Is your process one that can be replicated by those who are otherwise good thinkers and writers, but didn't have a clue about how to obtain, refine, pre-prompt, and deploy their own AI research, editing, and text production system?
Thanks for the insight into your process, it’s interesting. I’ll experiment with that myself. I don’t have a problem with the process and I can temper my jealousy over how prolific you are. However, the essays would hit harder than they do if there were examples and statistics. Good nonfiction writing is not just an author with ideas and a blank page - it’s also a “pile” of references. And I know that the models are weak in that area, I know that one can lie with tearjerker examples or phony statistics, I know that the velocity can’t be the same with fact-checking. I don’t have a solution, I’m just saying that instead of a “wow” my reaction is often “huh, okay, maybe”. The takes and sometimes even the subject matter are novel, but then suspicions arise about the extent of generalizations, the prevalence of the problem, etc. To give an example (ha!) one of the Epstein pieces argued that naive nouveau riche were drawn to his setup while old money folks had this long figured out - but then British and Norwegian royalty embroiled in the scandal is about as old as old money gets. So I’m hoping your approach can evolve to add some concreteness to the good but abstract arguments. Cheers!
I use Claude as a co-author on about half of my Substack posts. He's gotten so good at imitating my voice that I forget which passages are mine and which are his. For me he is simply a word-generator and not as a source of critique, though I probably try that too. Every post that had a significant contribution from AI ends with an Acknowledgements section that simply says "Thanks, Claude!" along with a link to our conversation.
I feel comfortable doing this because I'm already a good writer. I taught myself to become one back when staring at a blank page was the only way to do it, and now I know how to feel cadences, make arguments and tell jokes. I am confident that the machine is amplifying, not usurping, my skill.
Would it be a problem if I wasn't a good writer but just played one on Substack? From the reader's perspective no. Words are words. It doesn't matter how they got there. From my perspective, though, there is something lost. My process doesn't matter to a reader, but it does matter to me. I write to discover what I actually think. Ideas that are self-evident in my head melt away like mist the moment I try to put them down in words. That's step one. Writing isn't just about communicating thoughts—it's about generating them.
This process of generation is separate from the harsh external critique you describe (in the end it's still just me, alone in a room, confirming my biases) but is necessary in its own way. I'm glad I learned how to do it, and feel sorry for someone who doesn't because a machine makes the wordsmithing part too easy to bother with the rest.
"I find you ideas intriguing and I wish to subscribe to your newsletter"
More to the point, when you build your local models I will be very interested in any technical details you reveal because I'd be interested in cloning the concept
"You shouldn't use a forklift until you first can clean and jerk your own bodyweight, and shouldn't use a bike and then a car until you can run a marathon."
More directly, re writing, the cognitive bottleneck is almost never mechanical production. It's knowing what question to ask, recognizing a good answer, and having the judgment to edit. AI doesn't short-circuit any of those.
Is your process one that can be replicated by those who are otherwise good thinkers and writers, but didn't have a clue about how to obtain, refine, pre-prompt, and deploy their own AI research, editing, and text production system?
Thanks for the insight into your process, it’s interesting. I’ll experiment with that myself. I don’t have a problem with the process and I can temper my jealousy over how prolific you are. However, the essays would hit harder than they do if there were examples and statistics. Good nonfiction writing is not just an author with ideas and a blank page - it’s also a “pile” of references. And I know that the models are weak in that area, I know that one can lie with tearjerker examples or phony statistics, I know that the velocity can’t be the same with fact-checking. I don’t have a solution, I’m just saying that instead of a “wow” my reaction is often “huh, okay, maybe”. The takes and sometimes even the subject matter are novel, but then suspicions arise about the extent of generalizations, the prevalence of the problem, etc. To give an example (ha!) one of the Epstein pieces argued that naive nouveau riche were drawn to his setup while old money folks had this long figured out - but then British and Norwegian royalty embroiled in the scandal is about as old as old money gets. So I’m hoping your approach can evolve to add some concreteness to the good but abstract arguments. Cheers!
I actually did argue with one of the AIs about Andrews involvement, and that got factored into the resulting essay.
I use Claude as a co-author on about half of my Substack posts. He's gotten so good at imitating my voice that I forget which passages are mine and which are his. For me he is simply a word-generator and not as a source of critique, though I probably try that too. Every post that had a significant contribution from AI ends with an Acknowledgements section that simply says "Thanks, Claude!" along with a link to our conversation.
I feel comfortable doing this because I'm already a good writer. I taught myself to become one back when staring at a blank page was the only way to do it, and now I know how to feel cadences, make arguments and tell jokes. I am confident that the machine is amplifying, not usurping, my skill.
Would it be a problem if I wasn't a good writer but just played one on Substack? From the reader's perspective no. Words are words. It doesn't matter how they got there. From my perspective, though, there is something lost. My process doesn't matter to a reader, but it does matter to me. I write to discover what I actually think. Ideas that are self-evident in my head melt away like mist the moment I try to put them down in words. That's step one. Writing isn't just about communicating thoughts—it's about generating them.
This process of generation is separate from the harsh external critique you describe (in the end it's still just me, alone in a room, confirming my biases) but is necessary in its own way. I'm glad I learned how to do it, and feel sorry for someone who doesn't because a machine makes the wordsmithing part too easy to bother with the rest.
Thanks for telling us who to block. I appreciate your honesty.
"I find you ideas intriguing and I wish to subscribe to your newsletter"
More to the point, when you build your local models I will be very interested in any technical details you reveal because I'd be interested in cloning the concept
This is the sort of process that should be taught in school. But it’s also the sort of thing you shouldn’t do until you know how to do it yourself.
"You shouldn't use a forklift until you first can clean and jerk your own bodyweight, and shouldn't use a bike and then a car until you can run a marathon."
More directly, re writing, the cognitive bottleneck is almost never mechanical production. It's knowing what question to ask, recognizing a good answer, and having the judgment to edit. AI doesn't short-circuit any of those.
By doing it yourself, I mean within your own cognition. Then you can learn how to use the AI tool.
Oh, I do it to myself all the time.
I hope this is not a double entendre ...