Thanks. I think every writer should give it a try now, just to see how they feel about it. I may reject this way of working but for now I feel like it's unblocked my creativity.
I'll start with my own, umm, startling confession: I've yet to read beyond about the photo of "I Really Like Slop," yet I've got the nerve to write a comment!
(I do intend to finish the article, which I was really enjoying, partly because the writing is very good but also because this topic has been on my mind to write about, as well.)
Anyway, I just wanted to ask you: Who wrote the first two paragraphs?
I ask because that "you will do" ending of the second sentence really broke my brain. ("Yet if you don’t let them, and you use them like obedient assistants or training wheels on a bike, maintaining the intention to learn yourself, you will do.")
I ran both paragraphs through Grammarly and it didn't flag that, so then I asked ChatGPT if the grammar was okay on these two opening sentences. I won't repeat what it wrote about sentence #1 (unless you want me to), but it also flagged that "you will do ending."
Here, I'll post what it said, and like me, it's curious what your response will be.
Over the past few weeks, I've been publishing the journal I wrote in jail and using Grammarly and sometimes ChatGPT to edit, but the former certainly misses some big mistakes sometimes. Anyway, very thought-provoking article---what I've read so far has been, at least---I'll finish it later! Here was ChatGPTs response---and again, mostly curious who wrote that, or, if there's some way you can explain how that ending makes sense! you do"
I learned to speak Spanish partially by using Google Translate, and to find my way around Barcelona using CityMapper, both tools which will, if you let them, entirely hijack your brain to the point where you have outsourced everything to the machines.
Yet if you don't let them, and you use them like obedient assistants or training wheels on a bike, maintaining the intention to learn yourself, you will do.
---
CHAT GPT response
Sentence 2:
Original:
Yet if you don’t let them, and you use them like obedient assistants or training wheels on a bike, maintaining the intention to learn yourself, you will do.
Issues:
1. The phrase “you will do” is grammatically off. It’s likely a truncation of “you will do fine” or “you will learn,” but as it stands, it’s unidiomatic and confusing.
2. Minor style: “maintaining the intention to learn yourself” could be clearer as “maintaining your own intention to learn” or “continuing to learn yourself.”
Suggestion:
Yet if you don’t let them, and instead use them like obedient assistants or training wheels on a bike—while maintaining your own intention to learn—you’ll do just fine.
I wrote those Bryan! No AI, so that's why it's a bit off. But also that's a UK way of putting it. 'If you maintain the intention to learn, you will do (learn)' clarifies it a bit. I did think about changing it but it makes sense to me. I want to keep some little human idiosyncrasies if possible. Although... my next piece is edited to within an inch of its life, by both me and AI, just as an experiment.
And absolutely keep the idiosyncrasies. I spent a few weeks editing a memoir written by one of your countrymen (and fellow Rebel Wisdom “alum”), Alasdair Lord, and one thing I loved about it was learning all of the fascinating sayings you guys have, so this “you will do” is another new one to me! Hadn’t even considered it might be a regional term.
But yes, those regional elements are one of the ways we’ll keep our human voice, as is, of course any kind of writing from direct personal experience.
PS Thanks for the education on the painters of the past. I mean, it doesn’t surprise me all that much. After all, probably my favorite novel of the 21st century is “Cloud Atlas” by David Mitchell and one of those sub stories is all about an artist with fame basically stealing the ideas that his young “apprentice” comes up with and claims them as his own. (That’s obvious a more egregious example than what you’re sharing, though!).
Ultimately, my solution to this conundrum is to just be honest as best as you can. For example, a few weeks ago I took an essay I’d written about the tradition of cherry blossoms here in Japan for a native-English speaking audience and had ChatGPT cut the word count in half and write it for non-native English speakers at a certain level.
I then told my students that’s what the essay was. And then, a few of the higher level ones asked for the original so I sent them that link.
In a funny twist to this tale that I think a Rebel Wisdom grad/Daniel Pinchbeck fan could appreciate, the next week I started the class with questions and one student asked me about how in the essay I’d mentioned taking MDMA—-”how many times?” I said, “Well, I wasn’t counting…” LOL.
Anyway, again, this was a great piece! It’s going to be quite the journey forward, eh?
Thanks Bryan! Yes I think as long as one is transparent, it’s ok. People can judge whether they find the style too clinical or not. Half of my writing is just digging into things I am curious about, and I think it makes sense to use AI for that. The other half is more about my own life (not that I’ve published much of that yet, endless drafts abound of course), and a bit of poetry, and I won’t be using AI much, if at all, for those.
That’s a good idea to create simpler versions using AI for people learning English. I will suggest that to my sister who is an English teacher here in Spain.
I have an half-written essay about this same topic waiting for me in my Substack drafts :p
As a non-native English writer, I sit with the question of how to engage with LLMs in a way that feels integral to me a lot. It certainly gives me an advantage (for example, here I originally wrote "advance", my brain probably picked such a word because it translated directly from portuguese "avanço", that's the kind of thing that I have to keep questioning... I had a verbal intuition about which word to pick but it was coming from the wrong language!).
Anyway, I like to use it as an extension of my brain. I write my sentences, it helps me make them better, reflect on word choice, grammar, intended meaning... What I’ve also been observing is that it can backfire: I become too obsessive, my confidence begins to shake. Suddenly I'm using it to double-check pretty much anything I put on the page. That's when I stop, remind myself to put the tool at my service, not become enslaved by it. Working with it so closely, I believe, has made me a better writer: I've been training how to question each word choice, each sentence I generate. But again, confidence is perhaps as important as verbal sensitivity, so I use it with caution. My rule now is: everything you write that you actually care about--books, essays, emails--should come from you and you only. You're allowed to use the AI 1) in an early-writing stage, only when you don't know a particular term (like Anne Lamott describes in Bird By Bird, when she went on a mission to discover how to refer to the tip of a shoelace); 2) when you’re done writing, re-writing, editing and re-editing and you can't make the piece any better out of your own brain. Then I usually copy all the text and ask the AI for a reflection: any improvements I could make, from grammar to vocabulary? And, once again, I'm the one in control, making the final decision of which sections or sentences to alter. I find that otherwise it can easily destroy one's voice. And even as it improves (I notice that by now it can copy voice quite well, in fact even take it a step further), I don't want it to do the work for me because that defeats the purpose (and fun!) of creativity for me...
As I was doing some research for the essay still in my drafts, I landed on (another mistake, I wrote "into" here, prepositions are the worst for me :p) some news about certifications being created, ensuring publications are "Human Made", much the same way there are several certifications of quality control or eco products. Even "human made" writing though, seems to allow a minimum level of LLM editing. So yeah, I believe there's no way back!
That said, another thing on my mind is the amount of resources and energy needed to power all this data generation: storage, water to cool down the infrastructure... that's quite a conundrum--another one!
Ah, and the correct term to refer to the tip of the shoelace is "aglet"! What took Anne Lamott hours of research we can now discover in a couple of seconds... like many tech innovations throughout history, as you point out in the essay...
Chat GPT said:
The proper term is "aglet."
Anne Lamott famously describes looking for that word in Bird by Bird, so referencing it exactly might strengthen your point and give a little wink to readers who know the book:
"...when she went on a mission to discover how to refer to the tip of a shoelace—an aglet."
Thanks Carlota for the reflections! Yes, what I'm mainly using AI for is to take drafts on which I feel stuck and can't progress further on my own, and 'unstick' them. It's so good for that.
The current Library of Babel piece was a bit different in that it's an experiment on writing with an LLM. I gave it a load of ideas and it turned them into an essay, then I re-wrote large parts of it and added others. It's like a back and forth with another writer.
I have now created a system prompt which includes some of my own essays, so that it can 'write like me'. It's actually really uncanny how it takes some parts of my style and reproduces them. In fact it's like looking in a mirror and seeing one's flaws. My style has been too ponderous, too eager to cover every eventuality. (I am not going to let it write my essays for me, obviously, because it's not me: it's similar but not as good).
As an antidote to that, I created another system prompt which is a brutal critic and editor. It points out all the unnecessary verbiage and flaws. I don't always take its advice though because neither do I want a 100% 'optimised' piece of writing. If you check out the new piece you will see a change in style, I think. Does it sound 'like me'? Maybe not completely, but 'I' am a process, not an object. I always remember Carlos Santana talking about how he practices guitar: constantly throwing away the cliches that have crept into his playing (and you can see how his style changed over the years as a result of that).
For me personally, all this has been really illuminating. I think it will make me a better writer, just as CityMapper made me better at navigating Barcelona without it. But of course, I can see that to people with no integrity, churning out 'content' in someone else's style is now totally trivial.
Yes there is also the point that kids may now grow up unable to write. The invention of the camera also probably severely cut down the number of people who are good at painting, however it certainly didn't reduce it to zero. And I don't think LLMs will completely take over writing either.
What we need is to hone our sense of poetry, of intuition, our 'right brain' capabilities, and we will immediately spot who or what is churning out 'AI slop' and polluting the information commons, and who is creating more clarity (even if they use LLMs to help them do it). This is part of the Renaissance Of Discernment that I keep talking about. I do think that those 'right brain' capacities are going to become more and more important, because that is the part that AI doesn't have, at all.
I also worry about the use of energy and water for AI; it's probably way worse than we can imagine already. As much as I can I use LLMs on my own computer (and I get my electricity from renewable sources via a cooperative). For certain things they are not good enough, but for others they are just as good as ChatGPT or Claude or whatever.
And I learned something today: I didn't know the tip of a shoelace was called an 'aglet'! Thanks :p
I'm a total neophyte with using AI to write. I'm also neck deep in drafts left unpublished. I have a couple questions for you both, if you don't mind giving some insight.
1. Which model has helped you the most? Which one helps you preserve your "voice"? 2. Have you tried Notebook LLM? What do you think of it?
1. Claude Sonnet 3.7 Thinking model is the best I've tried. But it's also the most expensive. I am experimenting with system prompts to preserve voice because in general the output will be generic without specific instructions, and that's for all models.
2. Yes NotebookLM is amazing. So far I'm mostly using it for research and learning, like dump 20 documents, YT videos and websites into it, and ask it to summarise, read the summary, ask it questions on things it missed or need more depth. I'm not using it for writing because the output is quite generic and you can't change it as you can with other methods (I use Obsidian with the SystemSculpt AI plugin, which allows you to tinker with system prompts and different models, if you have accounts with OpenAI, OpenRouter, Grocq, Anthropic etc.)
But what NotebookLM is great for is taking, say, a folder of old notes or documents about a specific topic which never got resolved, or a course you took, or some old emails, and ask it to synthesise everything into something coherent, like an essay you can read, or a podcast you can listen to. That's really good for learning and digesting, which is the subject of a series of articles I'm writing at the moment.
That’s where I am too. It’s helped me put my ideas into structure I’m not generally good at. It also use it for feedback, and to flesh out articles after I have a skeleton. It’s been very useful. I’m also finding myself noticing AI writing now, and becoming more on the side of using it lightly. Thanks for being honest and transparent. I think we are all grappling with this and it will be interesting to watch it unfold.
Props for diving into this question, and revealing your process around it. I think it's important. I'm grappling with it too.
Thanks. I think every writer should give it a try now, just to see how they feel about it. I may reject this way of working but for now I feel like it's unblocked my creativity.
Hi guy,
I'll start with my own, umm, startling confession: I've yet to read beyond about the photo of "I Really Like Slop," yet I've got the nerve to write a comment!
(I do intend to finish the article, which I was really enjoying, partly because the writing is very good but also because this topic has been on my mind to write about, as well.)
Anyway, I just wanted to ask you: Who wrote the first two paragraphs?
I ask because that "you will do" ending of the second sentence really broke my brain. ("Yet if you don’t let them, and you use them like obedient assistants or training wheels on a bike, maintaining the intention to learn yourself, you will do.")
I ran both paragraphs through Grammarly and it didn't flag that, so then I asked ChatGPT if the grammar was okay on these two opening sentences. I won't repeat what it wrote about sentence #1 (unless you want me to), but it also flagged that "you will do ending."
Here, I'll post what it said, and like me, it's curious what your response will be.
Over the past few weeks, I've been publishing the journal I wrote in jail and using Grammarly and sometimes ChatGPT to edit, but the former certainly misses some big mistakes sometimes. Anyway, very thought-provoking article---what I've read so far has been, at least---I'll finish it later! Here was ChatGPTs response---and again, mostly curious who wrote that, or, if there's some way you can explain how that ending makes sense! you do"
I learned to speak Spanish partially by using Google Translate, and to find my way around Barcelona using CityMapper, both tools which will, if you let them, entirely hijack your brain to the point where you have outsourced everything to the machines.
Yet if you don't let them, and you use them like obedient assistants or training wheels on a bike, maintaining the intention to learn yourself, you will do.
---
CHAT GPT response
Sentence 2:
Original:
Yet if you don’t let them, and you use them like obedient assistants or training wheels on a bike, maintaining the intention to learn yourself, you will do.
Issues:
1. The phrase “you will do” is grammatically off. It’s likely a truncation of “you will do fine” or “you will learn,” but as it stands, it’s unidiomatic and confusing.
2. Minor style: “maintaining the intention to learn yourself” could be clearer as “maintaining your own intention to learn” or “continuing to learn yourself.”
Suggestion:
Yet if you don’t let them, and instead use them like obedient assistants or training wheels on a bike—while maintaining your own intention to learn—you’ll do just fine.
I wrote those Bryan! No AI, so that's why it's a bit off. But also that's a UK way of putting it. 'If you maintain the intention to learn, you will do (learn)' clarifies it a bit. I did think about changing it but it makes sense to me. I want to keep some little human idiosyncrasies if possible. Although... my next piece is edited to within an inch of its life, by both me and AI, just as an experiment.
Thanks for answering, Guy!
And absolutely keep the idiosyncrasies. I spent a few weeks editing a memoir written by one of your countrymen (and fellow Rebel Wisdom “alum”), Alasdair Lord, and one thing I loved about it was learning all of the fascinating sayings you guys have, so this “you will do” is another new one to me! Hadn’t even considered it might be a regional term.
But yes, those regional elements are one of the ways we’ll keep our human voice, as is, of course any kind of writing from direct personal experience.
PS Thanks for the education on the painters of the past. I mean, it doesn’t surprise me all that much. After all, probably my favorite novel of the 21st century is “Cloud Atlas” by David Mitchell and one of those sub stories is all about an artist with fame basically stealing the ideas that his young “apprentice” comes up with and claims them as his own. (That’s obvious a more egregious example than what you’re sharing, though!).
Ultimately, my solution to this conundrum is to just be honest as best as you can. For example, a few weeks ago I took an essay I’d written about the tradition of cherry blossoms here in Japan for a native-English speaking audience and had ChatGPT cut the word count in half and write it for non-native English speakers at a certain level.
I then told my students that’s what the essay was. And then, a few of the higher level ones asked for the original so I sent them that link.
In a funny twist to this tale that I think a Rebel Wisdom grad/Daniel Pinchbeck fan could appreciate, the next week I started the class with questions and one student asked me about how in the essay I’d mentioned taking MDMA—-”how many times?” I said, “Well, I wasn’t counting…” LOL.
Anyway, again, this was a great piece! It’s going to be quite the journey forward, eh?
Thanks Bryan! Yes I think as long as one is transparent, it’s ok. People can judge whether they find the style too clinical or not. Half of my writing is just digging into things I am curious about, and I think it makes sense to use AI for that. The other half is more about my own life (not that I’ve published much of that yet, endless drafts abound of course), and a bit of poetry, and I won’t be using AI much, if at all, for those.
That’s a good idea to create simpler versions using AI for people learning English. I will suggest that to my sister who is an English teacher here in Spain.
Thank you for touching me without knowing 🖊️
Great piece! It was fun to read.
I have an half-written essay about this same topic waiting for me in my Substack drafts :p
As a non-native English writer, I sit with the question of how to engage with LLMs in a way that feels integral to me a lot. It certainly gives me an advantage (for example, here I originally wrote "advance", my brain probably picked such a word because it translated directly from portuguese "avanço", that's the kind of thing that I have to keep questioning... I had a verbal intuition about which word to pick but it was coming from the wrong language!).
Anyway, I like to use it as an extension of my brain. I write my sentences, it helps me make them better, reflect on word choice, grammar, intended meaning... What I’ve also been observing is that it can backfire: I become too obsessive, my confidence begins to shake. Suddenly I'm using it to double-check pretty much anything I put on the page. That's when I stop, remind myself to put the tool at my service, not become enslaved by it. Working with it so closely, I believe, has made me a better writer: I've been training how to question each word choice, each sentence I generate. But again, confidence is perhaps as important as verbal sensitivity, so I use it with caution. My rule now is: everything you write that you actually care about--books, essays, emails--should come from you and you only. You're allowed to use the AI 1) in an early-writing stage, only when you don't know a particular term (like Anne Lamott describes in Bird By Bird, when she went on a mission to discover how to refer to the tip of a shoelace); 2) when you’re done writing, re-writing, editing and re-editing and you can't make the piece any better out of your own brain. Then I usually copy all the text and ask the AI for a reflection: any improvements I could make, from grammar to vocabulary? And, once again, I'm the one in control, making the final decision of which sections or sentences to alter. I find that otherwise it can easily destroy one's voice. And even as it improves (I notice that by now it can copy voice quite well, in fact even take it a step further), I don't want it to do the work for me because that defeats the purpose (and fun!) of creativity for me...
As I was doing some research for the essay still in my drafts, I landed on (another mistake, I wrote "into" here, prepositions are the worst for me :p) some news about certifications being created, ensuring publications are "Human Made", much the same way there are several certifications of quality control or eco products. Even "human made" writing though, seems to allow a minimum level of LLM editing. So yeah, I believe there's no way back!
That said, another thing on my mind is the amount of resources and energy needed to power all this data generation: storage, water to cool down the infrastructure... that's quite a conundrum--another one!
Ah, and the correct term to refer to the tip of the shoelace is "aglet"! What took Anne Lamott hours of research we can now discover in a couple of seconds... like many tech innovations throughout history, as you point out in the essay...
Chat GPT said:
The proper term is "aglet."
Anne Lamott famously describes looking for that word in Bird by Bird, so referencing it exactly might strengthen your point and give a little wink to readers who know the book:
"...when she went on a mission to discover how to refer to the tip of a shoelace—an aglet."
Thanks Carlota for the reflections! Yes, what I'm mainly using AI for is to take drafts on which I feel stuck and can't progress further on my own, and 'unstick' them. It's so good for that.
The current Library of Babel piece was a bit different in that it's an experiment on writing with an LLM. I gave it a load of ideas and it turned them into an essay, then I re-wrote large parts of it and added others. It's like a back and forth with another writer.
I have now created a system prompt which includes some of my own essays, so that it can 'write like me'. It's actually really uncanny how it takes some parts of my style and reproduces them. In fact it's like looking in a mirror and seeing one's flaws. My style has been too ponderous, too eager to cover every eventuality. (I am not going to let it write my essays for me, obviously, because it's not me: it's similar but not as good).
As an antidote to that, I created another system prompt which is a brutal critic and editor. It points out all the unnecessary verbiage and flaws. I don't always take its advice though because neither do I want a 100% 'optimised' piece of writing. If you check out the new piece you will see a change in style, I think. Does it sound 'like me'? Maybe not completely, but 'I' am a process, not an object. I always remember Carlos Santana talking about how he practices guitar: constantly throwing away the cliches that have crept into his playing (and you can see how his style changed over the years as a result of that).
For me personally, all this has been really illuminating. I think it will make me a better writer, just as CityMapper made me better at navigating Barcelona without it. But of course, I can see that to people with no integrity, churning out 'content' in someone else's style is now totally trivial.
Yes there is also the point that kids may now grow up unable to write. The invention of the camera also probably severely cut down the number of people who are good at painting, however it certainly didn't reduce it to zero. And I don't think LLMs will completely take over writing either.
What we need is to hone our sense of poetry, of intuition, our 'right brain' capabilities, and we will immediately spot who or what is churning out 'AI slop' and polluting the information commons, and who is creating more clarity (even if they use LLMs to help them do it). This is part of the Renaissance Of Discernment that I keep talking about. I do think that those 'right brain' capacities are going to become more and more important, because that is the part that AI doesn't have, at all.
I also worry about the use of energy and water for AI; it's probably way worse than we can imagine already. As much as I can I use LLMs on my own computer (and I get my electricity from renewable sources via a cooperative). For certain things they are not good enough, but for others they are just as good as ChatGPT or Claude or whatever.
And I learned something today: I didn't know the tip of a shoelace was called an 'aglet'! Thanks :p
I'm a total neophyte with using AI to write. I'm also neck deep in drafts left unpublished. I have a couple questions for you both, if you don't mind giving some insight.
1. Which model has helped you the most? Which one helps you preserve your "voice"? 2. Have you tried Notebook LLM? What do you think of it?
1. Claude Sonnet 3.7 Thinking model is the best I've tried. But it's also the most expensive. I am experimenting with system prompts to preserve voice because in general the output will be generic without specific instructions, and that's for all models.
2. Yes NotebookLM is amazing. So far I'm mostly using it for research and learning, like dump 20 documents, YT videos and websites into it, and ask it to summarise, read the summary, ask it questions on things it missed or need more depth. I'm not using it for writing because the output is quite generic and you can't change it as you can with other methods (I use Obsidian with the SystemSculpt AI plugin, which allows you to tinker with system prompts and different models, if you have accounts with OpenAI, OpenRouter, Grocq, Anthropic etc.)
But what NotebookLM is great for is taking, say, a folder of old notes or documents about a specific topic which never got resolved, or a course you took, or some old emails, and ask it to synthesise everything into something coherent, like an essay you can read, or a podcast you can listen to. That's really good for learning and digesting, which is the subject of a series of articles I'm writing at the moment.
That’s where I am too. It’s helped me put my ideas into structure I’m not generally good at. It also use it for feedback, and to flesh out articles after I have a skeleton. It’s been very useful. I’m also finding myself noticing AI writing now, and becoming more on the side of using it lightly. Thanks for being honest and transparent. I think we are all grappling with this and it will be interesting to watch it unfold.
Outstanding. ❤️
great piece, Guy!
Stew it up!
AI is here, how will we use it well?
If it can get us more in the flow of our own lives, that would be a good thing.
Thanks for playing the game to help us figure it out.
Interested to see where this takes your own Branchy Brian.
Bryan Winchell had distilled this article into pure essence.