Edward Tian (@edward_the6) header
Edward Tian (@edward_the6) profile picture

Co-founder @ GPTZero. Writer and em-dash lover.

Joined December 2015
Photo by Edward Tian (@edward_the6)Photo by Edward Tian (@edward_the6)Photo by Edward Tian (@edward_the6)Photo by Edward Tian (@edward_the6)Photo by Edward Tian (@edward_the6)Photo by Edward Tian (@edward_the6)Photo by Edward Tian (@edward_the6)Photo by Edward Tian (@edward_the6)Photo by Edward Tian (@edward_the6)Photo by Edward Tian (@edward_the6)
Today, we're launching AI Vision. The first AI slop detector that exposes content as you scroll.
34
37
28
178
406,910
gptzero, named for the false positive rate
It’s unfortunate that mostly black authors have been outed for using AI to write fiction, but does that mean AI detectors are racist ? Sadly, the answer is a resounding NO. With the help of @pangram and @GPTZeroAI I analyzed 187 African books published between 2001 & 2026. 👇👇
1
2
225
New research from Tuhin on @GPTZeroAI and @pangram bias in languages, one of the most comprehensive ones on this topic.
It’s unfortunate that mostly black authors have been outed for using AI to write fiction, but does that mean AI detectors are racist ? Sadly, the answer is a resounding NO. With the help of @pangram and @GPTZeroAI I analyzed 187 African books published between 2001 & 2026. 👇👇
2
4
9
864
I regret to inform you that we have found our first 300+ word false positive for GPTZero. It's a 309-word text with the same phrases, 'That through the intersession' and 'Thee, hear us' repeated again and again.
@edward_the6 @alexcdot After a lot of attempts and tricks, found a 2014 text of 309 words where GPTZero 5.0b returned 100% AI in two scans. I didnt know AI text detectors had gotten so good - i thought it would be much easier to solve it! The text that worked is from github.com/GITenberg/Mary-He… from “hope, we” [line 5897] through “the promise of” [line 5933]
2
12
2,276
Still waiting on anyone to claim our open challenge from @Nition for anyone to find a 750+ word false positive.
2
195
Started reading the lecture notes for advanced macro and had the horrifying realization that what I was reading might be AI-generated. My itch was confirmed. They're gonna have to put me in an asylum.
5
3
3
78
25,100
Love this prize idea. We'll match + double Nition's prize for @GPTZeroAI to $1000, and offer a professional plan and API refunds for anyone who wants to dig in.
This quoted post is unavailable.
4
1
3
15
3,126
Today, I'm excited to announce we're launching AI Vision 2.0! AI Vision is the first slop detector that exposes content as you scroll. Our 2.0 launch adds the ability to customize it completely. - Personalizable content tags powered by Jev models - Blocking content in your feed based on AI detection scores or custom rules. With the launch, we're also giving anyone who comments 'AI Vision' on this post 30 days of Premium GPTZero. Try it out on LinkedIn, Reddit, X, and more!
13
17
4
77
5,372
Edward Tian retweeted
You might think Jensen's apology letter was written by AI. You're wrong. If you think about it, Jensen created the GPUs that ran the LLMs that wrote the letter So technically, he wrote it himself, just with a few extra steps!
Here's what Nvidia's Jensen Huang sent to staff, admitting to affair: "To my colleagues at NVIDIA, I owe my family and all of you an apology. I was unfaithful to my wife. My actions hurt her and our children. There is no excuse for what I did. I am deeply sorry. Throughout my life, my family has given me love and support. They deserve my honesty and loyalty. Instead, I let them down and damaged the life we built together. I take full responsibility for what I did and the hurt I caused, and I accept the criticism that comes with it. To my wife and family: I love you. I know an apology cannot undo what I have done. I hope our family will heal and that, over time, I can earn back your trust. To my colleagues at NVIDIA, including those I have worked alongside for decades, and to our board, partners, and shareholders around the world: you deserve better from me. Leading NVIDIA is the privilege of my life. My responsibility is to serve you and be worthy of your trust. I have fallen short of that responsibility. I am fully committed to NVIDIA and will work to earn back your confidence. Regarding the national security concerns raised in news reports, NVIDIA’s Board of Directors has initiated an independent investigation. I support the investigation and will cooperate fully. I love our country and our company. I am committed to protecting NVIDIA and supporting America’s national security. I am proud and grateful to be an American. My family bears no responsibility for my choices. I do. Please give them privacy. Jensen"
3
24
1,246
Edward Tian retweeted
Friday was my last day at Anthropic. Two years ago, when Claude Code was a small internal tool, I asked it to build an RL env. I steered it for tens of hours but never manually wrote a line of code. It worked, I was hooked, and I joined the team. When we first launched, most people approved every permission prompt by hand and mostly used Claude to answer questions and write tests.  Today, the default is trusting Claude's judgment to auto-approve. Claude does large migrations and builds new products.  Helping build Claude Code into an industry-defining product has been the journey of a lifetime.  Thank you to the incredible team behind Claude Code, Cowork, and Tag. These products are the sum of thousands of great ideas from you. I'll miss working with you all. Thank you to our users for embracing a new way of working. Your constant feedback shaped so much of what we shipped and your stories made the work so meaningful. Excited to see what you all make next!
1,224
532
565
16,672
4,721,099
This is one of the reasons we decided to completely re-engineer our model @GPTZeroAI this year, changing from two classes (AI or Human) to three output classes (AI, Human, and Mixed). What should the output of something human-written + 60 words from AI even be? Think about it hard, and you'll realize all good detectors should move to three output classes.
This is more troubling than the fact that someone apparently found a pangram false positive (it was bound to happen). Adding ~60 words should not take a block of text from predicted 100% human to 100% AI.
4
12
1,237
Edward Tian retweeted
wow💔 @OpenAI
954
1,902
859
60,928
22,292,515
The provost of Dartmouth was ousted from the university after it surfaced he wrote op-eds completely with AI. It seems like a ridiculous overreaction and witch hunt on an issue where editorial norms are still forming. But it is ironic that the main one, in WaPo (screenshot from GPTZero), expressed concern that AI undermined the meaning of grades in college. gptzero.me/share/s/432338a1-…
4
1
28
5,419
Ran this same text through GPTZero, and it didn’t have the false positive Pangram gave. That being said, I'm against the shaming of individual writers happening on the internet. AI detectors are statistical signals, not tools for ending careers. gptzero.me/share/s/27160a91-…
So it happened. This is the first time I've given Pangram a text entirely written by me, about my own life, and it said it was AI-written (58% of it!). I'm sorry, @pangram, but this false positive should be enough to stop this from ruining people's literary careers on probability
5
1
1
30
3,703
New @business article linking GPTZero analysis on nearly 1 in 5 UN speeches are AI-generated
1
4
417
Time to switch off Notion.
I've never seen something go so viral internally Giving teams the power to build their own bespoke collaborative tools led to a level of creativity that was honestly unhinged and super fun to watch My fav examples in the thread (reply with your own!)… teapawt.net/Superhuman/status/2112…
1
12
856
Edward Tian retweeted
We’ve been dogfooding AI views, a new feature in Superhuman Docs, for the past few months. The whole team is a little bit obsessed, me included! I plan my trips on a one-year calendar and have never found one formatted the way I want. I wanted it aligned on week boundaries. Make it thin, and it’s 52 rows of one week. A little wider, and it’s 26 rows of two, then 13 rows of four, etc. The columns stay Monday, Tuesday, Wednesday, so I can shade in the weekdays. So I described it to AI views. Now it’s in a doc right next to my travel notes, and when I add a trip to the table with all my travel details, it shows up on the calendar. There are truly infinite use cases, and no code is needed. If it’s off, I just tell it what’s wrong and get a fix. If you build something you love, send it my way
8
8
1
63
6,225
Replying to @seconds_0
Plz use one of the accurate ones before trying to undermine a conversation this important. These rando ai writing tests fail everything. Test it on ur own writing. Pope passes on gptzero. Don't be a dick.
42
7
9
348
49,853
A recent paper benchmarked GPTZero's Hallucination Check against a synthetic benchmark and reported a large performance gap. We reran the benchmark, and most of that gap comes down to one question: what counts as a hallucinated citation? gptzero.me/news/what-counts-…
3
1
1
6
488
Edward Tian retweeted
Love seeing more people benchmark citation checkers, this space needs it. But "wrong page number" and "paper doesn't exist" shouldn't get the same verdict when careers are on the line! Great analysis from @dmytronikolai
A recent paper benchmarked GPTZero's Hallucination Check against a synthetic benchmark and reported a large performance gap. We reran the benchmark, and most of that gap comes down to one question: what counts as a hallucinated citation? gptzero.me/news/what-counts-…
2
1
3
428
Great analysis from @semrush with the GPTZero API
After analyzing 42,000 blog posts with an AI detector, content classified as fully human-written outperformed AI-generated or mixed content across the top 10 positions. The gap is most striking at position 1: 80.5% probability of being human-written vs just 10% AI-generated. This may appear to conflict with survey data showing that 72% of SEOs believe AI content ranks as well as human-written content. However, the data points to a more nuanced reality. The key difference emerges at the very top of the rankings. From position 5 onward, the performance gap between human-written and AI-generated content narrows significantly. In other words, AI content performs competitively when the benchmark is ranking on page one. However, in the highest-ranking positions, human-written content clearly leads. Full study: social.semrush.com/3VlVBwJ.
1
9
871