Though its founders are still in their 20s, GPTZero is a profitable AI detection startup with millions in the bank and a new $10 million Series A round.

Among the young AI startups being relentlessly pursued by venture capitalists these days, GPTZero is already profitable, generating millions of dollars in revenue, a year and a half after its founding. Founded by high school friends Edward Tian, ​​24, and Alex Cui, 26, GPTZero offers detection tools that help identify whether content is generated by AI.

The founders opted for a $10 million “preemptive” Series A led by Footwork co-founder Nikhil Basu Trivedi, the team told TechCrunch exclusively. (“Preemptive” is VC slang for investors landing a deal before founders try to raise money.)

That’s quite a success for Basu Trivedi. GPTZero has been attracting attention from top venture capital firms since Tian released the first version as a web app in December 2022, when 30,000 people instantly flocked to it, crashing the Streamlit-hosted website. (Tian notes that Adrien Trouil, a co-founder of Streamlit that sold to Snowflake for $800 million, later became an angel investor.) The company officially launched in January 2023.

As 2024 began, the client base expanded, with the young founders receiving four to five calls a week from venture capitalists, they said.

GPTZero’s founders told TechCrunch that they’ve seen a 500% increase in ARR over the past six months, and their user base has grown from 1 million to 4 million over the past 12 months, making GPTZero one of the fastest-growing consumer apps this year. By some criteria.

He added that the company has been profitable over the past few months and now has more money in the bank than it has raised in total to date — more than $13 million, including the $3.5 million seed round and the new $10 million.

And growth continues: User count and revenue have “more than doubled, maybe even tripled, since January,” Basu Trivedi said. He wouldn’t comment on valuation, but based on a typical 20% Series A round, the deal values ​​the company at about $50 million pre-money. Other investors in the round include education-focused (and female-led) Reach Capital, Jack Altman’s Alt Capital, Uncork Capital (Jeff Clavier’s fund), and Neo (Ari Partovi’s fund).

How the VC won the deal

Princeton University graduate Basu Trivedi won the lead on this deal by playing the long game. He met Tian at an annual event where a small group of Princeton University students visit companies in Silicon Valley in 2022, before the GPTZero craze began. Basu Trivedi usually takes the group on a hike to the Stanford Dishes.

Tian developed GPTZero while studying computer science, natural language processing and journalism at an Ivy League university, and during internships at the BBC and then The New York Times, where he wrote code to help journalists identify AI-generated content.

After the initial web app garnered strong response, Tian enlisted the help of his friend Cui, who had a master’s degree in machine learning from the University of Toronto and dropped out of his PhD program to become a co-founder.

The duo rewrote the app into its current standalone platform, gained about 1.5 million users in the first five months, and then raised $3.5 million in seed funding, mostly from angel investors including former Reuters CEO Tom Grosser, Carnegie Mellon University professor and former Apple AI research director Russ Salakhutdinov (who sold his own startup, Perceptual Machines, to Apple in 2016), and CNN CEO and former New York Times CEO Mark Thompson.

Basu Trivedi had heard buzz about GPTZero in VC circles, seeing it garnering media attention and attracting impressive angel investors. As a seed investor who has invested in companies like Canva, ClassDojo, and Frame.io, he had an eye for spotting noteworthy companies.

He texted Tian in January 2023 to ask how things were going. He attracted founders with his network and product know-how from fast-growing companies like Canva, and the background of his own fund co-founder Mike Smith, the former chief operating officer of Stitch Fix and Walmart.

The founders, both in their 20s, were “desperate for investors” with both product and operational experience, Tian said, “especially since Alex and I are still learning how to build a big company.”

To prove this point, shortly after closing the round, Footwork hosted a networking event with AI leaders, including Basu Trivedi’s college classmate Jack Altman (brother of OpenAI’s Sam Altman, who participated in the Seed A round), and Nvidia founder and CEO Jensen Huang.

“The benefits of big data“

GPTZero is not the only company working to identify AI-generated content: others include AI Writing Check, Copyleaks, GPT Radar, CatchGPT, and Originality.ai.

However, much of the AI ​​detection industry has very low accuracy. The researchersOpenAI, which launched its own AI detector in early 2023 under pressure from skeptics in the AI ​​industry, shut down the tool in July, about seven months later, after it was widely criticized for its poor performance.

Interestingly, when TechCrunch’s Kyle Wiggers ran his own experiments with these tools, they all failed except for GPTZero.

Naturally, GPTZero has its own benchmarks, particularly through its partnership with researchers at Penn State, to help back up its claims that its technology works well, regardless of general industry reputation.

Choi says GPTZero is more accurate because it has access to more data and builds its own LLM models using cutting-edge open-source tools (which he declined to disclose details).

“We have an advantage in big data. We have millions of human and AI text examples,” Cui said. “And we combine that with best-in-class models and deep learning. We actually use language models to detect language.”

The startup may be best known for helping teachers detect AI-generated student assignments (in October, GPTZero has reached an agreement The company’s customer base has expanded since its partnership with the American Federation of Teachers, and now includes government procurement agencies, grant-making bodies, hiring managers, and, most interestingly, AI training data labelers.

Ultimately, using AI-generated data for AI training “will cause the model to collapse,” Tian said, because teaching a model using fabricated examples is not the best way to make it work in the real world.

Naturally, the young founders have a grander long-term vision: They want to create a new, independent layer of an accountable internet that ensures human and AI content is properly attributed.

To that end, the team is currently working on AI hallucination detection. Hallucinations, AI-generated fiction presented as fact, are a thorn in the GenAI industry’s side. Their first step in addressing this issue is to offer a new, free AI text copyright check for the LLM training dataset, which can generate training data for hallucination detection more broadly.

“We’re just trying to avoid a world where the entire internet is AI-generated content,” Tian said. “An internet where everyone is using AI would eliminate the opportunity for people to continue posting creative and original content.”