now I sort of want there to be some kind of ai detective series where the detective could always solve the case, but ends up getting distracted by some small clue and doing down really long rabbit holes instead.
I use my tried-and-true technique. When I can't find a solution, I switch to reading books unrelated to my field (software development, not coding itself). For example, books on biology by Richard Dawkins. And as I read, a solution appears. The idea is simple: a problem arises, then I study the problem (subconscious downloading), then I switch contexts (reading the book), and the solution appears. Of course, it doesn't always work, but generally it does.
The suggestion that less knowledge helps you have good ideas is backwards. Creativity and understanding are built on a bedrock of automated facts and procedures. They're necessary but not sufficient for good ideas, and the richer your mental library, the more surprising the connections you can make. Knowledge is the raw material of creativity.
Viewed this way: experts don't see more because they're smarter than you. Their knowledge upgrades their perceptual apparatus, and it's the apparatus that does the seeing. A sparse board doesn't force interesting connections, it simply limits you to the ones you happen to have.
As I like to say: you can't conjecture from an empty head.
Except this: "A sparse board doesn't force interesting connections, it simply limits you to the ones you happen to have."
Necessity is the mother of invention, and creativity loves constraints / limitations breed creativity. Boundaries actually fuel the creative process. Anyone who has ever done improv knows this: an empty canvas is a lot less fun/creative than a constrained scenario. Completing a partial picture is a lot more fun than starting from scratch.
Constraints prevent choice paralysis: a blank canvas offers endless choices, so it's easy to get stuck, and generic ideas are the easiest way to get unstuck. But restrictions narrow our choices and make it easier to get started, or to branch out with a wacky idea. If we can't reuse generic ideas, we're forced to find something else. It's like cooking with unusual ingredient. What am I gonna do with this?
There is a fun game to play where a group of people write a story one word at a time, taking turns. We are forced to use the words that are already there and to try and direct the story in a direction that makes sense (or not). With all of the knowledge in the world, AI would suck at this because it would always use statistical probability to add the next word, and there's no fun in that. The diversity of each person's *constrained* knowledge is what makes the game fun / what makes the story unexpected. Ironically, the game is especially fun with children, *who have the least knowledge*, because young minds don't have the same guardrails as grown ones. Sometimes, knowledge is actually limiting, because we can't look outside of the box we have created for ourselves.
The world is a multiplayer game, but genius/intellect often restricts itself to a single-player experience.
Thanks for the thoughtful reply. I don't disagree constraints are helpful, but I think perhaps for different reasons e.g. providing clarity and focus.
The simpler version of what I'm saying (to address your point) still stands I think and would be the equivalent of trying to play the story game in a language no-one speaks.
As I said: knowledge is necessary but not sufficient for good ideas. I agree other things can get in the way of this (as you point out - and I agree with) but that's a separate effect in my view.
You can't have ideas about things you don't possess knowledge of. I've never had a good idea in microbiology, or quantum physics, or chemistry. Why? Because I know naff all about them.
I agree with this, to a point, I think. If you're clueless about everything, absolutely, you probably don't have many good ideas. If you're know a lot about lots of stuff, you probably have lots of good ideas; that was, like Da Vinci's whole thing. But I think I'd add one main details:
You can get stuck. If you know a ton about whatever thing, and spend all your time thinking about that thing, I've found it better to go run into some other stuff. In a sense, that proves your point - you're basically picking up other ideas - though I do think the timing of it matters, because *knowing* about something isn't quite the same as *thinking* about it. And really, that may be a better way of making the point - it's less about what you know, and more about putting different disparate ideas at the front of your mind.
And that's how AI goes wrong, I'd say. It knows tons of stuff, and could, in theory, do a really good job of running them into each other. But it doesn't on its own, because it doesn't need to.
Yeah, I guess there are (at least) three dimensions to getting (un)stuck:
1. Acquire more relevant knowledge;
2. Acquire other knowledge;
3. Go and do something else.
All can be helpful.
AFAUI the latter just helps you enter what they used to call "diffuse mode thinking" - no idea if it's still a thing these days. I think of this as just letting your brain escape local minima and don't see it as something that depends on the amount of knowledge you have. Kids routinely get stuck on one thing for ages with very little knowledge or exhibit "one track thinking".
Yeah, that's more or less exactly what it is, I think. Which clearly isn't a new or novel thought, though it does seem like something that's hard to convince yourself to do when you're stuck. It somehow always feels better to keep banging on the thing; it almost always works way better to do anything else.
Oh yeah - many a time* I've been locked-in banging my head against a wall with a problem knowing full well that taking a walk or break would help. Did I? Did I bollocks.
This is what brainstorming is all about! Brain*storming*, not brain-sheltered-in-a-tunnel.
And it's also why diverse groups tend to come up with better ideas. Homogenous groups are all in the same tunnel already, and they don't want to go out in the rain.
I would assume there's tons of real research out there for how groups are better than individuals for these sorts of things, but my very vibe-based sense is yeah, that's a big part of it. On your own, it's so much harder to run into interesting or unexpected stuff.
I tried a version of this by having bots go read like 10 random read a bunch of random wikipedia pages or something, and then doing it's work with that stuff in its mind. It didn't quite work, because it turns out that "random" wikipedia pages were mostly historical biographies or really obscure science stuff, but it did produce some pretty weird stuff.
You write beautifully, Benn. One thing I’ve always wondered is what your writing process looks like, especially given all the intricate connections you make to communicate an idea. I loved the codenames analogy!
I didn't really say this part that explicitly, but I do think this trick can be pretty effective for making things appear to be pretty connected. When you're writing the piece in tandem with the reference you're using to fit it, it makes it a lot easier to tie things together: https://benn.substack.com/p/how-to-have-ideas#footnote-7
This resonated with the focused vs. diffuse mode framing: focus gives you the problem and constraints; diffuse mode creates the unexpected connections. The trick is not to fully switch off the problem, but to hold one clear question lightly while exposing yourself to unrelated inputs. That seems to be where the useful collisions happen.
well, I am convinced that it is WAY more than just that... there are also Slacks and other tools to act as the first gate of all of information going to our mind?
I find it difficult to engage with the author's ideas because I'm so distracted by the fact that the writing style is a blatant rip off Matt Levine. I enjoy the conversational simplicity of Money Stuff but seeing it aped in this context feels crass and kinda lame. I'd encourage the author to push harder to find their own organic voice and style, which will be better anyway in the long run.
The sassy remark was just the product of frustration and lack of restraint. All the best of luck to this author, writing well is hard.
So, I'm obviously not exactly a neutral party here, but I think this sort of thing is worth thinking about some, especially in a world where AI can kinda-sorta-but-badly engage in some form of stylistic plagiarism (though i generally find AI to be quite bad at writing of any kind, including editing, and don't use it at all for writing. It's useful for research and as a kind of automated Google on a loop, but, so far, not at all for writing.) Anyway, a few thoughts:
This is quote from a recent interview with Marlon James, who just published the "The Disappearers," and I generally agree with it. I'm a believer that every creative endeavor is mostly a mashup of prior influences, and developing that mashup *is* what it means to develop a style. It doesn't come out of the ether; it comes out of things that rhyme with whatever you read and consume.
---
"I read tons of novels when I’m working on books. I stare at people funny when they go, “I don’t want to be influenced by something.” I’m like, “You can only hope!” I absolutely want to be influenced by what I’m reading.
I can tell which parts of this book I was reading Don Quixote, and which parts I was reading Virginia Woolf, and which part I was reading my designated Toni Morrison novel for the year, which would probably be my fourth or fifth reading of that novel. To me, it’s like looking at art while painting. Yes, it’s an influence, but also you want to be surrounded with creativity and reminded of creativity. And sometimes I want to see how they do it.
I would make basically the same argument here. Over the course of writing this thing, there are posts that have very different rhythms (some that are conversational; some that are not; in both cases, including posts like this one, I would generally say that I'm much more prone a kind of lofty melodrama and some structural weirdness than Matt Levine). Part of those differences, I'm sure, come from my mood and what feels good in that moment, which comes from other things I've read and liked recently.
But personally, I would rather write these posts in the way that feels natural both to me and the moment, and let that take each post where it may, rather than try to develop a "voice" and then edit each post to fit into that. This is actually a very annoying thing about AI to me; it frequently tells me things like "here's a way you might put." And I think that feels much more artificial - to chase a contrived character, be it your own, somebody else's, or the explicit avoidance of things that feel natural to you - than to write it as it comes.
Thanks Benn, I appreciate the thoughtful response, and I apologize for the snark above. Too much negativity on the internet
You are of course correct that our writing style is ultimately some kind of imitation and mashup of what we've consumed before + life experience (kind of like how llms do it minus the life experience haha).
For whatever reason I was triggered by the style in this post, and it has been my very-non scientific perception that there has been a stylistic shift in the substack toward the latent space dominated by Matt Levine's writing (at least in my mind). Take this as you will -- light feedback from a random on the internet.
Appreciate your ideas + writing, will look forward to what you put out in the future
This is a great point! It’s how Hercule Poirot usually solves his cases.
The code names part, not AI haha
now I sort of want there to be some kind of ai detective series where the detective could always solve the case, but ends up getting distracted by some small clue and doing down really long rabbit holes instead.
Oh man, we need a pilot episode for this concept!
"Now I need to find where this footprint came from."
Closest thing I can think of (in terms of rabbit holes) is... The Number 23? But that's not a comedy. ^_^'
Something in the wacky style of Silicon Valley or Big Bang Theory would work.
It could literally be called "AI Detective". :D
100% would watch.
@Netflix if you're reading this.
Maybe creativity is what happens when your knowledge graph stops optimizing for the shortest path.
I think that's more or less right. Nothing ever got more interesting by becoming more efficient.
Then originality has less to do with origin than with finding THE edge nobody else connected.
It reminded me of the movie Men in Black: if there are no solutions, then you have to eat apple pie...
And the thing is, when you feel desperate for a solution, stopping and eating apple pie never *feels* like the right thing to be doing.
I use my tried-and-true technique. When I can't find a solution, I switch to reading books unrelated to my field (software development, not coding itself). For example, books on biology by Richard Dawkins. And as I read, a solution appears. The idea is simple: a problem arises, then I study the problem (subconscious downloading), then I switch contexts (reading the book), and the solution appears. Of course, it doesn't always work, but generally it does.
The suggestion that less knowledge helps you have good ideas is backwards. Creativity and understanding are built on a bedrock of automated facts and procedures. They're necessary but not sufficient for good ideas, and the richer your mental library, the more surprising the connections you can make. Knowledge is the raw material of creativity.
Viewed this way: experts don't see more because they're smarter than you. Their knowledge upgrades their perceptual apparatus, and it's the apparatus that does the seeing. A sparse board doesn't force interesting connections, it simply limits you to the ones you happen to have.
As I like to say: you can't conjecture from an empty head.
Have written about this (and related topics) here: https://theinfinitehuman.substack.com/i/201352897/creativity-needs-a-foundation
+1
Except this: "A sparse board doesn't force interesting connections, it simply limits you to the ones you happen to have."
Necessity is the mother of invention, and creativity loves constraints / limitations breed creativity. Boundaries actually fuel the creative process. Anyone who has ever done improv knows this: an empty canvas is a lot less fun/creative than a constrained scenario. Completing a partial picture is a lot more fun than starting from scratch.
Constraints prevent choice paralysis: a blank canvas offers endless choices, so it's easy to get stuck, and generic ideas are the easiest way to get unstuck. But restrictions narrow our choices and make it easier to get started, or to branch out with a wacky idea. If we can't reuse generic ideas, we're forced to find something else. It's like cooking with unusual ingredient. What am I gonna do with this?
There is a fun game to play where a group of people write a story one word at a time, taking turns. We are forced to use the words that are already there and to try and direct the story in a direction that makes sense (or not). With all of the knowledge in the world, AI would suck at this because it would always use statistical probability to add the next word, and there's no fun in that. The diversity of each person's *constrained* knowledge is what makes the game fun / what makes the story unexpected. Ironically, the game is especially fun with children, *who have the least knowledge*, because young minds don't have the same guardrails as grown ones. Sometimes, knowledge is actually limiting, because we can't look outside of the box we have created for ourselves.
The world is a multiplayer game, but genius/intellect often restricts itself to a single-player experience.
Thanks for the thoughtful reply. I don't disagree constraints are helpful, but I think perhaps for different reasons e.g. providing clarity and focus.
The simpler version of what I'm saying (to address your point) still stands I think and would be the equivalent of trying to play the story game in a language no-one speaks.
As I said: knowledge is necessary but not sufficient for good ideas. I agree other things can get in the way of this (as you point out - and I agree with) but that's a separate effect in my view.
You can't have ideas about things you don't possess knowledge of. I've never had a good idea in microbiology, or quantum physics, or chemistry. Why? Because I know naff all about them.
I agree with this, to a point, I think. If you're clueless about everything, absolutely, you probably don't have many good ideas. If you're know a lot about lots of stuff, you probably have lots of good ideas; that was, like Da Vinci's whole thing. But I think I'd add one main details:
You can get stuck. If you know a ton about whatever thing, and spend all your time thinking about that thing, I've found it better to go run into some other stuff. In a sense, that proves your point - you're basically picking up other ideas - though I do think the timing of it matters, because *knowing* about something isn't quite the same as *thinking* about it. And really, that may be a better way of making the point - it's less about what you know, and more about putting different disparate ideas at the front of your mind.
And that's how AI goes wrong, I'd say. It knows tons of stuff, and could, in theory, do a really good job of running them into each other. But it doesn't on its own, because it doesn't need to.
Yeah, I guess there are (at least) three dimensions to getting (un)stuck:
1. Acquire more relevant knowledge;
2. Acquire other knowledge;
3. Go and do something else.
All can be helpful.
AFAUI the latter just helps you enter what they used to call "diffuse mode thinking" - no idea if it's still a thing these days. I think of this as just letting your brain escape local minima and don't see it as something that depends on the amount of knowledge you have. Kids routinely get stuck on one thing for ages with very little knowledge or exhibit "one track thinking".
Fascinating topic to think about!
Yeah, that's more or less exactly what it is, I think. Which clearly isn't a new or novel thought, though it does seem like something that's hard to convince yourself to do when you're stuck. It somehow always feels better to keep banging on the thing; it almost always works way better to do anything else.
Oh yeah - many a time* I've been locked-in banging my head against a wall with a problem knowing full well that taking a walk or break would help. Did I? Did I bollocks.
*pre-AI, when getting stuck was still a thing...
Great points. This is similar to Paul grahams essays about writing. The process of writing is the thinking.
This is what brainstorming is all about! Brain*storming*, not brain-sheltered-in-a-tunnel.
And it's also why diverse groups tend to come up with better ideas. Homogenous groups are all in the same tunnel already, and they don't want to go out in the rain.
I would assume there's tons of real research out there for how groups are better than individuals for these sorts of things, but my very vibe-based sense is yeah, that's a big part of it. On your own, it's so much harder to run into interesting or unexpected stuff.
literally introducing randomness to spark creativity: https://www.lennysnewsletter.com/p/how-to-turn-your-ai-into-a-world
I tried a version of this by having bots go read like 10 random read a bunch of random wikipedia pages or something, and then doing it's work with that stuff in its mind. It didn't quite work, because it turns out that "random" wikipedia pages were mostly historical biographies or really obscure science stuff, but it did produce some pretty weird stuff.
Keep your cup half-full so there is room to wander and see those flags, hanging just so.
You write beautifully, Benn. One thing I’ve always wondered is what your writing process looks like, especially given all the intricate connections you make to communicate an idea. I loved the codenames analogy!
I didn't really say this part that explicitly, but I do think this trick can be pretty effective for making things appear to be pretty connected. When you're writing the piece in tandem with the reference you're using to fit it, it makes it a lot easier to tie things together: https://benn.substack.com/p/how-to-have-ideas#footnote-7
This is an unbelievable read. Thanks for writing it Benn
This resonated with the focused vs. diffuse mode framing: focus gives you the problem and constraints; diffuse mode creates the unexpected connections. The trick is not to fully switch off the problem, but to hold one clear question lightly while exposing yourself to unrelated inputs. That seems to be where the useful collisions happen.
basically, yeah, at least that's how it works for me. It's keeping something in your head all the time, but always a bit in the background.
So we are back to this old problem: how do we defend our headspace to ensure the diffuse mode still works for ourselves?
something something try our ai bot to manage your emails?
well, I am convinced that it is WAY more than just that... there are also Slacks and other tools to act as the first gate of all of information going to our mind?
Oh for sure. It's just that when most all of these new tools come out, it feels like this is how it goes: https://x.com/thenanyu/status/2093411864667656633?s=20
Yeah, just boring old news. I wish for sth more groundbreaking
Dear Chat GBT make my sub stack sound like Matt Levine
Did it work? Where can we read your sub stack?
Let me rephrase in a less sassy way:
I find it difficult to engage with the author's ideas because I'm so distracted by the fact that the writing style is a blatant rip off Matt Levine. I enjoy the conversational simplicity of Money Stuff but seeing it aped in this context feels crass and kinda lame. I'd encourage the author to push harder to find their own organic voice and style, which will be better anyway in the long run.
The sassy remark was just the product of frustration and lack of restraint. All the best of luck to this author, writing well is hard.
I wasn't aware that Matt Levine owned this writing style. You don't have to read it if you don't like it.
So, I'm obviously not exactly a neutral party here, but I think this sort of thing is worth thinking about some, especially in a world where AI can kinda-sorta-but-badly engage in some form of stylistic plagiarism (though i generally find AI to be quite bad at writing of any kind, including editing, and don't use it at all for writing. It's useful for research and as a kind of automated Google on a loop, but, so far, not at all for writing.) Anyway, a few thoughts:
This is quote from a recent interview with Marlon James, who just published the "The Disappearers," and I generally agree with it. I'm a believer that every creative endeavor is mostly a mashup of prior influences, and developing that mashup *is* what it means to develop a style. It doesn't come out of the ether; it comes out of things that rhyme with whatever you read and consume.
---
"I read tons of novels when I’m working on books. I stare at people funny when they go, “I don’t want to be influenced by something.” I’m like, “You can only hope!” I absolutely want to be influenced by what I’m reading.
I can tell which parts of this book I was reading Don Quixote, and which parts I was reading Virginia Woolf, and which part I was reading my designated Toni Morrison novel for the year, which would probably be my fourth or fifth reading of that novel. To me, it’s like looking at art while painting. Yes, it’s an influence, but also you want to be surrounded with creativity and reminded of creativity. And sometimes I want to see how they do it.
(from here: https://www.deezlinks.com/p/how-i-wrote-this-book-marlon-james)
---
I would make basically the same argument here. Over the course of writing this thing, there are posts that have very different rhythms (some that are conversational; some that are not; in both cases, including posts like this one, I would generally say that I'm much more prone a kind of lofty melodrama and some structural weirdness than Matt Levine). Part of those differences, I'm sure, come from my mood and what feels good in that moment, which comes from other things I've read and liked recently.
But personally, I would rather write these posts in the way that feels natural both to me and the moment, and let that take each post where it may, rather than try to develop a "voice" and then edit each post to fit into that. This is actually a very annoying thing about AI to me; it frequently tells me things like "here's a way you might put." And I think that feels much more artificial - to chase a contrived character, be it your own, somebody else's, or the explicit avoidance of things that feel natural to you - than to write it as it comes.
Thanks Benn, I appreciate the thoughtful response, and I apologize for the snark above. Too much negativity on the internet
You are of course correct that our writing style is ultimately some kind of imitation and mashup of what we've consumed before + life experience (kind of like how llms do it minus the life experience haha).
For whatever reason I was triggered by the style in this post, and it has been my very-non scientific perception that there has been a stylistic shift in the substack toward the latent space dominated by Matt Levine's writing (at least in my mind). Take this as you will -- light feedback from a random on the internet.
Appreciate your ideas + writing, will look forward to what you put out in the future