Part of the reason some people like the score is that they don't trust writers to disclose, or to accurately disclose, their process. Ironically, the score itself is generated by a language model.
I totally agree, I wrote this a few days ago, and I am glad someone else is laying out a similar point: Pangram arriving on Substack was inevitable. This is not a good or bad thing. People were already scouting (policing) each other’s writing for AI signals—it’s just that now the thing they hate so much has become the easy-access tool to validate their suspicions. I use AI plenty (although not in this note), but apparently all I have to do, when I do use it on Substack, is provide you with a “How I make this” statement to explain my ridiculous process of wrestling with my self-doubt and the big three AI text editors, so I can “set expectations” for readers.
As if procedural disclosure (fighting the self-made three-model crucible over which edits to keep, revert, or synthesize into something I’ll put my name on) is going to reveal anything the writing itself cannot.
What they don't see (yet) is how the AI-detection tool will be turned against us. In the name of protecting us from AI slop, our capacity to read for understanding—to recognize meaning, or to judge at all—is now outsourced to yet another tool of administration.
The problem is not the writers but the readers. I am neither an AI technophiliac nor an AI technophobe. I refuse to submit to either position because I see both extremes as expressions of the larger problem of authoritarianism. In either case, we are granting technology an autonomous, alienated power before which the subjective freedom of human reflective judgment can only submit or retreat.
I will have no problem saying the responsibility is shared between writers and readers. :)
I certainly take the responsibility over what I put my name to. I am not using pseudo names: my journalistic writing and experimental, critical writing all carries my legal name.
I think I was really reacting to the sense that we are all engaging in the topic from the perspective of writers as producers, which we should, but the issue sometimes being neglected is also about reading, bc what matters as much is what is happening to the experience of reading (we are policing through reading) and the effect that is having on literacy, which already has been in marked decline since around 2010-2012 with social media.
Yes! I agree, to me though, I don’t like to say that this is on the readers, I think writers share some of the responsibility.
We have to own what we publish, and AI can certainly be used to manufacture empty authority at scale. But I agree that procedural disclosure cannot solve that problem. A detailed account of which tools touched a draft may describe the machinery without revealing where the judgment came from.
Part of the reason some people like the score is that they don't trust writers to disclose, or to accurately disclose, their process. Ironically, the score itself is generated by a language model.
True…
Also it’s a language model, but not the generative type like an LLM, I don’t think so. It’s probably more of a ML proximity algorithm
Their model card says it's a decoder-only transformer with a classification head: https://www.pangram.com/research/model-card/pangram-3-1
I totally agree, I wrote this a few days ago, and I am glad someone else is laying out a similar point: Pangram arriving on Substack was inevitable. This is not a good or bad thing. People were already scouting (policing) each other’s writing for AI signals—it’s just that now the thing they hate so much has become the easy-access tool to validate their suspicions. I use AI plenty (although not in this note), but apparently all I have to do, when I do use it on Substack, is provide you with a “How I make this” statement to explain my ridiculous process of wrestling with my self-doubt and the big three AI text editors, so I can “set expectations” for readers.
As if procedural disclosure (fighting the self-made three-model crucible over which edits to keep, revert, or synthesize into something I’ll put my name on) is going to reveal anything the writing itself cannot.
What they don't see (yet) is how the AI-detection tool will be turned against us. In the name of protecting us from AI slop, our capacity to read for understanding—to recognize meaning, or to judge at all—is now outsourced to yet another tool of administration.
The problem is not the writers but the readers. I am neither an AI technophiliac nor an AI technophobe. I refuse to submit to either position because I see both extremes as expressions of the larger problem of authoritarianism. In either case, we are granting technology an autonomous, alienated power before which the subjective freedom of human reflective judgment can only submit or retreat.
I will have no problem saying the responsibility is shared between writers and readers. :)
I certainly take the responsibility over what I put my name to. I am not using pseudo names: my journalistic writing and experimental, critical writing all carries my legal name.
I think I was really reacting to the sense that we are all engaging in the topic from the perspective of writers as producers, which we should, but the issue sometimes being neglected is also about reading, bc what matters as much is what is happening to the experience of reading (we are policing through reading) and the effect that is having on literacy, which already has been in marked decline since around 2010-2012 with social media.
Yes! I agree, to me though, I don’t like to say that this is on the readers, I think writers share some of the responsibility.
We have to own what we publish, and AI can certainly be used to manufacture empty authority at scale. But I agree that procedural disclosure cannot solve that problem. A detailed account of which tools touched a draft may describe the machinery without revealing where the judgment came from.