Pangram is a company that provides an advanced AI detection model to identify AI-generated text, aiming to give transparency into the authorship of the text, to see if it was written by a human or AI, or if it was of mixed authorship. Pulse 2.0 interviewed Pangram co-founder and CEO Max Spero to gain a deeper understanding of the company and how Pangram is different from existing AI-text detection technology
Max Spero is the Co-Founder and CEO of Pangram, an AI detector that detects AI in text by analyzing and understanding subtle cues in the writing.
Formation Of The Company
How did the idea for the company come together? Spero shared:
“My cofounder, Bradly Emi, was a classmate at Stanford. We were both studying in the AI/computer science fields and working on similar things. One relevant project was working on making robots better by collecting data from their operations and then using that data to improve robot performance. This feedback loop idea stuck with us and in part influenced our philosophy with training the algorithm powering Pangram.”
“After graduating from Stanford, we then went to work at Google and Tesla just a few years before generative AI became available to everybody through ChatGPT. By the time GPT-4 came out, we realized that the wide availability of generative AI was exciting and it had the potential to do good and make things easier for people, but it also had the potential to do harm, especially to do harm at scale. What was needed was a reliably accurate way to differentiate human from AI writing in an unlimited number of use cases: education, content moderation, publishing, authoring, journalism, law, and so much more.”
“We created Pangram to address that problem. We realized that our approach to smart detection and discovery solutions was new and potentially quite potent. After a few conversations and some research about where our time and technology could have the most impact and success – we decided there was a need for better systems for teachers and schools to help protect academic inquiry and integrity.”
“Pangram is still widely used in several other applications as well, especially in content moderation.”
(L to R) Max Spero and Bradley Emi co-founded Pangram, an AI detector with a near-zero false positive rate.
Favorite Memory
What has been your favorite memory working for the company so far? Spero reflected:
“I enjoy that the company has allowed me to meet so many new and interesting people across all kinds of career fields. In building Pangram, we’ve had the chance to meet educators, corporate leaders, journalists, security experts, as well as people who are curious about what we’re doing and the ways that we have trained and refined our model to make Pangram the most accurate detector on the market.”
“I also really enjoy when we get a chance to provide Pangram to academic researchers either so they can test our accuracy or use Pangram for other kinds of research projects. We’ve been part of projects studying misinformation, AI slop on Amazon, and product reviews, among many others. One of the most interesting is this very recent study by the American Association for Cancer Research to analyze abstracts and manuscript submissions for the prevalence of generative AI in abstracts and peer-review comments.”
Core Products
What are the company’s core products and features? Spero explained:
“Pangram is a way for people to see what and where AI-created text may be in any given document. We provide AI transparency for teachers, journalists, content moderators, and others who rely on written work. And we do that better than any other product because our core technology is better.”
“There are several notable independent university studies showing that Pangram is the most accurate, most reliable AI detector in the market. Studies that found Pangram to be the most accurate AI detector include: the University of Maryland, the University of Pennsylvania, joint research from the University of Houston, UC Berkeley, and UC Irvine, and this recent study from The University of Chicago.”
Challenges Faced
Have you faced any challenges in your sector of work recently? Spero acknowledged:
“In addition to the typical operational challenges of starting and growing a business, we have a communications challenge and are working to address the inaccurate and outdated view about our market.”
“Some people have the false impression that AI detection does not work, or that it is too inaccurate to be relied upon. Because of that perception, we find ourselves doing quite a lot of education about detection. That’s OK – we’re in the education business and support educators so we value learning and the learning process. Generally we overcome the entrenched perceptions and misconceptions as soon as someone uses Pangram, or when they see the long list of the well-regarded research documents verifying Pangram’s accuracy and our low FPR and FNR rates.”
Evolution Of The Company’s Technology
How has the company’s technology evolved since launching? Spero noted:
“Our model to detect the presence of AI in writing is evolving constantly because generative AI is constantly changing. It is essential to always be evolving to keep our position as the most accurate AI detector and most reliable provider of AI transparency on the market.”
“At the same time, we’re also continuing to add new features and give more robust information to our partners and customers. Compared to where we were when we started, the depth and quality of information we provide to our partners is the difference between a Ford F150 and Ferrari. And every day, we continue improving.”
Significant Milestones
What have been some of the company’s most significant milestones? Spero cited:
“There was a moment early on when we realized that scaling up our model with more parameters and more data was producing huge gains. Our first models had a 1 in 100 false positive rate (FPR), but by scaling up we were able to reach 1 in 1,000, then 1 in 10,000. We could have settled on a 1 in 100 FPR, but instead we kept working at it to get to 1 in 10,000 false positive rate. We knew upon reaching that milestone of 1 in 10,000, that Pangram outperformed any detector on the market. We also knew that we could extend our lead by relentlessly improving that number.”
Customer Success Stories
Can you share any specific customer success stories? Spero highlighted:
“This is very recent news from the last few months that illustrates a lot about Pangram’s success and our potential. Two companies, HARO and Qwoted, which help companies and other experts reach journalists with story ideas or expertise, started using Pangram. These are the two primary players in this space and landing them means that we essentially have full market share in a bedrock service to the news and information industry.”
“The use case here is critical to understand because it is a glimpse into the global issue of misinformation and its ability to wreck havoc on the world. Now; however, when journalists get a pitch or an answer from an expert on either of those sites, they have greater confidence in their ability to determine if the text was created by AI or not.”
Funding/Revenue
Are you able to discuss funding and/or revenue? Spero revealed:
“While I cannot get into too much detail, we are generating revenue, in fact a 10x increase since the start of this year. And we also announced a Series A funding round of $4 million. We have several large enterprise contracts closed and others in the pipeline.”
“Pangram’s goal is to be the go-to tool not only in education, but in all verticals where trust and authenticity matters.”
Differentiation From The Competition
What differentiates the company from its competition? Spero affirmed:
“Pangram’s technology is trained using a different premise from our competitors and that difference is one reason why our technology is better.”
“Our competitors, Turnitin and GPTZero, are focused on student writing. Turnitin’s whitepaper reports accuracy numbers only on essays and academic writing. On the other hand Pangram evaluates on a much broader range. We find—and our research and third-party research bears this out—that this broad approach in training produces a model that is more accurate on the full scope of AI writing. In other words, Pangram is just as accurate with academic writing as it is for Quora answers. This is because the act of writing is not a siloed, independent effort. A person’s writing reflects their experiences and unique world view and just as in education where breadth of knowledge is vital to developing a well-rounded person, the broader approach to training a detector produces a better equipped detector. The more diverse experiences help it to be more accurate regardless of the specificity of the application.”
“And, getting a bit into the weeds relative to education, the origins of our training set is different. Pangram’s AI detection model is trained entirely from pre-ChatGPT documents that are licensed for training and commercial use, not collected from a limited type of papers, or solely from synthetic mirrors.”
“There are more studies about Pangram coming out because we make Pangram available to researchers to verify its accuracy or to run their own studies. We feel this is important because our mission is not only to produce the best detector, it is to be an active partner in supporting and protecting authenticity, integrity, and transparency.”
Additional Thoughts
Any other topics you would like to discuss? Spero concluded:
“We think it’s important to mention that, although Pangram is focused on the education market and specifically in helping teachers, schools, and students understand and address AI challenges, we also know that what we’re doing is bigger than that.”
“Education is important, essential really. But being able to have good information about the information you are receiving is going to be fundamental to living in the AI age. Being able to see beyond and under the text has implications for reader trust in journalism, marketing, propaganda, deliberate disinformation, fraud, and the very value of information itself. Pangram may be helping teachers accurately assess written work from college students, but we’re also very consciously building the tools of self-preservation and protection that we think everyone, everywhere will need.”
