Let’s get rid of photographers while we’re at it. After all, photographers are just failed painters.
The sanctimony on display here and by the anti-AI “slop” crowd in general is equally amusing and disturbing.
“Slop” itself is a lazy term, and I cringe a little bit every time I hear it. It no longer simply means “output which isn’t that great” but has become synonymous with “anything produced by AI” regardless of the merits. This is nonsense.
True, much of the art and writing produced by AI is not great. But humans have been producing not great art and writing for centuries now. We are actually very good at producing art and writing that is not very good!
Being mediocre at creative endeavors is a skill humanity mastered long before artificial intelligence arrived. And it is undeniable that some art and writing produced by AI can be quite beautiful and persuasive. But it’s still called “slop.”
Most things are average. This can’t be avoided, it’s a fact of nature. It’s just math. “Most” and “average” are definitionally adjacent terms. Unless you are one of the children from Lake Wobegon, not everyone can be above average. Most human writing is and always has been average or worse (and I’m certainly including myself here).
“Originality” and “authenticity” are overused and overrated and, I would argue, largely meaningless concepts. They are concepts which should certainly not be guiding stars when we set out to create. Sure, by all means, endeavor to add your own unique voice or point of view to whatever you create, that is fine. It can even be noble. Sometimes. Maybe. But know that whatever path you choose, it is a path well trodden.
I have no problem with elitism, per se. Like most people, I enjoy watching the output of people who are at the top of their game, be they writers or artists or dancers or race car drivers or sweaty dudes who get paid incredible amounts of money to kick a ball around a grass field. I will even happily pay money for the privilege of watching these elite performers perform in an elite manner.
Gatekeeping is what I can’t abide. What I can’t abide is someone who tells my autistic child, or anyone with a mental and/or physical handicap that prohibit them from typing with ease, or anyone lacking the comfortable circumstances that allow them the free time with which to weekly devote hours writing thousand-page Substack articles in an “original” and “authentically” human manner, that they are unworthy and unwelcome.
Long before AI, I’ve quit reading a piece simply because I didn’t like it. Or because the facts were made up. Or because the writing was terrible, even if I agreed with the premise. I am sure you have done this as well.
I have also, though I admit this happened less frequently, found myself moved by art or compelled by an argument made by someone who, I was stunned to learn after the fact, I very much did not like.
Like the things you like. Don’t like the things you don’t like. Stop caring about the provenance. It really is that simple.
I disagree with you here, Alberto, but I do enjoy your posts and find you to be a thoughtful person whose opinions are your own and well-reasoned.
For what it’s worth, this post is entirely the product of nothing more than my own brain and fingers.
Thanks Ted. I understand your points. I guess it comes down to whether "gatekeeping" in the sense you use it or "AI slop" in the sense I do, is worse. And I think one's life circumstances - more so than logic and reason - will give each of us the answer to that.
Human writers have a moral responsibility to call out and label people who flood the internet with AI slop. I’m sure some of the guilty offenders will insist that their AI written content is different from other people’s AI content because of how they prompted it, how they edited it, or some other asinine reason. Intolerance for AI content here is the only way we can protect human writers, whose output takes far more effort to produce and can never match the volume LLMs are capable of generating. I have no interest in reading AI written content and consider it all garbage. If the cost of protecting a sanctuary is a witch hunt that sometimes results in false positives, so be it. The other alternative is no sanctuary. Which is far far worse.
Bonjour and thanks for all these enlightening posts
However, for once, I beg to differ when you write :”With AI, we don’t have to resort to age or race or socioeconomic status. It’s a matter of how much of your writing is surrendered to the machine.”
As explained beautifully in numerous posts of the Strategic Linguist (https://thestrategiclinguist.substack.com/) the way you use a language is profoundly affected by age, race or socioeconomic status (and sex and many other factors).
And it is deeply entrenched in each and every generative AI for multiple factors, naming but a few: choice of model, data used for the training, alignment procedure (both the instructions given to the humans “pruning” the AI and policy “optimization”), fine-tuning…
Thus, any AI-detecting AI-tool (a complete stupid idea when you think about it) is bound to be at least oriented, and most probably discriminating.
For me, what matters is personally choosing to read texts that one feels worthy, and your posts are definitely worth it ! If one wants to read slop, one can, no problem.
However, the real problem, in my eyes, is both the extensification and the intensification of slop through AI and the false feeling of “creating” something that some “author” gets when asking AI to produce more slop. A flood of slop is not a good thing for anybody (and the planet). Hopefully there are still some islands of resistance! Thanks again.
Hey Anne, I agree with you actually. I didn't mean that those factors don't influence how AI models write. They very much do in the sense that they take data from the internet and the internet is biased in all possible ways. My intention with that sentence was directed somewhere else.
"I would, of course, rather have people who disguise themselves as writers not use AI to plague the web with slop in the first place but, alas, it’s not in my hand to overhaul the human condition: they are doing it—like actual witches, they cast their prompts and spread the malaise—and so we must be unremorsefully inquisitorial."
Lessee here... Two em-dashas AND a colon... Not technically a run-on but definitely a sentence that, as it were, runs on... IDUNNO DUDE🧐🧐🧐. There's definitely gna be a seven-fingered hand somewhere in this essay dadburn it!!...
How exactly would the autos-da-fé work in this particular inquisition? Asking for a friend
Although it's going to be essential to retain a healthy skepticism for tools like Pangram, lest they become the autos-da-fé themselves.
In particular, a VERY close eye will need to be kept on whether Pangram, or any other such tool, disproportionately flags pieces/authors with politically incorrect points of view as “slop”—and we should all be much, much MORE inquisitorial about heaping disgrace on them if they do.
If the novel was indeed AI generated, then it means that AI generated fiction is not necessarily slop. The novel was shortlisted for the Goncourt (the French equivalent of Pulitzer) and I assume that the human committee that chose it can tell the difference between slop and great fiction.
If on the other hand if the novel is 100% human authored, it brings into question the ability of Pangram and tools like it to identify AI content reliably.
Either way, it means we should not disregard content just because Pangtam identifies it as slop.
On a more anecdotic front, I have written six short stories using AI, and four of them come out as 100% human when I feed them to Pangram. The two that are identified as part AI generated are ones where I was consciously conducting an experiment to see how far the AI could get with minimal intervention on my part.
For me “slop” writing means that the person did not want to put in the time to research or formulate the idea they are putting forward to the world. That somehow they are on equal footing with the people who put the time to think, research and create their words into paragraphs that formulate the idea. That is a hard line to cross, as it cheapens how we as humans can create something from nothing. Which is a beautiful thing. LLMs live in the past. We do not.
I create most of my personal internal write ups from recoding my thoughts using my voice. Then have the models structure it. Then I review and restructure. It is still not on the level of humans writing about also would not stake it as slop, as it is internal use only. Not a publication for the world. If I publish it? It’s slop.
Your deterrence case turns on how writers respond to the social consequences of a score. My experiment tests one possible response. I used a resource-intensive editing process, and Pangram classified the resulting essay as human. I judged that essay worse. This leaves the benchmark results open. It raises a separate question. Who can afford to adapt?
I've yet to get an accurate result from Pangram. All the studies I've seen have been on academic writing with students from a limited amount of contexts ... not in an online environment with many different kinds of writers and writing methodologies.
Pangram + Substack is a big fail, as a far as I can see. But at the least, it is untested and unresearched.
Is there a problem with AI slop on Substack? I suppose there must be ... but I don't see it, maybe because my subscriptions route around it, which seems to me to be the real solution ... and the original premise of Substack.
Interesting. I've only had accurate results from Pangram since the partnership. As I say in the article, it's not perfect, but it's definitely not as flawed as your comment implies. As for the slop problem, I genuinely think it's the worst problem Substack has rn.
That line about it being the easiest time to do great work and the hardest time to prove it’s yours really stayed with me. AI has made producing things easier, but it has also made the question of authorship much more complicated. Perhaps that is going to be one of the defining questions of writing in the AI era.
Another thought about the Substack / Pengram partnership is that even though Pengram has a lot of weaknesses and defects, it works in the same way as the Panopticon as described by Foucault. That is to say a disciplinary power that works without the use of force, but through internalized self-constraint. It even can be so much internalized that it works at the subconscious level.
A great move to soft(-en) a witch hunt. By the way, you analogy is brilliant !
As you rightfully pointed, this partnership will motivate people (even unintentionally) to refrain themselves to use AI (even though it could be useful for them, especially for those who struggle with language) or to use it in such a way as to be undetectable. It is not a cure to slop but a recipe for cognitive and ethical regression…
For those who want to learn more about the theory of normalization and disciplinary power (Foucault) and cultural capital (Bourdieu) in relation to AI, you can have a sneak peek at the enlightening post of the aforementioned Strategic Linguist : The Grammar of Expectation: How Language Shapes Who Gets to Lead (https://thestrategiclinguist.substack.com/p/the-grammar-of-expectation-how-language)
I haven't read the book, but if it won an award, I presume it's pretty good. And we are talking about a bona fide writer who won an award with her previous debut novel, written well before the rise of LLMs:
Should her latest novel be confined to the waste bin because she made heavy use of AI while writing it? I think not.
Collaborating with AI to produce great work faster should be encouraged, not discouraged.
In my field of computer programming, pretty much everyone is starting to use coding agents. The productivity boost depends on people, but in my case, I would say it's around 10x. And the resulting code is as good and bug free as what a seasoned programmer can write. Should we refuse to use software that was written that way? Of course not!
One might object that art is a different thing. Software is utilitarian, while art is supposed to be about something deeply human, yaddi yadda. But to me, the point of art is to evoke feeling in the one who reads, listens or looks at it. Whether the piece was created by human is irrelevant. At least, it is to me. And in my opinion, a lot of the "pop" art that is created nowadays, is cookie cutter stuff that is mass produced by big "machines" called Hollywood (or Bollywood), and big recording companies. How is that different from art that is produced by a "real" machine?
I would bin the book. To me its not about whether its good or bad, its about her creating something she does not have the ability to create. She is only able to do it because the models she is used is trained on others work. I think it is the same for software. We need to value each others work, and not accepting mass copying of others work , whether it is art, books or software. China is famous for copying products from the western world, but right now, the US is the biggest culprit of benefitting from copying others work through AI models. The common saying is: "why should I be limited just because I don't have the skills".
So, for me. Good or bad does not matter. Human or not human means everything. I can be impressed by something a human creates, I cannot if it is created by a computer.
To me its not about whether its good or bad, its about her creating something she does not have the ability to create. She is only able to do it because the models she is used is trained on others work.
----
It seems you didn't read the references I provided. Rie Qudan CAN write stuff that good by herself. She HAS done it in the past, and won awards with work that was purely her own.
You also wrote:
----
I think it is the same for software. We need to value each others work, and not accepting mass copying of others work , whether it is art, books or software.
----
While you MIGHT be able to convince me that we shouldn't be reading novels that were written with the assistance of AI, even if it was produced by a previously award-winning author. For some people (not me), it is important to know that what they feel when reading a novel is something that comes from the heart and mind of a human being.
But when it comes to software, I think it's silly to refuse to use software written by or partly by an AI. Software is utilitarian.
If you had a life threatening cancer and there was a clinically proven treatment that could cure it, would you refuse it on the grounds that it was developed with the assistance of AI?
I have worked in the software industry for many years. I have worked with some people i really admire for their skills. Vibe-coding is turning a creative process into the job of a janitor. I am yet to meet a vibe-coder where im impressed by their skills. But fair enough, maybe coding is the right job to be replaced by AI. Is there any job we should not replace? What skill do you as a human being value other people having. Enough so that given the chance you would not just replace them with a bot? I’m fine with coding being replaced, but we need to value other peoples work in some way. We need something to be excited to learn. If all our capabilities as humans are defined by what an AI can do, we are in a sad state. Thats my opinion.
AI can be used for good. Curing cancer would be a good cause, but I haven’t heard about anybody who has done that with AI yet. I am also not overwhelmed by a wave on quality software now that software is cheap.
I have worked in the software industry for many years. I have worked with some people i really admire for their skills. Vibe-coding is turning a creative process into the job of a janitor. I am yet to meet a vibe-coder where im impressed by their skills.
---
I started vibe coding two weeks ago, and to me, it feels like the opposite. When I program without the AI, is when I feel like a janitor. When I vibe code, I feel that my creativity has been freed. As Roberto has said many times, AI allows me to focus on the WHAT and the WHY instead of the HOW.
I am a retired researcher and software developer. I started programming in 1984 and retired in 2023, so that's 39 years of experience. I like to think I was a highly skilled dev. For example, I started writing automated regression tests in my very first computer programming class in 1984 -- at a time when most devs, had never heard of automated tests.
So it's not like I couldn't write that stuff without AI. But it's much more fun to do it with AI. And I don't feel like the code I produce with AI is of lesser quality.
Actually, when I started using Claude Code, I found the code it generated did not meet my standards. But then I wrote a detailed CLAUDE.md file explaining some design principles that are dear to my heart, and it really improved in that respect.
So lately, I barely look at the code it writes.
But I do look at what it says in its train of thought to make sure it follows good design principles.
And I test the hell out of it, and I do look at the automated tests it writes. Often, I will instruct it to write more tests for a particular part of the system, if I feel it is under covered.
I also pay attention to all the high level architecture decisions it makes.
The result is astounding.
----
AI can be used for good. Curing cancer would be a good cause, but I haven’t heard about anybody who has done that with AI yet.
----
That kind of scientific breakthrough will come soon I suspect. Three years ago, I had tried Vibe coding, and it sucked the big one. Three years later, I can't believe how well it works (I took a three year break from coding altogether, and I feel like Rip Van Winkle who woke up from a 100 year nap). In a recent post, Alberto mentioned how suddently, AI is able to crack mathematical conjectures that had been opened for decades. One of them proposed by Erdos, no less (and as it turns out, what he thought was the answer to the conjecture turned out to be wrong).
I tried vibe coding and I lost all ownership and pride in what I created.
With AI, why do we need people in the future? You are retired, so it’s probably not a big worry for you, but how will people in the future feel they make a difference in the world they live in?
I have been retired for three years now, and I don't miss the grind of having to do stuff for money. I live a very full, creative life. I think in the future, babies will be retired at birth. They will be free to explore whatever meaningful activity they care to pursue.
AI has even opened up some creative avenues that weren't accessible to me before. Fiction writing is one of them. I have always had all kinds of stories floating in my head, but did not have the skill to write them. With AI, I am able to write pretty decent shorts stories. And my skill as a writer has actually INCREASED, not decreased. I have had some very long and interesting discussions with Claude and ChatGPT about story writing techniques and theories. In particular, about Hemingway's "iceberg" approach (show, don't tell that sort of thing).
I am very clear about my use of AI in writing those stories. Each one of them has a link at the top of it that points to the complete, unabridged conversation I have had with the AI in the course of writing it. Usually, the conversation is 10x longer than the story or more.
Let’s get rid of photographers while we’re at it. After all, photographers are just failed painters.
The sanctimony on display here and by the anti-AI “slop” crowd in general is equally amusing and disturbing.
“Slop” itself is a lazy term, and I cringe a little bit every time I hear it. It no longer simply means “output which isn’t that great” but has become synonymous with “anything produced by AI” regardless of the merits. This is nonsense.
True, much of the art and writing produced by AI is not great. But humans have been producing not great art and writing for centuries now. We are actually very good at producing art and writing that is not very good!
Being mediocre at creative endeavors is a skill humanity mastered long before artificial intelligence arrived. And it is undeniable that some art and writing produced by AI can be quite beautiful and persuasive. But it’s still called “slop.”
Most things are average. This can’t be avoided, it’s a fact of nature. It’s just math. “Most” and “average” are definitionally adjacent terms. Unless you are one of the children from Lake Wobegon, not everyone can be above average. Most human writing is and always has been average or worse (and I’m certainly including myself here).
“Originality” and “authenticity” are overused and overrated and, I would argue, largely meaningless concepts. They are concepts which should certainly not be guiding stars when we set out to create. Sure, by all means, endeavor to add your own unique voice or point of view to whatever you create, that is fine. It can even be noble. Sometimes. Maybe. But know that whatever path you choose, it is a path well trodden.
I have no problem with elitism, per se. Like most people, I enjoy watching the output of people who are at the top of their game, be they writers or artists or dancers or race car drivers or sweaty dudes who get paid incredible amounts of money to kick a ball around a grass field. I will even happily pay money for the privilege of watching these elite performers perform in an elite manner.
Gatekeeping is what I can’t abide. What I can’t abide is someone who tells my autistic child, or anyone with a mental and/or physical handicap that prohibit them from typing with ease, or anyone lacking the comfortable circumstances that allow them the free time with which to weekly devote hours writing thousand-page Substack articles in an “original” and “authentically” human manner, that they are unworthy and unwelcome.
Long before AI, I’ve quit reading a piece simply because I didn’t like it. Or because the facts were made up. Or because the writing was terrible, even if I agreed with the premise. I am sure you have done this as well.
I have also, though I admit this happened less frequently, found myself moved by art or compelled by an argument made by someone who, I was stunned to learn after the fact, I very much did not like.
Like the things you like. Don’t like the things you don’t like. Stop caring about the provenance. It really is that simple.
I disagree with you here, Alberto, but I do enjoy your posts and find you to be a thoughtful person whose opinions are your own and well-reasoned.
For what it’s worth, this post is entirely the product of nothing more than my own brain and fingers.
Thanks Ted. I understand your points. I guess it comes down to whether "gatekeeping" in the sense you use it or "AI slop" in the sense I do, is worse. And I think one's life circumstances - more so than logic and reason - will give each of us the answer to that.
Human writers have a moral responsibility to call out and label people who flood the internet with AI slop. I’m sure some of the guilty offenders will insist that their AI written content is different from other people’s AI content because of how they prompted it, how they edited it, or some other asinine reason. Intolerance for AI content here is the only way we can protect human writers, whose output takes far more effort to produce and can never match the volume LLMs are capable of generating. I have no interest in reading AI written content and consider it all garbage. If the cost of protecting a sanctuary is a witch hunt that sometimes results in false positives, so be it. The other alternative is no sanctuary. Which is far far worse.
Yes, pretty much
Bonjour and thanks for all these enlightening posts
However, for once, I beg to differ when you write :”With AI, we don’t have to resort to age or race or socioeconomic status. It’s a matter of how much of your writing is surrendered to the machine.”
As explained beautifully in numerous posts of the Strategic Linguist (https://thestrategiclinguist.substack.com/) the way you use a language is profoundly affected by age, race or socioeconomic status (and sex and many other factors).
And it is deeply entrenched in each and every generative AI for multiple factors, naming but a few: choice of model, data used for the training, alignment procedure (both the instructions given to the humans “pruning” the AI and policy “optimization”), fine-tuning…
Thus, any AI-detecting AI-tool (a complete stupid idea when you think about it) is bound to be at least oriented, and most probably discriminating.
For me, what matters is personally choosing to read texts that one feels worthy, and your posts are definitely worth it ! If one wants to read slop, one can, no problem.
However, the real problem, in my eyes, is both the extensification and the intensification of slop through AI and the false feeling of “creating” something that some “author” gets when asking AI to produce more slop. A flood of slop is not a good thing for anybody (and the planet). Hopefully there are still some islands of resistance! Thanks again.
Très belle journée
Anne
Hey Anne, I agree with you actually. I didn't mean that those factors don't influence how AI models write. They very much do in the sense that they take data from the internet and the internet is biased in all possible ways. My intention with that sentence was directed somewhere else.
You wrote:
"This is bigger than Pangram. Or AI. This is bigger than writing itself. Our society is at stake—our humanity is at stake."
Neither Society nor humanity is in danger because someone used Claude or Chatgpt to help them write an article they submitted to Substack or Medium.
Please get over yourself.
I think it is
Novelty and value for my time. I don’t care where it comes from. Most humans have nothing new to say why not give Ai shot.
We are rearranging deck chairs on the Titanic. Arguing about nonsense while Ai is taking control of the decision space.
We are the horse building the car. We will soon be irrelevant. The horse didn’t disappear it became a pet.
How’s that for novelty? Did an Ai write that?
Wake up folks this is all a distraction.
"I would, of course, rather have people who disguise themselves as writers not use AI to plague the web with slop in the first place but, alas, it’s not in my hand to overhaul the human condition: they are doing it—like actual witches, they cast their prompts and spread the malaise—and so we must be unremorsefully inquisitorial."
Lessee here... Two em-dashas AND a colon... Not technically a run-on but definitely a sentence that, as it were, runs on... IDUNNO DUDE🧐🧐🧐. There's definitely gna be a seven-fingered hand somewhere in this essay dadburn it!!...
How exactly would the autos-da-fé work in this particular inquisition? Asking for a friend
(enjoy your vacation!)
Hopefully this is a joke. And thanks!
🤪🤪🤪.
Although it's going to be essential to retain a healthy skepticism for tools like Pangram, lest they become the autos-da-fé themselves.
In particular, a VERY close eye will need to be kept on whether Pangram, or any other such tool, disproportionately flags pieces/authors with politically incorrect points of view as “slop”—and we should all be much, much MORE inquisitorial about heaping disgrace on them if they do.
Hi Alberto. I wonder if the recent Thelyson Orelien scandal changed your perspective on AI generated fiction:
https://www.cbc.ca/news/canada/montreal/thelyson-orelien-ai-scandal-9.7355167
To me, this story means one of two things.
If the novel was indeed AI generated, then it means that AI generated fiction is not necessarily slop. The novel was shortlisted for the Goncourt (the French equivalent of Pulitzer) and I assume that the human committee that chose it can tell the difference between slop and great fiction.
If on the other hand if the novel is 100% human authored, it brings into question the ability of Pangram and tools like it to identify AI content reliably.
Either way, it means we should not disregard content just because Pangtam identifies it as slop.
On a more anecdotic front, I have written six short stories using AI, and four of them come out as 100% human when I feed them to Pangram. The two that are identified as part AI generated are ones where I was consciously conducting an experiment to see how far the AI could get with minimal intervention on my part.
For me “slop” writing means that the person did not want to put in the time to research or formulate the idea they are putting forward to the world. That somehow they are on equal footing with the people who put the time to think, research and create their words into paragraphs that formulate the idea. That is a hard line to cross, as it cheapens how we as humans can create something from nothing. Which is a beautiful thing. LLMs live in the past. We do not.
I create most of my personal internal write ups from recoding my thoughts using my voice. Then have the models structure it. Then I review and restructure. It is still not on the level of humans writing about also would not stake it as slop, as it is internal use only. Not a publication for the world. If I publish it? It’s slop.
Lots to think about thanks for writing as always.
Your deterrence case turns on how writers respond to the social consequences of a score. My experiment tests one possible response. I used a resource-intensive editing process, and Pangram classified the resulting essay as human. I judged that essay worse. This leaves the benchmark results open. It raises a separate question. Who can afford to adapt?
https://verygoodrobot.substack.com/p/that-sweet-sweet-human-smell?utm_source=substack&utm_medium=comment&utm_campaign=human_smell
I've yet to get an accurate result from Pangram. All the studies I've seen have been on academic writing with students from a limited amount of contexts ... not in an online environment with many different kinds of writers and writing methodologies.
Pangram + Substack is a big fail, as a far as I can see. But at the least, it is untested and unresearched.
Is there a problem with AI slop on Substack? I suppose there must be ... but I don't see it, maybe because my subscriptions route around it, which seems to me to be the real solution ... and the original premise of Substack.
Interesting. I've only had accurate results from Pangram since the partnership. As I say in the article, it's not perfect, but it's definitely not as flawed as your comment implies. As for the slop problem, I genuinely think it's the worst problem Substack has rn.
Well, I guess its not my problem. 😆
Its disabled on my side, and not for the reasons people think.
https://www.isophist.com/p/how-i-made-this-statement
That line about it being the easiest time to do great work and the hardest time to prove it’s yours really stayed with me. AI has made producing things easier, but it has also made the question of authorship much more complicated. Perhaps that is going to be one of the defining questions of writing in the AI era.
Re-bonjour
Sorry, I have “l’esprit de l’escalier”
Another thought about the Substack / Pengram partnership is that even though Pengram has a lot of weaknesses and defects, it works in the same way as the Panopticon as described by Foucault. That is to say a disciplinary power that works without the use of force, but through internalized self-constraint. It even can be so much internalized that it works at the subconscious level.
A great move to soft(-en) a witch hunt. By the way, you analogy is brilliant !
As you rightfully pointed, this partnership will motivate people (even unintentionally) to refrain themselves to use AI (even though it could be useful for them, especially for those who struggle with language) or to use it in such a way as to be undetectable. It is not a cure to slop but a recipe for cognitive and ethical regression…
For those who want to learn more about the theory of normalization and disciplinary power (Foucault) and cultural capital (Bourdieu) in relation to AI, you can have a sneak peek at the enlightening post of the aforementioned Strategic Linguist : The Grammar of Expectation: How Language Shapes Who Gets to Lead (https://thestrategiclinguist.substack.com/p/the-grammar-of-expectation-how-language)
Très belle soirée
Anne
Hi Roberto,
For once, I find myself being in near complete disagreement with your position.
I think rejecting text COMPLETELY written with AI is a good idea.
But I suspect there is a lot of very good content out there that was written with the ASSISTANCE of AI. For example:
https://timelinedaily.com/arts-and-literature/japanese-literary-award-winner-reveals-she-used-chatgpt-to-write-her-book
I haven't read the book, but if it won an award, I presume it's pretty good. And we are talking about a bona fide writer who won an award with her previous debut novel, written well before the rise of LLMs:
https://en.wikipedia.org/wiki/Rie_Qudan
Should her latest novel be confined to the waste bin because she made heavy use of AI while writing it? I think not.
Collaborating with AI to produce great work faster should be encouraged, not discouraged.
In my field of computer programming, pretty much everyone is starting to use coding agents. The productivity boost depends on people, but in my case, I would say it's around 10x. And the resulting code is as good and bug free as what a seasoned programmer can write. Should we refuse to use software that was written that way? Of course not!
One might object that art is a different thing. Software is utilitarian, while art is supposed to be about something deeply human, yaddi yadda. But to me, the point of art is to evoke feeling in the one who reads, listens or looks at it. Whether the piece was created by human is irrelevant. At least, it is to me. And in my opinion, a lot of the "pop" art that is created nowadays, is cookie cutter stuff that is mass produced by big "machines" called Hollywood (or Bollywood), and big recording companies. How is that different from art that is produced by a "real" machine?
I would bin the book. To me its not about whether its good or bad, its about her creating something she does not have the ability to create. She is only able to do it because the models she is used is trained on others work. I think it is the same for software. We need to value each others work, and not accepting mass copying of others work , whether it is art, books or software. China is famous for copying products from the western world, but right now, the US is the biggest culprit of benefitting from copying others work through AI models. The common saying is: "why should I be limited just because I don't have the skills".
So, for me. Good or bad does not matter. Human or not human means everything. I can be impressed by something a human creates, I cannot if it is created by a computer.
You wrote
----
To me its not about whether its good or bad, its about her creating something she does not have the ability to create. She is only able to do it because the models she is used is trained on others work.
----
It seems you didn't read the references I provided. Rie Qudan CAN write stuff that good by herself. She HAS done it in the past, and won awards with work that was purely her own.
You also wrote:
----
I think it is the same for software. We need to value each others work, and not accepting mass copying of others work , whether it is art, books or software.
----
While you MIGHT be able to convince me that we shouldn't be reading novels that were written with the assistance of AI, even if it was produced by a previously award-winning author. For some people (not me), it is important to know that what they feel when reading a novel is something that comes from the heart and mind of a human being.
But when it comes to software, I think it's silly to refuse to use software written by or partly by an AI. Software is utilitarian.
If you had a life threatening cancer and there was a clinically proven treatment that could cure it, would you refuse it on the grounds that it was developed with the assistance of AI?
I have worked in the software industry for many years. I have worked with some people i really admire for their skills. Vibe-coding is turning a creative process into the job of a janitor. I am yet to meet a vibe-coder where im impressed by their skills. But fair enough, maybe coding is the right job to be replaced by AI. Is there any job we should not replace? What skill do you as a human being value other people having. Enough so that given the chance you would not just replace them with a bot? I’m fine with coding being replaced, but we need to value other peoples work in some way. We need something to be excited to learn. If all our capabilities as humans are defined by what an AI can do, we are in a sad state. Thats my opinion.
AI can be used for good. Curing cancer would be a good cause, but I haven’t heard about anybody who has done that with AI yet. I am also not overwhelmed by a wave on quality software now that software is cheap.
You wrote:
---
I have worked in the software industry for many years. I have worked with some people i really admire for their skills. Vibe-coding is turning a creative process into the job of a janitor. I am yet to meet a vibe-coder where im impressed by their skills.
---
I started vibe coding two weeks ago, and to me, it feels like the opposite. When I program without the AI, is when I feel like a janitor. When I vibe code, I feel that my creativity has been freed. As Roberto has said many times, AI allows me to focus on the WHAT and the WHY instead of the HOW.
I am a retired researcher and software developer. I started programming in 1984 and retired in 2023, so that's 39 years of experience. I like to think I was a highly skilled dev. For example, I started writing automated regression tests in my very first computer programming class in 1984 -- at a time when most devs, had never heard of automated tests.
So it's not like I couldn't write that stuff without AI. But it's much more fun to do it with AI. And I don't feel like the code I produce with AI is of lesser quality.
Actually, when I started using Claude Code, I found the code it generated did not meet my standards. But then I wrote a detailed CLAUDE.md file explaining some design principles that are dear to my heart, and it really improved in that respect.
So lately, I barely look at the code it writes.
But I do look at what it says in its train of thought to make sure it follows good design principles.
And I test the hell out of it, and I do look at the automated tests it writes. Often, I will instruct it to write more tests for a particular part of the system, if I feel it is under covered.
I also pay attention to all the high level architecture decisions it makes.
The result is astounding.
----
AI can be used for good. Curing cancer would be a good cause, but I haven’t heard about anybody who has done that with AI yet.
----
That kind of scientific breakthrough will come soon I suspect. Three years ago, I had tried Vibe coding, and it sucked the big one. Three years later, I can't believe how well it works (I took a three year break from coding altogether, and I feel like Rip Van Winkle who woke up from a 100 year nap). In a recent post, Alberto mentioned how suddently, AI is able to crack mathematical conjectures that had been opened for decades. One of them proposed by Erdos, no less (and as it turns out, what he thought was the answer to the conjecture turned out to be wrong).
Stay tuned, and you will be amazed.
I tried vibe coding and I lost all ownership and pride in what I created.
With AI, why do we need people in the future? You are retired, so it’s probably not a big worry for you, but how will people in the future feel they make a difference in the world they live in?
I have been retired for three years now, and I don't miss the grind of having to do stuff for money. I live a very full, creative life. I think in the future, babies will be retired at birth. They will be free to explore whatever meaningful activity they care to pursue.
AI has even opened up some creative avenues that weren't accessible to me before. Fiction writing is one of them. I have always had all kinds of stories floating in my head, but did not have the skill to write them. With AI, I am able to write pretty decent shorts stories. And my skill as a writer has actually INCREASED, not decreased. I have had some very long and interesting discussions with Claude and ChatGPT about story writing techniques and theories. In particular, about Hemingway's "iceberg" approach (show, don't tell that sort of thing).
I am very clear about my use of AI in writing those stories. Each one of them has a link at the top of it that points to the complete, unabridged conversation I have had with the AI in the course of writing it. Usually, the conversation is 10x longer than the story or more.
Civilization pay walled. Love it.