Having read a few drafts of this essay, version 7.2 is proof of the thesis. Lots of darlings killed, but this is far and away the best version of the essay.
This insight will stick with me: "Once you grasp the nature and magnitude of an original sin, the only way forward is destruction. There is no degree of polish or patchwork that can fix a flawed foundation."
I love this—it’s worthy of the Delphic Oracle: “If you believe in the spirit of iteration, then design is 50% destruction.”
Hemingway’s first wife accidentally left a satchel with the manuscripts of all his as-yet unpublished short stories on a train in France. They were never recovered.
So Hemingway had to re-write them all from scratch, putting the material through the imaginative crucible you describe and initiating the alchemical dance of thesis and form, essence and appearance, substance and accident.
That’s how Hemingway’s art reached greatness.
The prospects for Essay Architecture are just as dazzling!
I loved the Fountainhead. Loved Roark's single-minded obsessiveness. Though Ayn Rand went way overboard with the court monologue, I forgave her anyway, since she bought me Roark and that guy in the bar he met up with from time to time who, as I recall, used to call HR red.
I'm writing from decades old memory here, but if I recall correctly, the manipulative Elworth Toohey personifies Woke culture in hauntingly relevant terms.
To your point here, Michael, I smile. "At the core of creative destruction is a simple idea: be thesis-driven, not form-driven. This means you’re not tied to any one particular form; you burn through them to find one that embodies your thesis."
About two weeks ago I unpublished every post I made here with the exception of my landing page, because the time had come to repackage my own thesis in a different domain, now that Substack has served its primary purpose.
I had not anticipated how strangely exhilarating it is to do that, after I posted about 15 lengthy essays only to unpublish them for my own "thesis-driven" purposes.
i've never heard of anyone getting an LLM-system to give essentially the same numeric score to a qualitative thing like an essay across multiple attempts...
Congrats with shipping! I can't wait to see both going live. I like the distinction of being thesis-driven vs form-driven because, especially with products, it's easy to get caught up in 'I have to launch a SaaS' or 'my book needs to be 100+ pages'
You mentioned you can't actually know your thesis until the piece is formed (though imperfect) - is a 'thesis' exactly the same as an essay thesis in this piece? I'm seeing a lot of parallelism with wiki and tool while reading this - but I wonder if it's a pure 1:1 translation
This essay has given me much to chew on. To answer your prompt at the end, there have been a few times I've started over on significant projects—feature articles, the first 3 episodes of a narrative podcast—but only after much delay and hand-wringing. In every case, I had a sense of the problem but I stalled, hemmed and hawed, and it was only when the flaws became irreconcilable that I started over. Usually, though, by that point I had clarity and the revisions went quickly. What your essay has reminded me is that this is part of the process, and perhaps I can identify such issues in the future quicker and not be as resistant to starting over. There's a lot of the sunk cost fallacy at play here.
My second impression from this post is: holy cow, you've been busy and pulled off a lot. Your software sounds fascinating. One question I have is: will it only be limited to essays? Or might it also have some utility if applied to other writing such as narrative journalism or writing that's more episodic in nature, like chapters of a narrative non-fiction book? (I suppose in the latter's case, the context container would need to be even wider)
Thank you Michael for this wonderful essay, and for sharing your work-in-progress with us as you build this tool! It’s fascinating to read the evolution of your project.
Former exec from Scale here, now working on evals in a specific domain - lemme know if you want to speedball ideas around the evals and pipelines in general
Congratulations on knowing when to not stop and abandon it. I’m impressed with how much you’ve been able to improve the written feedback, considering the limitations of LLMs on that front. I’m excited to play with it more, and to read the final ebook!
Having read a few drafts of this essay, version 7.2 is proof of the thesis. Lots of darlings killed, but this is far and away the best version of the essay.
This insight will stick with me: "Once you grasp the nature and magnitude of an original sin, the only way forward is destruction. There is no degree of polish or patchwork that can fix a flawed foundation."
I love this—it’s worthy of the Delphic Oracle: “If you believe in the spirit of iteration, then design is 50% destruction.”
Hemingway’s first wife accidentally left a satchel with the manuscripts of all his as-yet unpublished short stories on a train in France. They were never recovered.
So Hemingway had to re-write them all from scratch, putting the material through the imaginative crucible you describe and initiating the alchemical dance of thesis and form, essence and appearance, substance and accident.
That’s how Hemingway’s art reached greatness.
The prospects for Essay Architecture are just as dazzling!
I loved the Fountainhead. Loved Roark's single-minded obsessiveness. Though Ayn Rand went way overboard with the court monologue, I forgave her anyway, since she bought me Roark and that guy in the bar he met up with from time to time who, as I recall, used to call HR red.
I'm writing from decades old memory here, but if I recall correctly, the manipulative Elworth Toohey personifies Woke culture in hauntingly relevant terms.
To your point here, Michael, I smile. "At the core of creative destruction is a simple idea: be thesis-driven, not form-driven. This means you’re not tied to any one particular form; you burn through them to find one that embodies your thesis."
About two weeks ago I unpublished every post I made here with the exception of my landing page, because the time had come to repackage my own thesis in a different domain, now that Substack has served its primary purpose.
I had not anticipated how strangely exhilarating it is to do that, after I posted about 15 lengthy essays only to unpublish them for my own "thesis-driven" purposes.
The next phase of my own roll out begins here:
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5379106
i've never heard of anyone getting an LLM-system to give essentially the same numeric score to a qualitative thing like an essay across multiple attempts...
…i exist to destroy your machine…tell me when i can try…
Congrats with shipping! I can't wait to see both going live. I like the distinction of being thesis-driven vs form-driven because, especially with products, it's easy to get caught up in 'I have to launch a SaaS' or 'my book needs to be 100+ pages'
You mentioned you can't actually know your thesis until the piece is formed (though imperfect) - is a 'thesis' exactly the same as an essay thesis in this piece? I'm seeing a lot of parallelism with wiki and tool while reading this - but I wonder if it's a pure 1:1 translation
This essay has given me much to chew on. To answer your prompt at the end, there have been a few times I've started over on significant projects—feature articles, the first 3 episodes of a narrative podcast—but only after much delay and hand-wringing. In every case, I had a sense of the problem but I stalled, hemmed and hawed, and it was only when the flaws became irreconcilable that I started over. Usually, though, by that point I had clarity and the revisions went quickly. What your essay has reminded me is that this is part of the process, and perhaps I can identify such issues in the future quicker and not be as resistant to starting over. There's a lot of the sunk cost fallacy at play here.
My second impression from this post is: holy cow, you've been busy and pulled off a lot. Your software sounds fascinating. One question I have is: will it only be limited to essays? Or might it also have some utility if applied to other writing such as narrative journalism or writing that's more episodic in nature, like chapters of a narrative non-fiction book? (I suppose in the latter's case, the context container would need to be even wider)
Fascinated and following. How can I access the prompt library?
“Now that I know what I know, what could I make if I start over?” Your questions are often unparalleled.
Thank you Michael for this wonderful essay, and for sharing your work-in-progress with us as you build this tool! It’s fascinating to read the evolution of your project.
I'll admit I still have no clue what the hell you're building, but am curious to find out... 😄
Former exec from Scale here, now working on evals in a specific domain - lemme know if you want to speedball ideas around the evals and pipelines in general
Congratulations on knowing when to not stop and abandon it. I’m impressed with how much you’ve been able to improve the written feedback, considering the limitations of LLMs on that front. I’m excited to play with it more, and to read the final ebook!
Congrat, Michael. What an accomplishment. Can't wait to see the book and new tool mid Sept.
Loved this essay. What an engaging and creative way to update on your project and discuss creative destruction.
maybe my fav of ur essays so far