OM DENNE EPISODE
Patrick McKenzie (patio11) is joined by Max Spero, co-founder and CEO of Pangram Labs, which builds models that identify AI-generated text and images. Max goes under the hood to explain how AI detection actually works. They also discuss the institutions that quietly assumed writing was expensive: credit bureaus that negotiated the right to shred templated dispute letters, grant committees where reading the application is five percent of the work, and unemployment offices staffed for a world where claims cannot plausibly rise two hundredfold.
–
Full transcript available here: https://www.complexsystemspodcast.com/how-ai-detection-actually-works-with-pangrams-max-spero/
–
Presenting Sponsors: Mercury & Granola
Complex Systems is presented by Mercury—radically better banking for founders. Mercury Spend hands your team and agents their own cards with limits you set once, so nobody waits on you to approve a SaaS invoice and nobody chases a receipt. Apply online in minutes at https://mercury.com/.
If meetings consistently leave you with hazy action items and lost context, Granola handles the transcription so you can actually participate and gives you searchable notes afterward. Try it free at granola.ai/complexsystems with code COMPLEXSYSTEMS
–
Links:
- Pangram Labs: https://textguard.ai/
–
Timestamps:
(00:00) Intro
(00:27) Why AI detection exists
(01:42) Friction as a rate limiter
(05:42) What Pangram actually does
(06:57) Authorship identification at scale
(11:01) Common Crawl and the crawling arms race
(15:52) Being in the weights
(18:08) Ghostwriting letters to banks
(19:45) Templated letters and the FCRA loophole
(22:38) The anti-memetic property of LLM text
(25:28) Value above the weights
(27:03) What happens to programmers
(30:19) Sponsors: Mercury | Granola
(33:39) Jevons Paradox and the compiler
(36:44) What comes after the coding harness
(40:19) Mode collapse and the dinner recommendation problem
(43:00) The patio11 fan blog
(45:58) Cognitive surplus and the dead internet
(49:40) Is this fight winnable?
(54:35) Reputation and disclosure norms
(58:02) Training on books and moral standing
(01:01:52) Attention, phones, and what we've gained
(01:04:17) Wrap
