media coverage, Tian pointed out, there’s still no clear resolution on attribution or consent.

This raises a critical media literacy challenge for educators: how should AI be taught when the tools themselves raise ethical concerns?

Tian also addressed AI hallucinations—instances where AI produces false information with high confidence. He referenced a joint investigation by GPTZero and The New York Times into AI-generated books on Amazon. The findings were alarming: entire guidebooks filled with fabricated content.

One example was a travel guide to Uganda that listed attractions and hotels that didn’t exist. Only after purchase did a buyer discover the deception. Following the investigation, Amazon executives contacted GPTZero to discuss solutions and prevent further damage to the platform’s credibility.

To combat misinformation, GPTZero has partnered with Wikipedia, providing editors with access to its detection technology. “Wikipedia is a rare example of a human-curated, community-built knowledge base,” Tian explained. “Yet it was heavily used to train early AI models like ChatGPT—without consent.”

New Features from GPTZero

Tian unveiled several key product updates. One of the most anticipated is the Lightly Edited by AI tag, developed in response to teacher feedback. It distinguishes between fully AI-generated content and student writing that has been lightly polished using grammar tools like Grammarly.

“This came directly from educators,” Tian said. “They see a big difference between AI writing an essay from scratch versus a student using a tool for minor editing.”

Another major update is a reimagined writing feedback system, structured around three core areas:

1. Content – assessing the overall structure and depth of thought.
2. Clarity – improving how ideas are communicated.
3. Grammar – refining language mechanics last.

“This reflects what teachers are telling us,” Tian explained. “In the past, poor grammar might have cost students dearly. Now, educators care more about ideas and reasoning than perfect syntax.”

GPTZero has also launched a bibliography and sources checker, addressing a widespread concern from educators last semester. “Teachers keep finding AI-generated references—fake articles, broken links, or real links that have nothing to do with the essay,” Tian said. “During grading season, no one has time to verify each one.”

The new tool automatically scans bibliographies, identifying hallucinated sources and checking whether cited content actually aligns with the essay. “It’s not enough that a link works,” Tian noted. “If the article is about self-driving cars but the paper is on aviation, that’s a mismatch.”

Educators can try the tool at gptzero.me. “Feel free to test it,” Tian invited. “Go to gptzero.me, click on the ‘check bibliography’ option, try one of our sample essays, or submit your own. See how it works firsthand.”

Another standout feature is Authorship Replay, which allows both teachers and students to visualize the writing process. “You can now see how a document evolved—who wrote what, when, and how it changed over time. It’s valuable for teachers, but also empowering for students.”

Wrapping up, Tian reiterated GPTZero’s mission. “We’re not building detectors just to catch AI use. We have a research team dedicated to understanding how this technology evolves.”

He highlighted one of GPTZero’s research advisors, who collaborates with their machine learning team and also leads AI model training for Meta’s LLaMA project. “Having a lab focused on real academic research in this space—that’s what sets us apart.”

The above is the detailed content of Inside GPTZero's Office Hours with Founder Edward Tian. For more information, please follow other related articles on the PHP Chinese website!