AI definitely has its place, and I totally understand why you would use it for writing posts in English. But I don’t want it anywhere near my own writing, and that includes any AI-driven AI detectors. My question is, do I now have to go back through nearly four years of Substack posts to make sure it’s turned off on every one?
No, Dan, it’s only applicable to posts published from July 21 onwards. Everything posted after that date is automatically eligible to be checked by the AI checker. You can access it via the three dots on the right-hand side.
Anything published before July 21 isn’t available to be checked... for now.
I’ve also emailed them with my concerns. Never got a reply.
I should add though, Dan, that your content is automatically ingested as well. All the content is. You can't turn that off. So even if you deactivate the AI checker, your content will still be ingested by the Chatbot. The only thing you are switching off is the evaluation possibility itself. Substack always wins
I also completely understand their reasoning behind this. I think Substack as a community has a great fear of Ai posts turning the site into slop, which has happened on LinkedIn.
I do find it troublesome though, particularly for cases like yourself, none native English speakers. As well as for individuals whose writing skills are not great but still have interesting ideas that they want to share, and for the most part, I may want to hear.
I think potentially both sides could be satisfied, if the detection tool was looking more for original thought, and value in the work it scans (something that could be done, and with a lower carbon footprint). We’re then provided with a score, that’s potentially tailored to the individual reader’s interests. Although there are other potential dangers with that also!
If it’s any consolation, it found every chapter of my 2004 PhD dissertation to be Ai.
I’ve been thinking about this for the last few weeks, Howard. I don’t think there’s a good solution, because a “detection tool” always risks turning the discussion into a black & white one, and that’s not particularly helpful when we’re talking about AI.
Furthermore, from what I’m seeing in the music business, I’m not at all convinced we even need these tools. The slop will eventually be rooted out automatically once the economic incentive to create it disappears. I can usually tell when something isn’t human-written anyway, and then it’s my choice whether I want to read it or not.
The thought that kept coming back to me these past few weeks was: what if it’s not about the detection, but about the training?
Substack sits on a mountain of extremely valuable content, but they can’t simply sell all of that to an AI chatbot. This way, they could potentially use it and reap the benefits in the long term without having to ask permission or offer any financial incentive. I really hope I’m wrong about that, but I can’t completely rule out the scenario. Substack wouldn’t be the first brave new tech company to take the shortest route to the money, egged on by the VCs funding it.
I’m not too bothered about the detection itself. People read my stuff or they don’t. You can’t be everything to everyone. But the process this morning, getting three different results on exactly the same text, really made me angry and pushed me to write this post. I now know that I write extra fast when I’m all worked up 😁 That was a new sensation too.
And what you say about your PhD is crazy, btw. But it also proves that what we currently have on the detection front raises some very serious questions.
Well said, Pe. It's absolutely bonkers that the percentages differ so much by just changing one or two things like an emoji and a video. I loved how you framed this in terms of your views on AI in general, the practical use you benefit from, the reliability of AI detection scanners, and the fuckery and noise that can really distract us from the joy of writing. Bravo.
Bravo Pe. Substack needs to make shutting this off a global setting so it need be done only once. I don't use AI. It never would had occurred to me you did. So this human reads you as 100% human. Checking grammar is no different than editing tools.
It seems Pangram is wreaking havoc in the publishing industry, generating results that have led to allegations that authors have used AI when those authors insist they haven't.
Such an interesting article, Ellen. It's a perfect illustration of the Twilight Zone we're currently in. There's this black-and-white discussion raging around AI, and the one thing we know is that black-and-white discussions aren't helpful. They turn it into a Luddite debate at exactly the moment when we should be discussing how we can use this new technology to make things better, and what guidelines and boundaries we need to put in place. Just like we did with nuclear technology after WW2.
Zooming in on my own little rant, it reinforces what I said about Pangram. And, by extension, the necessity to look beyond what is happening on Substack and ask ourselves why it is happening.
They (as in the writers of the piece you shared) don't have the answers. Nobody has. But they recognise that we're living through interesting times. In many ways, publishing hasn't been disrupted that much by technology in the past. This is probably their Napster moment. And looking at industries that have already had their Napster moments can help us understand what is happening today.
If I look at the music industry, it's going through the same convulsions, amplified by the whole DSP/fair remuneration discussion. But I do see a consensus slowly building around using AI to speed up, help and democratise parts of the creative process, while rejecting AI slop.
The same goes for photography, which has been completely transformed by technological evolution over the past 20 years. But that doesn't mean there aren't still great photographers producing great work and doing well today.
What people often forget is that we're living in a world that moves fast. Standing still yourself is never a good option.
What bothered me as someone who's published books is 1) conducting an AI witch hunt with a pitchfork called Pangram when who said you can't use any AI whatsoever (who made that rule?! which is what you're also saying), and 2) using a platform that has not been proved to be reliable and legitimate as a weapon for enforcing compliance.
The careers and incomes of authors are what's at stake here, as usual the artists the ones being held hostage while these 'experts' engage in long-winded debates. If their own income were on the line, they'd cut through all this nonsense and get right to the heart of the matter. Which is agreeing on those guidelines and boundaries you talked about.
"After Pangram identified three of his writers' articles as AI-written, editor James Taranto of The Wall Street Journal called the tool a "defamation machine". Two of the authors admitted they used AI for revision of some of their work, which Taranto said is "inaccurate and unfair to characterize" as AI-generated...
According to Matteo Wong of The Atlantic: "While Pangram is accumulating the power to end reputations and careers, the tool does make mistakes, perhaps to a greater extent than is currently understood. In turn, AI accusations could very quickly spiral into a witch hunt...Just like chatbots, AI-detection tools have become effective enough for widespread use, but not reliable enough to fully trust." Wong also raised concerns about the black box nature of Pangram's algorithm, similar to other neural networks; while Pangram can show what part of a text is likely to be generated by an LLM, it cannot describe why it was flagged."
It's quite problematic to use a program like this that's not tested and proved.
Thanks for all the info, Ellen. There’s one sentence on Pangram’s rather sparse website that really made me laugh out loud: “We will aid the effort to ensure that the new generation of powerful language models will be a net positive on the world.”
It depends on what “net positive” means to them, of course, and that’s something they don’t say. 😁
Sites like Medium (and, of course, LinkedIn) were overrun with slop. Both have finally taken measures to counter it, butI'm not sure they moved in time to save themselves. Substack probably saw that and saw integrating Pangram as a preventative strike. If your business model is built on quality writing- and people paying for the same- having a platform flooded with shit writing represents an existential threat. My assumption is that it was meant more to prevent AI slop from showing up in the first place than to have all of us run litmus tests on each other.
You nailed it further upthread when you said it will disappear as soon as the financial incentive to create it disappears. The sooner that happens, the better. If having Pangram here accelerates that fall, then I'm okay with it.
I share your opinion on the sooner AI slop disappears, the better, Kevin. I had to look into the Medium and LinkedIn solutions because I didn’t know what their approach was. The latter we can probably classify as window dressing, I think. But then again, LinkedIn has always felt a bit like Facebook with some different accents.
The Medium approach I found much more interesting, because they make a clear distinction between AI slop and AI assistance. The latter isn’t classified as AI-generated content. I’m not on the platform, so I can’t judge whether it actually works or not, but the basic principle seems to me to be much better than what Substack is doing.
On paper, Medium's strategy might look better; the problem is they waited far too long to take any action. They also continue to waffle, condemning AI slop one minute and offering premium subscribers access to an AI-based writing tool the next.
Thanks for this Pé! I’m no fan of AI in the creative space, meaning for visual, audio or textual content created solely or primarily through AI. But there’s clearly a place for some usage of AI tools, particularly for non-native English speakers or for people trying to improve their grammar with things like Grammarly. Or even for people creating headers or other images with AI that they want to include in their posts.
I don’t mind the idea of an AI detection tool — I certainly wish all music streaming sites let me opt out of AI music — but I do need to believe I can trust the tools and that there’s some level of consistency. Your research tells me there isn’t.
Having said that, I’d like to think that people know what I write isn’t created by AI. I’ve certainly NEVER suspected any of your work was in any way AI-produced (nor for any other music writers I follow here). But if someone feels like they want to check my work for AI, I’m not sure I care. Feel free to believe my work is written by me. Or not. And act accordingly. It doesn’t really bother me.
My bigger concern is Pangram is potentially ingesting my work. It’s not a concern from a creative perspective because I’m not the world’s next great music writer 😆. But I don’t want anything I’m writing to be training the next generation of AI in any way, shape or form. That would be the reason I might turn off permissions for Pangram scanning of my work.
All valid points Mark. As for turning off Pangram, you touch on the central issue there: it isn’t possible. Your work is ingested, whether you like it or not. The only thing we’re allowed to do is turn off the evaluation tool for readers. But that doesn’t mean the evaluation itself isn’t happening. We have absolutely no control over that, and Substack hasn’t asked for our permission.
Which begs the question: why?
As a historian, I’m trained to look beyond the source, to ask questions about what it is telling me, but also about what it isn’t telling me. And that’s exactly what I find myself doing here. But anyway, sadly there’s very little we can do about it.
As for AI chatbots themselves, when used correctly they can be tremendous learning tools. My English was already pretty good when I started writing, but the way I work with a chatbot is that it corrects me and explains why it has made those corrections. It’s incredible how much that has helped me deepen my knowledge of English grammar and expand my vocabulary.
I also don’t think the kind of unlimited access we currently have to these chatbots will remain this way forever. It might end sooner than we think because the economic model behind it simply doesn’t make much sense. So, for once in my life, I’ve used the time available to me extremely well 😁 and learned whatever I can from it. And that learning process has become part of the fun of creating for me.
The reason I’m saying all this is to illustrate my central point: AI is not a black & white discussion. Yet the way Substack, and other platforms, are handling it is turning it into exactly that kind of discussion.
Excellent post, PeDupre. Substack should say thank you for using available tools to make your writing more accessible and informative for your readers. If they don't say it, I will.
I’m with you completely, Pé. I think the pangram is completely unnecessary and detracts from time spent reading and enjoying our writings, even if they were written 47 percent by Martian. It never would have occurred to me to wonder about your AI usage. Your stories and deep dives are so compelling — that’s what stands out.
It’s intriguing, your testing with the images. Were the images AI created? It does seem like the percentages are random. Your results are proof that this is a garbage tool designed to….what? Appease a disgruntled populace? Or maybe, as you say, it’s a way to train AI on our stories (and sell the content).
The percentages are indeed random, Steve. I did a lot of testing over the past few weeks because I really wanted to understand how it worked. The images weren’t AI-generated either. They were images I found on the net, like a photograph of Montreal in the fifties or a postcard from Expo 67. So it’s completely incomprehensible to me why Pangram reacted the way it did.
That said, the moment I decided to block any further “evaluations” actually felt quite liberating. We should probably all do it, if only to make Substack understand that this isn’t working and that they need to change tactics.
Not that I’m convinced they would, because, as I said in my reply to Howard, I’m beginning to strongly suspect that the real agenda here may not be quite what they want us to believe it is.
Excellent points!! Ai is here to stay! For non native speakers it is a great help, more even if one is dyslexic ... like me). Perhaps our challenge is to write content that is so unique, that it reveals our singular intelectual contributions.
I find that it is essential, to command the machine "do not add adjetives" "do not editorialize". Then carefully revise that the text in your non-native language really shows your personality, you insight into the topics ... Just my 0.02 Best José
I constantly change the prompts I’m using, José. That’s one of the things I find so useful about chatbots: it’s a learning curve. And although my English was already quite good when I started, I’ve learned a lot since I began working with ChatGPT and Claude. I find them incredibly useful tools for deepening your knowledge of a language.
I even use them now to improve my other languages, including my French. That’s exactly what I meant when I said this isn’t a black & white discussion, so let’s stop turning it into one.
Anyway, thanks for your reaction. Really appreciate it!
Yep, it is, and it’s been a constant factor these past few weeks, Jonathan. It simply doesn’t work properly. And on top of that, I’m not even convinced we need it. The AI slop will eventually be rooted out once the economic incentive to create it disappears.
Which begs the question: why introduce it in the first place?
I can’t help wondering if we’re looking at this the wrong way. What if it isn’t really about detection, but about training? Substack can’t simply use the mountain of valuable content sitting on its platform directly, but could something like this provide an indirect route to doing so? It would certainly make the decision to launch it without properly announcing it or asking writers for permission look rather different.
I hope I’m wrong, but I can’t rule it out. And if that were the case, it would be beyond the pale, if you ask me.
Well, I can't speak to Substack's corporate motivations. It sure seems though that they want to gesture toward supporting actual human authors by creating a "checkmark," except that the tool is too blunt for the purpose.
But really, there is a category difference between videos of animals wearing silly hats and mixes of disco songs from a half century ago. If you're still fascinated by fabricated videos of animals wearing silly hats (and over the age of 13), there's no hope for you. Happily, people's perceptions of disco mixes, which used to be on the level of animal flicks, have been changing, for the better. I hope you can keep making new episodes of The Twelve Inch and keep informing your readers (and listeners) about great music of the past. I don't think it's something that we need AI to appreciate.
AI definitely has its place, and I totally understand why you would use it for writing posts in English. But I don’t want it anywhere near my own writing, and that includes any AI-driven AI detectors. My question is, do I now have to go back through nearly four years of Substack posts to make sure it’s turned off on every one?
No, Dan, it’s only applicable to posts published from July 21 onwards. Everything posted after that date is automatically eligible to be checked by the AI checker. You can access it via the three dots on the right-hand side.
Anything published before July 21 isn’t available to be checked... for now.
I’ve also emailed them with my concerns. Never got a reply.
Good to know — thanks!
I should add though, Dan, that your content is automatically ingested as well. All the content is. You can't turn that off. So even if you deactivate the AI checker, your content will still be ingested by the Chatbot. The only thing you are switching off is the evaluation possibility itself. Substack always wins
Well said, and I feel your pain.
I also completely understand their reasoning behind this. I think Substack as a community has a great fear of Ai posts turning the site into slop, which has happened on LinkedIn.
I do find it troublesome though, particularly for cases like yourself, none native English speakers. As well as for individuals whose writing skills are not great but still have interesting ideas that they want to share, and for the most part, I may want to hear.
I think potentially both sides could be satisfied, if the detection tool was looking more for original thought, and value in the work it scans (something that could be done, and with a lower carbon footprint). We’re then provided with a score, that’s potentially tailored to the individual reader’s interests. Although there are other potential dangers with that also!
If it’s any consolation, it found every chapter of my 2004 PhD dissertation to be Ai.
I’ve been thinking about this for the last few weeks, Howard. I don’t think there’s a good solution, because a “detection tool” always risks turning the discussion into a black & white one, and that’s not particularly helpful when we’re talking about AI.
Furthermore, from what I’m seeing in the music business, I’m not at all convinced we even need these tools. The slop will eventually be rooted out automatically once the economic incentive to create it disappears. I can usually tell when something isn’t human-written anyway, and then it’s my choice whether I want to read it or not.
The thought that kept coming back to me these past few weeks was: what if it’s not about the detection, but about the training?
Substack sits on a mountain of extremely valuable content, but they can’t simply sell all of that to an AI chatbot. This way, they could potentially use it and reap the benefits in the long term without having to ask permission or offer any financial incentive. I really hope I’m wrong about that, but I can’t completely rule out the scenario. Substack wouldn’t be the first brave new tech company to take the shortest route to the money, egged on by the VCs funding it.
I’m not too bothered about the detection itself. People read my stuff or they don’t. You can’t be everything to everyone. But the process this morning, getting three different results on exactly the same text, really made me angry and pushed me to write this post. I now know that I write extra fast when I’m all worked up 😁 That was a new sensation too.
And what you say about your PhD is crazy, btw. But it also proves that what we currently have on the detection front raises some very serious questions.
I agree with you 100%
I wonder what AI would do with mine from 1987 ;-) ;-)
Well said, Pe. It's absolutely bonkers that the percentages differ so much by just changing one or two things like an emoji and a video. I loved how you framed this in terms of your views on AI in general, the practical use you benefit from, the reliability of AI detection scanners, and the fuckery and noise that can really distract us from the joy of writing. Bravo.
Thanks Andy, I now know that I write very fast when I'm super worked up as I was this morning. 😁
Bravo Pe. Substack needs to make shutting this off a global setting so it need be done only once. I don't use AI. It never would had occurred to me you did. So this human reads you as 100% human. Checking grammar is no different than editing tools.
Ah, this explains dancers instead of clubgoers / partygoers. 🤣😂😆🤪
Pe, wonder if you've seen this about Pangram - https://post.substack.com/p/an-authors-2-million-book-deal-was
It seems Pangram is wreaking havoc in the publishing industry, generating results that have led to allegations that authors have used AI when those authors insist they haven't.
Such an interesting article, Ellen. It's a perfect illustration of the Twilight Zone we're currently in. There's this black-and-white discussion raging around AI, and the one thing we know is that black-and-white discussions aren't helpful. They turn it into a Luddite debate at exactly the moment when we should be discussing how we can use this new technology to make things better, and what guidelines and boundaries we need to put in place. Just like we did with nuclear technology after WW2.
Zooming in on my own little rant, it reinforces what I said about Pangram. And, by extension, the necessity to look beyond what is happening on Substack and ask ourselves why it is happening.
They (as in the writers of the piece you shared) don't have the answers. Nobody has. But they recognise that we're living through interesting times. In many ways, publishing hasn't been disrupted that much by technology in the past. This is probably their Napster moment. And looking at industries that have already had their Napster moments can help us understand what is happening today.
If I look at the music industry, it's going through the same convulsions, amplified by the whole DSP/fair remuneration discussion. But I do see a consensus slowly building around using AI to speed up, help and democratise parts of the creative process, while rejecting AI slop.
The same goes for photography, which has been completely transformed by technological evolution over the past 20 years. But that doesn't mean there aren't still great photographers producing great work and doing well today.
What people often forget is that we're living in a world that moves fast. Standing still yourself is never a good option.
Sorry to inundate you, Pe, but this might be of interest -- https://www.substackwritersatwork.com/p/the-truth-about-substacks-new-ai
Also note that Pangram is only two years old. You might be interested in the founders' backgrounds - https://www.pangram.com/about-us
Great assessment, Pe.
What bothered me as someone who's published books is 1) conducting an AI witch hunt with a pitchfork called Pangram when who said you can't use any AI whatsoever (who made that rule?! which is what you're also saying), and 2) using a platform that has not been proved to be reliable and legitimate as a weapon for enforcing compliance.
The careers and incomes of authors are what's at stake here, as usual the artists the ones being held hostage while these 'experts' engage in long-winded debates. If their own income were on the line, they'd cut through all this nonsense and get right to the heart of the matter. Which is agreeing on those guidelines and boundaries you talked about.
Pangram has been called out in its wikipedia entry (https://en.wikipedia.org/wiki/Pangram_(AI_detector)):
"After Pangram identified three of his writers' articles as AI-written, editor James Taranto of The Wall Street Journal called the tool a "defamation machine". Two of the authors admitted they used AI for revision of some of their work, which Taranto said is "inaccurate and unfair to characterize" as AI-generated...
According to Matteo Wong of The Atlantic: "While Pangram is accumulating the power to end reputations and careers, the tool does make mistakes, perhaps to a greater extent than is currently understood. In turn, AI accusations could very quickly spiral into a witch hunt...Just like chatbots, AI-detection tools have become effective enough for widespread use, but not reliable enough to fully trust." Wong also raised concerns about the black box nature of Pangram's algorithm, similar to other neural networks; while Pangram can show what part of a text is likely to be generated by an LLM, it cannot describe why it was flagged."
It's quite problematic to use a program like this that's not tested and proved.
Thanks for all the info, Ellen. There’s one sentence on Pangram’s rather sparse website that really made me laugh out loud: “We will aid the effort to ensure that the new generation of powerful language models will be a net positive on the world.”
It depends on what “net positive” means to them, of course, and that’s something they don’t say. 😁
Good noticing! Not to mention 'on' the world like they're going to cover it rather than 'for' the world.
Sites like Medium (and, of course, LinkedIn) were overrun with slop. Both have finally taken measures to counter it, butI'm not sure they moved in time to save themselves. Substack probably saw that and saw integrating Pangram as a preventative strike. If your business model is built on quality writing- and people paying for the same- having a platform flooded with shit writing represents an existential threat. My assumption is that it was meant more to prevent AI slop from showing up in the first place than to have all of us run litmus tests on each other.
You nailed it further upthread when you said it will disappear as soon as the financial incentive to create it disappears. The sooner that happens, the better. If having Pangram here accelerates that fall, then I'm okay with it.
I share your opinion on the sooner AI slop disappears, the better, Kevin. I had to look into the Medium and LinkedIn solutions because I didn’t know what their approach was. The latter we can probably classify as window dressing, I think. But then again, LinkedIn has always felt a bit like Facebook with some different accents.
The Medium approach I found much more interesting, because they make a clear distinction between AI slop and AI assistance. The latter isn’t classified as AI-generated content. I’m not on the platform, so I can’t judge whether it actually works or not, but the basic principle seems to me to be much better than what Substack is doing.
On paper, Medium's strategy might look better; the problem is they waited far too long to take any action. They also continue to waffle, condemning AI slop one minute and offering premium subscribers access to an AI-based writing tool the next.
Thanks for this Pé! I’m no fan of AI in the creative space, meaning for visual, audio or textual content created solely or primarily through AI. But there’s clearly a place for some usage of AI tools, particularly for non-native English speakers or for people trying to improve their grammar with things like Grammarly. Or even for people creating headers or other images with AI that they want to include in their posts.
I don’t mind the idea of an AI detection tool — I certainly wish all music streaming sites let me opt out of AI music — but I do need to believe I can trust the tools and that there’s some level of consistency. Your research tells me there isn’t.
Having said that, I’d like to think that people know what I write isn’t created by AI. I’ve certainly NEVER suspected any of your work was in any way AI-produced (nor for any other music writers I follow here). But if someone feels like they want to check my work for AI, I’m not sure I care. Feel free to believe my work is written by me. Or not. And act accordingly. It doesn’t really bother me.
My bigger concern is Pangram is potentially ingesting my work. It’s not a concern from a creative perspective because I’m not the world’s next great music writer 😆. But I don’t want anything I’m writing to be training the next generation of AI in any way, shape or form. That would be the reason I might turn off permissions for Pangram scanning of my work.
Certainly lots to think about on this front.
All valid points Mark. As for turning off Pangram, you touch on the central issue there: it isn’t possible. Your work is ingested, whether you like it or not. The only thing we’re allowed to do is turn off the evaluation tool for readers. But that doesn’t mean the evaluation itself isn’t happening. We have absolutely no control over that, and Substack hasn’t asked for our permission.
Which begs the question: why?
As a historian, I’m trained to look beyond the source, to ask questions about what it is telling me, but also about what it isn’t telling me. And that’s exactly what I find myself doing here. But anyway, sadly there’s very little we can do about it.
As for AI chatbots themselves, when used correctly they can be tremendous learning tools. My English was already pretty good when I started writing, but the way I work with a chatbot is that it corrects me and explains why it has made those corrections. It’s incredible how much that has helped me deepen my knowledge of English grammar and expand my vocabulary.
I also don’t think the kind of unlimited access we currently have to these chatbots will remain this way forever. It might end sooner than we think because the economic model behind it simply doesn’t make much sense. So, for once in my life, I’ve used the time available to me extremely well 😁 and learned whatever I can from it. And that learning process has become part of the fun of creating for me.
The reason I’m saying all this is to illustrate my central point: AI is not a black & white discussion. Yet the way Substack, and other platforms, are handling it is turning it into exactly that kind of discussion.
Excellent post, PeDupre. Substack should say thank you for using available tools to make your writing more accessible and informative for your readers. If they don't say it, I will.
Thanks RATW! Much appreciated!
I’m with you completely, Pé. I think the pangram is completely unnecessary and detracts from time spent reading and enjoying our writings, even if they were written 47 percent by Martian. It never would have occurred to me to wonder about your AI usage. Your stories and deep dives are so compelling — that’s what stands out.
It’s intriguing, your testing with the images. Were the images AI created? It does seem like the percentages are random. Your results are proof that this is a garbage tool designed to….what? Appease a disgruntled populace? Or maybe, as you say, it’s a way to train AI on our stories (and sell the content).
Lots of food for thought. Thanks for this piece!
The percentages are indeed random, Steve. I did a lot of testing over the past few weeks because I really wanted to understand how it worked. The images weren’t AI-generated either. They were images I found on the net, like a photograph of Montreal in the fifties or a postcard from Expo 67. So it’s completely incomprehensible to me why Pangram reacted the way it did.
That said, the moment I decided to block any further “evaluations” actually felt quite liberating. We should probably all do it, if only to make Substack understand that this isn’t working and that they need to change tactics.
Not that I’m convinced they would, because, as I said in my reply to Howard, I’m beginning to strongly suspect that the real agenda here may not be quite what they want us to believe it is.
Excellent points!! Ai is here to stay! For non native speakers it is a great help, more even if one is dyslexic ... like me). Perhaps our challenge is to write content that is so unique, that it reveals our singular intelectual contributions.
I find that it is essential, to command the machine "do not add adjetives" "do not editorialize". Then carefully revise that the text in your non-native language really shows your personality, you insight into the topics ... Just my 0.02 Best José
I constantly change the prompts I’m using, José. That’s one of the things I find so useful about chatbots: it’s a learning curve. And although my English was already quite good when I started, I’ve learned a lot since I began working with ChatGPT and Claude. I find them incredibly useful tools for deepening your knowledge of a language.
I even use them now to improve my other languages, including my French. That’s exactly what I meant when I said this isn’t a black & white discussion, so let’s stop turning it into one.
Anyway, thanks for your reaction. Really appreciate it!
That's so weird, how you get a different number each time
Yep, it is, and it’s been a constant factor these past few weeks, Jonathan. It simply doesn’t work properly. And on top of that, I’m not even convinced we need it. The AI slop will eventually be rooted out once the economic incentive to create it disappears.
Which begs the question: why introduce it in the first place?
I can’t help wondering if we’re looking at this the wrong way. What if it isn’t really about detection, but about training? Substack can’t simply use the mountain of valuable content sitting on its platform directly, but could something like this provide an indirect route to doing so? It would certainly make the decision to launch it without properly announcing it or asking writers for permission look rather different.
I hope I’m wrong, but I can’t rule it out. And if that were the case, it would be beyond the pale, if you ask me.
Well, I can't speak to Substack's corporate motivations. It sure seems though that they want to gesture toward supporting actual human authors by creating a "checkmark," except that the tool is too blunt for the purpose.
But really, there is a category difference between videos of animals wearing silly hats and mixes of disco songs from a half century ago. If you're still fascinated by fabricated videos of animals wearing silly hats (and over the age of 13), there's no hope for you. Happily, people's perceptions of disco mixes, which used to be on the level of animal flicks, have been changing, for the better. I hope you can keep making new episodes of The Twelve Inch and keep informing your readers (and listeners) about great music of the past. I don't think it's something that we need AI to appreciate.
Haha... don’t worry, Jonathan. I’m surely not going to stop writing & mixing. It’s far too much fun and has become a valued part of my weekly routine.
And Substack is what it is. I knew upfront what I was getting into. Which doesn’t mean we shouldn’t call out the things that need to be called out!