Pangram is now the purity test for ethical writing. I'm a person who uses Grammarly extensively due to my dysgraphia. I assure you, even though every thought is my own, every concept in the sentence is mine, it almost always gets tagged with high AI numbers at Pangram.
And if you're a person for whom English is a second language, god help you if you use ChatGPT to correct your grammar. This policing and litmus testing that says "writing is thinking" is really quite wrong and harmful to large numbers of folks.
I feel like you are missing the point. Both you, and someone for which English is a second language, are employing thinking to write the pre-AI edited version. Grammar correcting tools causing Panagram positives is not evidence of "writing is thinking" being "wrong and harmful", but rather a limitation of Panagram.
Second, I am doubtful that chatGPT/Grammarly fixing spelling errors / conjugations / punctuation would cause false positives. If those tools are restructuring your sentences, I guess I can understand but I would argue restructuring sentences also subtly changes their meaning. Just the camera caused some of the best paintings to no longer pursue realism, so to might AI cause the best writing to no longer be slick, "correct" and clean prose.
Sure, AI writing is mediocre. But what about people's writing? I have been in the world of professional work for decades and have seen that most people cannot write well. And by that I mean their words do not convey a clear idea efficiently. They ramble, they add extraneous material, they don't structure arguments well.
Your challenge is not that AI is slop. It's that AI is "good enough" to allow most people to communicate better. Is it fair to those people to stigmatize them because they used a tool to write better or more quickly?
From now on we should assume that all writing uses AI. Writers are free to certify themselves 100% AI free. And Pangram can suss out the liars.
This discussion lacks nuance, and the real substance lies precisely in that omitted nuance.
First, there are writers with dyslexia, who speak English as a second language or struggle with dense phrasing. Clearly, they benefit from these tools. If the standard was 100% solo authenticity, we would not have human editors, not to mention speechwriters and ghostwriters. Is a piece that goes through a back-and-forth iteration with a human editor or peer reviewers ever truly authentic?
Second, not all writing is created as equal. Some writing is about wordsmithing; but often one needs to communicate ideas clearly and efficiently. It is too high of a standard to expect for most professional writing to take us on a linguistic adventure.
Finally, this tool cannot distinguish if the idea transmitted was authentic and/or worthwhile and that is what ultimately matters. Absent that, we are unduly stigmatizing those who use AI responsibly.
My favorite quote from this episode is ‘as a lover of language…’ because I resonate with that and would love to hear more about Derek’s thoughts on loving language and how AI is making language…not as lovable.
Great episode. I’m surprised by some of the comments.
There are ALWAYS edge cases - “what about people with dyslexia? What about people for whom English is a second language?” - but I think readers/listeners are missing the broader point if that’s what you’re taking away from that episode. I felt like he was talking about writing as a means of communicating between two humans - the writer and the reader. I know if read a piece by Derek or Ezra Klein or Helen Lewis etc that they have done the work to understand what they’re talking about, and have used judgment and, most humanly of all, personal taste to create something they consider to be compelling and persuasive. AI cannot do that. It cannot develop taste, though it can try to mimic it.
I’m happy to be corrected, but Derek’s episodes/essays can be looked at on their own, AND as a corpus of work which is aimed at understanding and then explaining the Big Topics of our modern world. To me, it wasn’t about what a minority of people gain from ai writing, but what does humanity lose.
I don’t think the argument being made is that “writing is thinking”. However, it’s unarguable that “writing expands thinking”. Derek and others don’t write 7,000 word essays for the good of their health. If Abundance could have made its points as effectively in 300 words, it would have been 300 words long.
I’m just surprised by some being surprised that a human writer - which is a short way of saying someone who is curious for a living and thinks deeply about how to widen understanding and how to convey that best to a human reader - is defending human writing.
I can both agree with your take that AI reduces authenticity and think that a badging system will always have its flaws. I’m not a fan, at all, of the 100% AI-written drivel that is continually pushed into my various feeds.
But a mix can be helpful. I have incredible human editors when I publish in other outlets. Here, I do use Claude as an editor, but then I also ignore Claude in the ways I NEVER would with one of my human editors (even after tuning a Claude project).
My columns here tend to run 100% human to 60% human, but the variable seems less how I use AI than it is my subject matter. The more “business-like” the subject, the higher my odds are of a score that says “he used AI for some of this.”
So once watermarking happens, it sounds like I'm ... doomed with the mixed authorship if I happen to clip in an improved sentence?
At some point won't the AI models be able to incorporate pangram scores into their metrics and intentionally change their style to pass the test? I'm sure pangram has some method to detect bots iterively checking their scores, but there's probably a database of existing examples they can train on.
Pangram is now the purity test for ethical writing. I'm a person who uses Grammarly extensively due to my dysgraphia. I assure you, even though every thought is my own, every concept in the sentence is mine, it almost always gets tagged with high AI numbers at Pangram.
And if you're a person for whom English is a second language, god help you if you use ChatGPT to correct your grammar. This policing and litmus testing that says "writing is thinking" is really quite wrong and harmful to large numbers of folks.
I feel like you are missing the point. Both you, and someone for which English is a second language, are employing thinking to write the pre-AI edited version. Grammar correcting tools causing Panagram positives is not evidence of "writing is thinking" being "wrong and harmful", but rather a limitation of Panagram.
Second, I am doubtful that chatGPT/Grammarly fixing spelling errors / conjugations / punctuation would cause false positives. If those tools are restructuring your sentences, I guess I can understand but I would argue restructuring sentences also subtly changes their meaning. Just the camera caused some of the best paintings to no longer pursue realism, so to might AI cause the best writing to no longer be slick, "correct" and clean prose.
Sure, AI writing is mediocre. But what about people's writing? I have been in the world of professional work for decades and have seen that most people cannot write well. And by that I mean their words do not convey a clear idea efficiently. They ramble, they add extraneous material, they don't structure arguments well.
Your challenge is not that AI is slop. It's that AI is "good enough" to allow most people to communicate better. Is it fair to those people to stigmatize them because they used a tool to write better or more quickly?
From now on we should assume that all writing uses AI. Writers are free to certify themselves 100% AI free. And Pangram can suss out the liars.
- I did not use AI in writing this post -
This discussion lacks nuance, and the real substance lies precisely in that omitted nuance.
First, there are writers with dyslexia, who speak English as a second language or struggle with dense phrasing. Clearly, they benefit from these tools. If the standard was 100% solo authenticity, we would not have human editors, not to mention speechwriters and ghostwriters. Is a piece that goes through a back-and-forth iteration with a human editor or peer reviewers ever truly authentic?
Second, not all writing is created as equal. Some writing is about wordsmithing; but often one needs to communicate ideas clearly and efficiently. It is too high of a standard to expect for most professional writing to take us on a linguistic adventure.
Finally, this tool cannot distinguish if the idea transmitted was authentic and/or worthwhile and that is what ultimately matters. Absent that, we are unduly stigmatizing those who use AI responsibly.
My favorite quote from this episode is ‘as a lover of language…’ because I resonate with that and would love to hear more about Derek’s thoughts on loving language and how AI is making language…not as lovable.
Great episode. I’m surprised by some of the comments.
There are ALWAYS edge cases - “what about people with dyslexia? What about people for whom English is a second language?” - but I think readers/listeners are missing the broader point if that’s what you’re taking away from that episode. I felt like he was talking about writing as a means of communicating between two humans - the writer and the reader. I know if read a piece by Derek or Ezra Klein or Helen Lewis etc that they have done the work to understand what they’re talking about, and have used judgment and, most humanly of all, personal taste to create something they consider to be compelling and persuasive. AI cannot do that. It cannot develop taste, though it can try to mimic it.
I’m happy to be corrected, but Derek’s episodes/essays can be looked at on their own, AND as a corpus of work which is aimed at understanding and then explaining the Big Topics of our modern world. To me, it wasn’t about what a minority of people gain from ai writing, but what does humanity lose.
I don’t think the argument being made is that “writing is thinking”. However, it’s unarguable that “writing expands thinking”. Derek and others don’t write 7,000 word essays for the good of their health. If Abundance could have made its points as effectively in 300 words, it would have been 300 words long.
I’m just surprised by some being surprised that a human writer - which is a short way of saying someone who is curious for a living and thinks deeply about how to widen understanding and how to convey that best to a human reader - is defending human writing.
Early in the podcast, Derek (who I respect and support) repeated bad misinformation from the New York Times.
If Spero hadn't been there, that lie would have just continued to spread via Derek.
This is not an attempt to dunk of Derek. Rather, question the unquestioned assumption that "human = good, AI = slop."
The internet was filled with harmful garbage way before ChatGPT. We are the slop.
I can both agree with your take that AI reduces authenticity and think that a badging system will always have its flaws. I’m not a fan, at all, of the 100% AI-written drivel that is continually pushed into my various feeds.
But a mix can be helpful. I have incredible human editors when I publish in other outlets. Here, I do use Claude as an editor, but then I also ignore Claude in the ways I NEVER would with one of my human editors (even after tuning a Claude project).
My columns here tend to run 100% human to 60% human, but the variable seems less how I use AI than it is my subject matter. The more “business-like” the subject, the higher my odds are of a score that says “he used AI for some of this.”
So once watermarking happens, it sounds like I'm ... doomed with the mixed authorship if I happen to clip in an improved sentence?
At some point won't the AI models be able to incorporate pangram scores into their metrics and intentionally change their style to pass the test? I'm sure pangram has some method to detect bots iterively checking their scores, but there's probably a database of existing examples they can train on.