The one AI detector people actually trust
Summary
The Verge Cast interviews Pangram CEO Max Spiro about his company’s highly accurate AI text detector, which uses active learning to achieve a 1 in 10,000 false positive rate. They discuss Pangram’s methodology, its superiority over perplexity-based tools, and its role in verifying authorship for educators and publishers. The episode also covers recent tech news, including OnePlus exiting the US market, EU regulations forcing Google to open Android, and FCC proposals to relax broadcast ownership caps.
Topics discussed
Intro: The rise of reliable AI detectors
90 Seconds: OnePlus exits, EU vs Google, FCC rules
Sponsors: ServiceNow and MongoDB
How Pangram became the trusted AI detector
The 'Synthetic Mirror' training method
Why perplexity-based detectors fail
Confidence scores and partial AI detection
Sponsor: Even Realities smart glasses
Sponsor: MongoDB for AI data layers
Case Study: The AI plagiarism prize controversy
Who uses Pangram and how to handle positives
The future of human writing and trust
Adversarial arms race with AI humanizers
Conclusion: AI text as cognitive offloading
Listen ad-free on Castria