This morning, Substack announced that it would be integrating Pangram scans to detect AI content in its posts. I wanted to spend my day revising a chapter of my next book, and later this week, I was thinking of writing a post on the “Memex” I have created using plugins for Zotero and Obsidian that I have coded using AIs to help me in my scholarly work. But instead, I have had to do some serious thinking about the future of my place on the Substack platform.
Now there were plenty of issues with Substack before: it has monetized Nazis, it prioritizes itself over its authors, and—although it billed itself as supporting discovery of small voices
it has recently followed the way of all social media corporations, shoveling large accounts down our throats through its crappy discovery mechanism and creating a social media platform (Notes) that nobody wanted.
The problem is what to do instead—Ghost or a WordPress-based newsletter? Patreon? what? (I would absolutely love your suggestions here)—as hardly anybody uses RSS readers anymore and the old days of sites like Archinect or Things Magazine driving new visitors to my site are over. I just don’t know. So in the meantime, I will be using this feed to announce my articles, but from now on, I will be hosting them where they belong—on varnelis.net, a site that I have owned and built over 25 years.
Now on to my deep, abiding hatred for Pangram and why you should hate it too.
Pangram is an AI for the AI haters. Pangram’s very young CEO, Max Spero, failed to sell his product to universities. At this point, dozens of universities, including MIT, NYU, Georgetown, the University of California, Berkeley, and Yale, have banned the use of AI detectors. The reasons are rather obvious. Much as Luddites on both the far left and far right would hope, we are not going to stuff the AI genie back in the bottle. AI, like Google and the Web, spelling and grammar check, Wikipedia, printed books, or writing itself, is a tool. What matters is whether we choose to use it intelligently. AI detection is notorious for false positives, and universities have wisely understood that calling out a student for using too many em-dashes, negative constructions, or triads is a ticket to a lawsuit. That AI detectors are especially prone to flagging neurodivergent students and students for whom English is a second language only makes it worse. So what was young Mr. Spero’s response? To shame journalists—especially those critical of him—instead. Calling himself a “slop janitor” on X, Spero has accused journalists of using AI in their writing, most notably a journalist for the Guardian.
Clearly, Pangram is trying to exploit the rampant fear of AI in the US, which is, frankly, a bit off the deep end these days. Death threats against people who use AI (even those who acknowledge using it) are not uncommon on left-leaning social media platforms such as Threads and Bluesky. The pathological fear of this technology by many deserves a deep dive of its own, and the Generative Unconscious takes that fear seriously, excavating its origins through a historical narrative; but one also has to ask: cui bono? Pangram uses AI to profit from such fear, a deeply cynical project.
Using AI—even to write an article—isn’t a guarantee of slop, nor is it required to produce slop. On the contrary, Substack’s own On Substack newsletter is full of slop, like “A Business Model That Works for Creators” or “The Global Ideas Exchange,” both of which are proof that you don’t need an AI to write like one. Substack also encourages low-effort use of AI through an image generation tool it offers writers. It even offers AI-generated “podcasts” to authors. So AI-generated images and sound are ok at Substack, but the use of AI in writing is not. Or at least for now. It’s all too easy to imagine that Pangram or Substack will offer an AI humanizer, as Grammarly does, to dodge the AI detectors, for a fee, of course.
Do I use AI? Of course. I am writing two books on AI in contemporary culture and, frankly, if I did not use AI, you should deeply mistrust anything I write. As anyone who has looked at my artwork can see, I have worked extensively with technology for two decades now—from photography to print to code to AI. I have also written pieces such as “Fables of Accelerationism” in express collaboration with AI. Did I use AI to write this article? No, I wrote every word, although I did have Claude Opus look for and correct grammatical and spelling mistakes. But what’s the next step? If we don’t get outright attempts to shame writers or death threats from the wingnuts, I imagine this corporation, valued at over $1 billion, will start evaluating its creators on the basis of whether or not they used AI in their writing even as it encourages us to use AI in our posts. It’s all a bit rich.
Like all tech hysteria, this too will pass. AI writing will get better (Anthropic’s Fable, for example, is much better at writing than any other model) and doubtless someone will come up with a way to post-train AIs to emulate any given author’s work. The war against the AI detectors will come to an end, and these scammy startups will end where they belong, in Chapter 11 bankruptcy protection.
I had other things to do today, and those all got derailed by this nonsense, but between the platform’s support of Nazis and its desire to shovel large accounts down my throat (and yours if you are reading this there), I’m done. It’s been a colossal pain dealing with Substack’s restrictions on length, the way it butchers HTML code, and how it deals with images. You will hear from me soon—assuming there are no more major home appliances breaking down (thanks, Miele), insane thunderstorms spawning tornadoes in town, or other disasters in my life. And if you are on Substack, you will still be able to follow when I publish something. But Substack doesn’t get my content anymore. It cost me a day of research and writing, but thanks to the billion-dollar corporation for making it easy for me to finally take a step I have been pondering for a while.