Welcome to The Closing Argument, our verdict on the news, plus everything The Argument published and appeared in this week.
The Verdict, by Kelsey Piper
A lot of people hate AI for drowning the internet in slop and displacing human creatives. You might think that these people would be thrilled by a program that can identify and flag AI-generated work.
But many people who hate AI also hate Pangram, an impressively accurate AI detector that produces very few false positives, according to high-quality independent research. This loathing found its most recent expression in a bizarre Wired article that amplifies a wide range of false arguments from Pangram skeptics: that AI detectors are racist (they’re not) or ableist (they’re not)1 or don’t work (this one does)2.
Rather than starting from the premise that it would be really valuable to have a working AI detector and scrutinizing Pangram to make sure it lives up to that promise, the article starts from the premise that Pangram obviously has bad vibes and sets out to discover why. (One answer: The CEO microwaved lunch during an interview!)
Why aren’t writers who resent Pangram more eager to distinguish themselves from posters who are having ChatGPT ghostwrite all their work? There are a few different things at play, but I think one key dynamic is this: These are often the same people. A core constituency for arguments like Wired’s is “writers” who are using AI to write their posts but resent this fact being made public. They stand to lose from AI detectors because AI detectors call them out.
When Substack announced a partnership with Pangram, a sizable fraction of the platform’s authors rose up in outrage. Some people had experience with other AI detectors that didn’t work, and some felt that AI writing ought to be encouraged — both of which are fair enough. But the fury was fueled by authors passionately insisting that Pangram kept wrongly accusing them on almost every post they published.
Pangram can be wrong, especially on short passages. But in the overwhelming majority of cases where Pangram looked at a whole blog post rather than a single paragraph, the program correctly detected work as AI. Even if we can’t be 100% sure in any individual case, we can notice the trend in the aggregate.
Wired quotes Sam Illingworth, a professor of critical AI literacy at Edinburgh Napier University, who said, “I don’t think [AI detectors] work. … I think that detectors are prejudiced against certain people.” This spring, he said that he used AI for research and editing; this summer he acknowledged that in addition to research and editing, he also uses it “now and then for a turn of phrase I keep.” Pangram informs us that most of his posts are 100% AI-written (some are mixed AI-and-human) and that 100% AI-written posts started appearing last year.
The favorable interpretation is that, as someone whose recent work Pangram consistently mistakes for AI, Illingworth sees Pangram’s flaws firsthand. The unfavorable interpretation is that the tool’s credibility trades off against his own, because its account of who writes his posts is not consistent with his.
Illingworth is not alone in insisting the detector does not work. “It’s a Monday night in late July, and I’m pasting my own Substack draft into an AI detector for the first time. Not because I’m worried. Because I’m curious. I wrote every word of it myself, the way I write everything, out loud in my head first, in a rhythm that still carries the accent of a language I learned to think in before I learned to write in this one,” begins one Substack post by an author indignant that Pangram flags her writing as AI. “That’s the part I had to sit with before I could write this.”
If you, like me, have worked with AIs a lot, you won’t even need Pangram there. The author of that piece is AI. It is obviously AI! I checked 10 other posts on the blog, and they were also obviously AI!
“I have spent 22 years as an award-winning author and essayist. And now this bullshit tool that learned from MY writing is going to accuse me of plagiarism,” popular Substack writer Carlyn Beccia wrote in an anti-Pangram post. “For the last and final time, AI DETECTORS DO NOT WORK.”
Here’s what her writing sounded like in 2020: “Most success stories revolve around one person who believed enough in you to say — give it a try. We call these people muses, but they are so much more than that. They are the strong man giants whose shoulders we stand on so we can see the horizon. Throughout the editing process, I am sure my editor felt less like a giant of greatness and more like Atlas with the world on her shoulders.” Pangram says it was 100% human-written. (So were other articles of hers in 2020.)
And here’s what it sounds like now: “And then there is the reshuffling, which sounds like paperwork until you remember that empires have always loved paperwork. The Trump regime recently moved special education oversight from the Education Department to HHS. That is not a minor bureaucratic transfer. It changes the frame.”3 Pangram says this is mostly AI. I am not convinced that Pangram is wrong.
In my own experience, Pangram is very effective on long texts — if it marks a long essay as AI, the essay was either written by AI or deliberately written to sound like AI. On tweet-length or paragraph-length texts, it’s not perfectly reliable. It seems to work just as well for non-native English speakers as for native speakers. While I have seen dozens of claims of pre-2022 text returning a positive result on Pangram, I have tried repeatedly to corroborate those and have never once succeeded.
To be clear, I’m absolutely not saying that complaining about Pangram suggests a writer is a cheat. There are plenty of other concerns. Most are mistakenly premised, such as the false claim that Pangram trains on work you put into it, but others are reasonable: It’s unreliable on sufficiently short text segments; it creates an internet witch-hunt mentality when there are defensible uses of AI; people who are open about their AI use are sometimes blamed, when the only people really doing something wrong are the ones who hide it; and it isn’t perfect, so a single positive result shouldn’t be taken as decisive proof of misconduct.
I’m happy to concede that Pangram is imperfect and possible to over-rely on. If an author’s posts are mostly non-AI and one scans as partially AI, this could be a mistake. It’s also possible to defeat Pangram with intentional effort. If you think we need something more robust and more reliable, fine! But if you don’t like AI and don’t like Pangram, your attitude really ought to be that it’s a shame that our existing tools for AI detection are imperfect, and it’s important to invent better ones.
If your stance is instead that AI is evil and so is looking for AI, then ultimately you’re contributing to the sloppification of the internet. You’re making it harder for people who write their own text to distinguish themselves, and providing cover for authors who are definitely not writing their own text.
Top stories this week, by Kobe Yank-Jacobs
Commentators often worry that the decline of reading could mean the decline of liberal democracy — but deliberative governance actually predated widespread literacy. This week, Maibritt Henkel pointed out that many liberal cultures thrive on a culture of raucous, interactive debate, which can happen in-person as easily, if not more so, than in text. Read on, dear reader, to learn more. It may actually cheer you up:
It used to be bad to downplay climate change (if we mean denying its reality). Today, it’s actually good to downplay climate change (if we mean keeping the issue out of political fights so we can make real progress). This counterintuitive view is called “climate hushing,” and contributor Alex Trembath argued that it’s the key to enacting good policy going forward. Get up to speed on the latest in climate politics:
Canada’s Prime Minister Mark Carney is standing up for the forces of liberalism against Donald Trump’s assault on our closest ally. So, why is one prominent left-wing commenter attacking Carney? Columnist Matthew Yglesias explains:
🌟Abundance Wins of the Week🌟
Shout out to my home state of California, whose legislature passed two housing bills this week. One requires a housing density minimum in transit-oriented areas of the state’s most transit-rich cities (Los Angeles, Sacramento, San Francisco, Oakland, San Jose, San Diego, and Long Beach). Another bill closes two common loopholes used by third parties to delay housing projects, including an opportunity to challenge maps of a given lot created by the developer after a project has been approved.
Shout out to our big-state rival, Texas, where pro-housing SB 785 took effect this week. The new state law requires municipalities to have at least one district permitted for manufactured housing without a special permit. This will promote zoning changes in 44% of municipalities.
The company Fervo Energy signed a deal with Google for what is set to be the world’s largest enhanced geothermal facility, proving a technology can work at a scale that could be crucial to a decarbonized, energy-abundant economy.
Worth watching…
This week, forecasting legend Nate Silver joined Jerusalem, Matt, and Jeremiah on a crowded episode of the pod. It was a party. And speaking of parties, our round table debated the factional lines within the Democratic Party: Is it possible to be center-left and also anti-establishment? Watch or listen to find out:
The Rio Grande Valley in Texas is one of the areas that swung hardest toward Trump in 2024, but musician Bobby Pulido is running to flip it back. Right now, The Argument/Split Ticket’s model has him running 1.1 points behind his Republican opponent in a district Trump won by 18 points. Tune in to find out how he differs from national Democrats and what he thinks he can to do win this seat.
On Thursday, Bernie Sanders was among the first politicians to call for a ban on superintelligence. My guest this week, Andrea Miotti, has been lobbying to ban superintelligence for several years, most recently in the U.K., U.S., and Canada. I asked him how he gets legislators to take speculative risks seriously, and we dug into what plans might look like to actually pass a ban. Listen to learn more:
What’s News with The Argument
The Argument recommends, by Kobe Yank-Jacobs
If we ever lose track of Justin Zuckerman, I’ll have the team split up and check all the movie theaters in the city. I’m shocked (and a little jealous) at how often he gets out to see good cinema.
This week, he brings us It Ends, a “horror-ish movie,” which he described as a “Gen Z Waiting for Godot.” I was intrigued before he added that it was “funny, dark and existential” — at which point I was fully locked in.
I guess I also had existential themes on the mind this week. One morning, a single line from a poem I read long ago just came to me: “Is death a stretch of time in which a life is just a flash?” The poem is called “From 20,000 Feet” by Heather McHugh. I suggest the version in The Vintage Book of Contemporary American Poetry, which, by tradition, my best friend and I always bring backpacking, extra weight be damned.
We’ll stay thematically dark for just one more rec from senior editor John Robert Thomason, whose upcoming trip to Baltimore reminded him to suggest Michael Clune’s memoir White Out (about being addicted to heroin in Baltimore in the 1990s). Ben Lerner’s blurb called it “Disturbing, brilliant, hilarious—it’s as if Proust had written Jesus’ Son.” John told me this “seems accurate, with the caveat that I’ve never read Proust.”
I, personally, have also shelved Proust until retirement.
At this point, I was a little concerned for the team since all of us tossed out such psychologically grim material — maybe we’re all down about the end of summer — but thankfully, at least a few of us were into more upbeat selections.
Milan Singh recommended Made In’s 5-quart stainless clad saucier, suggesting he had some fun in the kitchen.
Jeremiah Johnson asked if he could recommend yet more Carly Rae Jepsen. Once denied, he offered up Game Changer instead, calling it “the most wildly creative game show, maybe in the entire history of the genre.”
Eli Richman mentioned Lucky, “a B-tier show with an A-tier cast and an S-tier theme song.” I suppose he was recommending the song, Horns of a Bull.
Still, despite these cheerier recs, I couldn’t help but be concerned that we were subconsciously worried about racing toward summer’s end — the end of swim trunks and splashing around. Then a message landed reminding me of the three-day weekend to come. A whole extra day for sunshine. One more swim in the lake.
I was grateful this called to mind yet another line from a poem, one by Mary Oliver, the patron saint of every gorgeous summer day (and about whom a documentary is now out). Speaking to God in the poem “Thirst,” she wrote:
grant me, in your mercy,
a little more time. Love for the earth
and love for you are having such a long
conversation in my heart.
We have merch!
We have quarter-zips, keychains, hats, and stickers. Each one is a great conversation starter in its own way. Buy them here.
More to read:
Seeing like an elite
What do Americans hear when they watch Sen. Bernie Sanders or Vice President JD Vance rail against the elite? Are they thinking of right-wing tech billionaires? Imperious college professors? Liberal magazine writers?
God is back on the timeline
This week, we’re talking about Substack’s new religious debates, the continuing MAGA influencer scandal, how to retire from YouTube, and aura battles.
This was a legitimate thing to worry about, but doesn’t seem to be a problem given the actual technology: Marijke Van Vlasselaer, Filip Van Droogenbroeck, and Bram Spruyt’s 2026 paper looks at whether Pangram is less accurate when evaluating the work of non-native speakers of English. It isn’t. That’s because Pangram isn’t really looking at how “unusual” a piece of writing is but whether it has the characteristic patterns of LLMs.
The Wired article makes an odd gesture at a different Pangram-is-racist argument. There were several controversies this summer during which upcoming books were canceled after being discovered to be AI-written: “The three novels mentioned earlier — Shy Girl, Daggermouth, and Call Me, easily publishing’s biggest AI-detection scandals — were all written by writers of color.” But come on! As Zvi Mowshowitz put it: “They were not primarily written by people of color. They were primarily written by AIs.” That’s the whole problem!
And as my colleague Jeremiah Johnson recently observed with respect to Cambridge’s Arday scandal, a community being unwilling to call out fraud if it’s perpetuated by people of color is not an anti-racist tendency; it is ultimately quite racist.
Many non-Pangram detectors don’t work, but many possible airplane designs also don’t work, and no one considers this a general argument against aviation; the thing that matters is whether any detection works.
The phrase “empires have always loved” mysteriously became all the rage in the last few years. See here, here, here, here and here. Before 2025, the only use of that phrase I have been able to find is a single Tweet. The logic of the sentence, too, is incoherent in an AI-characteristic way. “...sounds like paperwork, until you remember that empires have always loved paperwork”? Does remembering that make it sound less like paperwork? It’s a craft-ful sentence, but what is it actually saying??
1. This kind of technology is inevitable and necessary. I'm glad Pangram exists.
2. Like just about every supportive piece on this topic, this one underestimates the technological challenges to what AI writing detection is doing. I think this is a vestige of the fact that wordcels are having these arguments, frankly. Pangrams one in 10,000 claimed false positive rate is impressive, but too many people are treating that as though it's indistinguishable from one in 1 million. A false positive rate of one in 10,000 means that every single day millions of false positives are being produced by humans beings. And I have no idea why discussing that basic technological reality is treated like a sign of bad faith.
3. Pangrams own founder has been much more cautious and limited in his claims then most of the boosters of his product. It's all very strange.
Okay I’m at least somewhat persuaded by this reasoning, but that then nudged me to ask - Kelsey, what do you make of people who are overtly/publicly anti-AI but also using AI to do their writing?
Is it that their public position is performative? Cognitive dissonance? Viewing only their own use of AI as acceptable? (I’m skeptical of the last one because using AI to author whole essays goes way beyond using it for research — and I say this as someone who uses AI a ton).
I guess this just ties into a larger theme that I’ve been wondering about, which is how salient anti-AI / anti-data center views really are. It’s easy to oppose something on the internet, and slightly harder but not actually that much of a burden to show up to a city council meeting to scream nonsense at your local government… but when actual push comes to shove? Perhaps all these people using AI to write suggests something.