Post

Will Oremus
@willoremus.com

People have real qualms with AI detectors, and so do I, but the evidence is mounting that Pangram in particular very rarely gets it wrong (especially on longer chunks of text, but increasingly even on short ones). @brianx.bsky.social tested it and it didn't miss: www.nytimes.com/2026/08/13/t...

nytimes.comI Tested a Popular A.I. Slop Detector. It Felt Empowering.Pangram excels at distinguishing chatbot-generated words from human writing. But it’s not reliable for spotting artificial images.
Eryk Salvaggio
@eryk.bsky.social

I wrote about this as a critic of slop & detectors. To your good Q’s: even rare false positives scale up (1 in 100 is a lot in a million); second, it’s an arms race: LLMs will learn what not to do, breaking any detection infrastructure we opt to depend on. mail.cyberneticforests.com/its-not-just...

mail.cyberneticforests.comIt's Not Just X. It's Y.Against the Quantification of Integrity When the measure of language becomes its target, it ceases to be good language. 💡Nerd Rating: 1/5. I discuss the origins of certain linguistic tics in LLMs a...
Eryk Salvaggio
@eryk.bsky.social

But top of my list is the effect detection has on the way we naturally write; which is secondary to detection, but that it turns us away from writing that “sounds like an LLM,” even though LLMs are optimized via RLVR on reasoning language and so produce efficient, *logical sounding* outputs.

Will Oremus
@willoremus.com

That's a fair worry. FWIW like a lot of journalists I use em-dashes, negative parallelism and triads regularly. So far I have not seen Pangram flag any of my articles as being other than 100% human-written. I wonder if it could actually save us from relying so much on crude heuristics to clock AI?

0 reposts6 likes

Eryk Salvaggio
@eryk.bsky.social

For pro writers, less of an issue. From the pov of its most-used domain, student writing — kids growing up in this environment without having their skin thickened or language mastered — if the risk of automated false positives influences how they write, that’s a net loss.