Pangram Secures $9M to Detect AI-Generated Content Amid Online Surge
Image Credits:Pangram
Pangram: Battling AI-Generated Content
New York-based startup Pangram is dedicated to tackling the overwhelming influx of AI-generated content flooding the internet. Recently, Pangram secured $9 million in funding, underscoring the growing demand for tools that effectively differentiate human-created text from AI-generated material.
Recent Funding and Product Launches
The funding round was led by Menlo Ventures, with contributions from Haystack, ScOp, Script Capital, and Cadenza. This investment comes as Pangram introduces its cutting-edge AI text detection model, Pangram 4, alongside an AI image detection tool known as Pangram Image. Pangram claims its new text detection system boasts over 99% accuracy in detecting AI-assisted writing and mixed human-AI content, while also being adept at identifying AI-humanizer programs. The image detection model is currently in a research preview, with plans for a broader release in the near future.
The Origins of Pangram
Founded by Stanford graduates Max Spero and Bradley Emi about two years ago, Pangram emerged in response to the chaos triggered by ChatGPT. This particularly opened the gates to a torrent of bots and poorly crafted AI-generated SEO content. Spero highlights the dangers of AI-generated misinformation, drawing attention to issues like fabricated Russian disinformation campaigns on social media.
The Value of AI Detection
“I think it’s incredibly valuable to know whether what you’re looking at is AI-generated or not,” Spero shared in a recent TechCrunch interview. Understanding the origins of text can significantly alter a reader’s approach—whether it requires skepticism or can be trusted as well-researched journalism.
Pangram’s AI detection model is built on extensive training using millions of known human documents. This allowed the development of a “synthetic mirror” for each document, simulating topics, lengths, and tones using advanced large language models (LLMs).
How Pangram Detects AI Content
The model’s success lies in its ability to identify specific stylistic choices that AI makes consistently, enabling it to distinguish AI-generated content with high accuracy. Importantly, Pangram’s detection does not rely on metadata or hidden watermarks, making it a robust tool for individuals and organizations.
For Pangram, detecting AI content involves more than simply determining if a piece is entirely AI-generated; it also requires assessing levels of AI assistance. For instance, a person might write a draft but use AI to edit or refine it. Spero believes these nuances are crucial, emphasizing that transparency about AI assistance is vital.
Institutional Backlash Against AI Content
As AI-generated content proliferates, backlash is beginning to surface in institutional rules. The open-access archive arXiv enacted a new policy this year that bans submissions if authors fail to review LLM output adequately, thus signaling a demand for vigilance and accountability in content creation.
Competition in AI Detection
Pangram is not alone in this market; competitors like Winston AI, Originality.ai, Copyleaks, and GPTZero are also developing their own detection tools. While none of these technologies are flawless, they collectively contribute to the push against the pervasive acceptance of AI-generated content in various fields, including legal and academic settings.
User Access and Features
Pangram is accessible through a $20-per-month subscription model and provides a Chrome extension that labels posts in real-time on platforms like X, LinkedIn, Substack, Reddit, and Medium. This extension offers users a “feed health score,” detailing the percentage of human versus AI content they encounter. Furthermore, Pangram’s technology is available via API to several clients, including Substack, Quora, universities, and publishers.
Performance Testing
In testing Pangram’s capabilities, Spero indicated that the model incorrectly labels about one in 10,000 human documents as AI-generated. My testing revealed that Pangram effectively flagged entirely AI-generated news articles from ChatGPT and Claude. However, it sometimes misidentified completely rewritten sentences as AI-written content.
When I submitted a personal article that had been polished by AI, Pangram assigned it a 13% AI-assisted score—mostly accurate, though it occasionally flagged human-written sentences as AI-generated. However, when I submitted the entire article in its original form, it scored 100% human.
Image Detection Capabilities
Pangram’s new image detection model also showed promise. The model is designed to identify AI-generated images across different platforms, unlike watermark checks from other companies that focus on their outputs. It uses pixel-level analysis to discern subtle differences between real and AI-generated images. In my tests, it successfully detected AI imagery in various styles, though it misclassified one real photo as human content when it featured an AI creation.
A Balanced Approach to AI Content
Spero emphasizes that while the technology is potent, it aims not to instigate a witch hunt against creators using AI but to establish a mechanism for maintaining quality amid the chaos of AI content proliferation. He warns that without active efforts to prioritize human-generated content, the online landscape could become increasingly dominated by AI, obscuring genuine human voices.
Conclusion
As Pangram continues to roll out its tools and expand its reach, it stands at the forefront of a significant movement aimed at preserving the integrity of content online. The demand for transparency in AI use points to a broader societal need for accountability as we navigate the complexities of technology and information in the digital age.
Thanks for reading. Please let us know your thoughts and ideas in the comment section down below.
Source link
#content #floods #internet #Pangram #raises #detect
About The Author
