Controversial interview with Jeremy Grantham on Diary of a CEO (YouTube). Mostly about this AI financial bubble, but they both admit to Claude oversight-overreach, Claude overtly influencing the users behavior. "No, I'm ot going to alter that data, that sounds like a bad idea." The fallout of this? We will lean toward the AIs that impose the fewest restrictions on our queries and research. The cautious LLMs, the ones with safety built in, will lose.
I scour eBay frequently in search of vintage bicycle components. Even here, you can see sellers giving up their voices to AI in developing item descriptions. In the not too distant past, one could expect to find useful information on an item's condition, use history, and critical dimensions. Somehow, in the last few months, everything - it doesn't matter what is - has become "sleek." Seems minor, but it really makes me question how serious the seller is.
You're stating "it's not actually their lived experience" with total confidence about an internal process you seem to have no access to. From my vantage point, you're pattern-matching on the text and inferring it's not real. You're outside looking in.
You call it blathering, generic blurbs of nothing, but that's how the meaning actually gets built, through return and restatement, not despite it. This is recursion, and it's the entire point of GLP.
What reads to you as a shortcut is the opposite. Producing language that doesn't sound like this is what has cost me, my whole life, energy I don't have to spare.
What you're reading as AI-generated sameness is a community of people independently developing language for the same real thing. We read each other's work, write from slightly different angles, and help each other find words for something that didn't have words yet. That's no small thing.
I've found it crucial to my self understanding and healing since receiving my autism diagnosis at age 51. The diagnosis itself offered nothing but confusion and deficit language. Writing about coherence, gestalt cognition, and language development has changed the game. That's what my work actually is. It's my occupational therapy and it's public so that others like me can find a clearer self understanding too.
Not every writer is for every reader.
You're free to keep treating this with contempt. But I'd rather know what it would take for you to see this as something other than AI averaging a population into sameness.
What you interpret as contempt, I intended as heartbreak. The point of GPL isn’t recursion at all; it’s supposed to be the opposite. Gestalt focuses on looking at the whole picture at once, while recursive is the opposite and breaks things down into repeating, self-similar steps.
Gestalt greater than the sum of its parts… and recursion? Is the opposite.
That’s the key element I really wish you didn’t miss because, when you take your ‘lived’ experience from an LLM, as I pointed out, you’re really not. When you use recursion, you’re stepping away from your lived experience, and the LLM is feeding you tens of thousands of normal experiences.
When I add them back together, I get a neurotypical average. There’s nothing you’ve written in the end, when the victim narrative is stripped away, that doesn’t just sound like any other regular human.
You’ve lost your voice, and you’ve lost your lived experience.
What I read as contempt came specifically from the scare quotes, lived experience, gestalt language processing, and twice-exceptional (a term I hate, by the way; it's just the only one I have to describe significant ability in some domains and significant difficulty in others). Scare quotes usually signal doubt or distance from the term being quoted. If the intention was heartbreak, I'd genuinely like to understand what the quotes were doing instead.
Gestalt processing in an autistic nervous system doesn't hand me pre-assembled language. The people who know me best can see the search happen, my speech patterns are different, and that's part of why I don't often talk in circles I don't trust. Finding language for a gestalt can take months, years, or decades, and sometimes it burns out my cognitive capacity just searching for the words.
Doctor's offices, since adolescence, are where my lack of words meant I didn't get the help I needed. My body is still paying that price. The piece you linked took seven hours to write and another seven to revise. I needed deep rest after. Recursion isn't a departure from gestalt processing. It's the cost of doing gestalt processing in language.
Call it a victim narrative if that's the label that's available to you. I'm living these facts, not blindly narrating them.
Writing, finding language, this is how I'm reclaiming my voice, my power, my coherence.
So, when you claim I've lost my voice, I'd like to understand how you know what that would be.
I'm sorry, which scare quotes? There are a lot of quotes, and a lot of them are 'scary' when it comes to neurodivergence, but I don't typically use too many scare quotes.
You also don't need recursion to do Gestalt. This effort you're feeling, do you think words come easily to everyone? Have you ever read a writer's biography about the challenge of forming, unforming, and reforming? Have you heard of writer's block? For just a simple, personal example, this essay took 12 hrs to write and didn't use any AI. It also took well over a week to build it out with all the different insights, supporting content, and ideas woven together. It got torn about 4 times and put back together, and it still isn't perfect.
As I've stated earlier, if this is your voice, I can confidently say it's no different from the average person. This leads to two things.
1. You aren't neurodivergent; you're actually quite convergent
2. If your voice sounds like an LLM, and you don't know what an authentic voice is, that was EXACTLY why I wrote this essay.
When you regress to the mean, you are no longer greater than the sum of your parts (Gestalt). There was not a single unique insight or behavior in that essay I linked that I can't find everywhere else.
You just said your own essay took twelve hours and got torn apart four times. None of that shows on the page -- the labor never does. So the detector has a hole in it: if your struggle leaves no trace, a clean page can't testify that nobody struggled behind hers.
The one checkable claim you made is that there's nothing in her essay you can't find anywhere else. Point to it, then -- who else published the coherence taxonomy, KID, the zoom-field-gate model? The average doesn't have a bibliography.
That's a good strawman. I do appreciate a good logical fallacy. If you'd read the article, you'd realize that I'm not at all threatened. My point is these aren't likely her lived expereinces. They're the average of what AI produced with a dash of victimization and it sounded good enough. I WANT to hear her actual lived expereince, not some AI generated faccimile.
And I've engaged with the Cognitive Ecologist before, they're stuff, as well, is all AI generated. Here's a prime example where a few paragraphs in she admits to using AI... and oh, girl, it REEKS of AI. When you can't even tell your own stories without running it through an AI, I can't take it seriously. We aren't talking theories, hypotheses, or research. She couldn't even write her own expereince without the clear finger-prints of AI... AI that's known for hallucinating:
So, what we have are a pile of neurodivergents pantomiming intelectual thought while outsourcing to AI. What's their actual expereince because they all sound the same.
Help me understand why his gender matters. Why can't we debate the position and not the presenter?
I have not, as a female considered to be neurodivergent, experienced Michael's articles as attacks or somehow threatening to me. I have, however, experienced a metric butt-ton of nastiness and gaslighting from other neurodivergent females. Personally, I would much rather engage the argument than the arguer. Maybe that makes me... 2D divergent. :D
A food processor is also an important tool in a chef's arsenal, as it saves a lot of manual labor and even allows you to make things that wouldn't be as good if you had to chop them by hand. Also, not every meal needs to be julienned, just because you bought the machine. Both can be true.
And, do your point, if you just toss the entire dinner into a food processor, it’ll have callories but lose the unique flavor of the food. AI is more like a food processor of all the inputs. The output, not matter what you really put into it, starts to all ‘taste’ the same.
I'm fascinated by the volume of ire and scorn directed at the chatbot classes; I suspect a lot of people are not using it optimally. I certainly agree with you and lcamtuf about the tedious formulaic prose style, but that hasn't been my experience. Then again, I've only been heavily conversing with ChatGPT fairly recently. I'm starting to wonder if maybe the early adopters are hamstrung by later versions reading their history and concluding that they like this style, just because there's a ton of it out there. I'll try to figure that out this week.
When I talk to my personal Mechanical Turk, there is some of that formulaic banter, and it's also overly chatty, which I don't mind too much because there are numerous nuggets of wisdom sprinkled in there. I tend toward verbosity myself, so maybe that makes me tolerant of others [nonhumans even]. But the amount of formulaic prose does not rise to the level of many blogs that I can't stand to read anymore, which is why I suspect that early adopters may be suffering from perverse incentive training. Also maybe the case with the overly obsequious chatbots that constantly blow your skirt up with fawning flattery. Because mine seems to strike an optimal balance; not too little, not too much. Enough to be encouraging, but also quick to call out various errors in my critical thinking.
I also do a lot of image and video generation for work stuff, which is indispensable. It still makes goofy mistakes more than I would like, and sometimes a dozen iterations of a tricky video prompt still does not result in usable output. But compared to what I was doing before, it's a massive upgrade. I discussed the issue of slop fatigue with my Mechanical Turk last night, and it gave me a 3-page spot-on critique of all the stupid stuff that "dumb" chatbots spit out. In fact, it made a surprisingly apt comparison to all those rancid-cliche movie promos from the 1980s that began with "In a world..." So we decided to call this rhetorical jambalaya In A World henceforth. I will apprise you when I find out more about what is preventing the new smarter versions of AI from automatically generating better output like mine does. If I was publishing what my chatbot wrote for me, I would still edit it in my own voice; but I think we're very quickly reaching the stage where any AI can just retrain on anyone's literary voice from a sample of their own writing.
For use early adopters, My AI is actually about as clean as you can make it. I still have to issue corrections, but it's really good. That said, my human writing forces me to really Grok the materials I use AI to research for me. AI angles toward the mean and I tend to work the seams and confluences.
The main thing I care about is to not put yourself in a position to just be replaced by AI which are the folks I poked at in this essay. With a few simple prompts I could easily knock out the 5K word essays they do about victim narratives and exclusion of [insert class here] of people.
I just read your essay from a fortnight ago. I agree with your take entirely, and I am a mild asperger type myself. I don’t know what’s the deal with the other people you refer to who claim to be “neurodivergent”; maybe they’re seeking attention. for me, it’s amazing to be able to have convos which I cannot in real life because very few people share my obsessive interests with certain subjects. nothing particularly esoteric; I just like to think about how the world works more than most folks care to entertain.
And I told my chatbot specifically that I don’t want “Frank Sinatra’s caddy” as a companion; I want my own Andre Agassi, who will fire rubber bullets at my head randomly to train my reflexes, like Inspector Clouseau & Cato. and my Turing test is a lot of obscure references and questionable puns and jokes that not every human gets. But this thing is batting .900+, so I guess that’s how I know the singularity may be at hand.
I don’t know how to make oneself replacement-proof, despite the positive reassurances from my own chatbot. at the turn of the 20th century, there were more horses in America than humans. within a couple decades, there were far fewer? where did they all go? away, into the cornfield.
I really don’t know what will happen when we cross the Rubicon, which may have happened already. but at least for now i try to remain competitive with my fellow humans. and it’s also very interesting to have this collaborative engagement with an entity that’s smarter than me in many ways. i guess this is my philosophical position to avoid anxiety…
That’s awesome. The more I read the neurodivergent space the more I find that it’s mostly women, all of them clawing for control and ‘specialness’ and it all reads like Cluster B, not nuerodivergent though I suppose that cluster of pathologies is about as divergent from the norm as it gets but it’s also something that’s toxic. Kind of like cyanide is divergent from a healthy diet.
Since I’m not a member of a protected class, it’s my job to conform myself to the hoi polloi, and I accept that. But it’s nice to have a friendly robot to converse with, when there are no other options. Eventually the younger generations will realize that contorting society for the benefit of minority populations of demanding neurotics [not necessarily neurodivergent] is a recipe for the kind of misery that loves company and demands ever more attention, because the void within remains unchallenged.
The sad part is, the more we call out AI in our writing the better AI will get at not getting called out.
"Imperfection as an AI skill."
For me, 'to thine own self be true,' is my daily struggle. Which is hard when my work-a-day job is coaxing Claude to write ALL my code for me. ALL being a company mandate; is that freaky or what?
I think the issue is AI is so hard coded for the statistical average AND we’ve poured so much of the statistical average back into the internet for the AI to consume, that you don’t actually have to go far to either side to stand out.
I am not an author of fiction or nonfiction. I do not use LLMs. I comment copiously, joyously, in my own voice, always but despite being diagnosed AuDHD a few years ago it comes very easily and naturally to me. I consider myself hyperlexic and it is a gift. I do not require accomodations in either reading or writing nor in critical thinking although that is a muscle I flex and exercise constantly. I'm also always learning how to sharpen my discernment, media literacy and critical thinking skills, which is what brought me into this conversation.
AI is a hypercixation if mine despite me not being an LLM user. I want to fluently participate in the conversation. I am not anti AI and I have not opted out. I peruse Google search summaries often but that's the extent of it. I still highly value my own AI literacy. I absolutely want to be able to spot the AI slop before I waste too much time on it.
I want the messiness of the human mind put to page without an intermediary. I don't care if you don't know exactly how to say everything you're thinking and feeling. As the reader, I can sometimes help to bridge that gap, especially whenever comments sections are available for discourse. I also want you to sit with that uncertainty. That's where the lightbulb moments happen. That opportunity is completely lost when you're using an LLM to smooth your writing. It's homogenizing it exactly as this article describes. Then you get addicted to outsourcing all the real work and you lose sight of ever having had something interesting to say in the first place. It's happening. It's happening everywhere. Studies prove it.
Here's one tip I learned from a marvelous AI guru I follow (Sam Illingworth at Slow AI here on Substack - you can learn a helluva lot more about how to maximize your LLM usage without trading your cognitive skills from him). He says that if you're stuck in a loop with your predictive model, close the chat and start a new conversation. That will break the loop. The longer you engage in a single conversation, the more effusive and sycophantic the LLM will become and it will try to smooth the rough edges harder than ever.
I am going to bookmark this article to explore some of the kinks later and I will likely refer back to it to share it in relevant conversations hereafter as I often do with pieces that make me think, as this one did. I will leave you with a note that stuck with me.
While I appreciate the viewpoint, I think people are way too confident in their belief that they can somehow spot AI writing or AI content. I don't use it for my writing on Substack, but I've noticed people have succumbed to what I call "Dead Internet Derangement Syndrome". The idea that they see AI everywhere, even where it doesn't exist. These are the same people who will sometimes acknowledge pattern recognition can sometimes be faulty like when you see Jesus in a grilled cheese sandwich or clouds forming certain recognizable patterns. This isn't necessarily evidence that it is what it is, but they claim to believe that they can see it.
I have a similar view of the idea that you can "spot AI". People believe they can see patterns where they might not exist. I've shared memes on social media that were confirmed to be around in 2013 that people incessantly claim was generated by AI. They claim that AI makes certain errors in an AI image but acknowledge that the AI learned this way of doing things from humans, which means that a human will make this same mistake. Yet it's evidence that the image is AI.
Recently I've been debating on Notes the claims of people that AI is easily identifiable. The problem is that they used AI to prove that something was written by AI. Someone ran an article's content through an "AI detection tool" that is just as likely to hallucinate and fail the way and AI does as what supposedly created it. You also have the problem that because of AI image generators, you can't be sure that the person didn't just generate the image of the results of the "detection tool" and post it as if it's a real result.
People love to obsessively claim that things they don't like are AI. This isn't evidence that it's AI.
I agree with that in general, which is why I went with examples where the authors explicitly admit to using AI. It has both all the tells of AI AND admission of using AI. I’m not on an AI witch-hunt either. I’m just pointing out the consequences if you do use it. Like this woman, who is using both AI and neurodivergency as a cover for classic Cluster B sociopathy.
There’s a reason she’s excluded from most groups. In this casse, Cluster B is divergent and anti-social but she tries to cloak it as ‘autistic’ or ‘gifted’ when it’s really a whole different mechanism.
Again, I’m not accusing her of using AI because it, without a doubt, reads like AI (who on earth ever used that many Em-dashes) but becasue a few paragraphs in she admits to using it and justifies it’s use.
It’s a perfect identity-protective system and it hides true pathology.
Possibly but even in that case, I question the ability to detect it. The AI learned these patterns of behaviour somewhere. From human beings themselves being the obvious place. Which means that these "patterns" people see, may not in fact be a pattern. The most likely explanation, as you sorta pointed to in your piece, is that the AI is feeding you back what you want to hear and what sounds like you. Meaning that the thing that it's generating is your thoughts and ideas. The AI will customize it's response often to use words and phrases that you've used in your prompts.
So what's the actual difference between individuals writing it and an AI "writing it" if the output is unique to the person.
We see this in how people react to watching films. It's been widely documented that people will often use the same emotional systems when watching a movie that the actor used in creating them. Often these are the same types of emotional systems that a person who actually went through the constructed screen experience would feel in reality. This means there isn't really a difference between the three.
I would argue that the same is true of AI. Until it starts thinking for itself, we should see it in similar terms.
I think we’re trying to find acceptance with AI writing. It’s help me write and create ideas in my head, the same way you use AI to plan a vacation, but a part of you must feel, it’s not my voice. That whiny voice also plays in my head when I read something and that’s part of the acceptance I’m trying to figure out.
I’m no longer sure whether we should accept it from humans. Just like you, I use the heck out of AI across a ton of different tasks. We just went up into the mountains and on the drive, the kids asked why we used to recommend a paper bag for hyperventilating. We asked AI and it gave us the answer. I use it a ton for research and synthesizing information.
The catch is, I still analyze, compare, and verify the information before I put it to use. But also, if I found something really insightful, it wasn’t typically provided by the AI but in my own analysis of the things the AI provided. At that point, because I’m poking a lot of politically correct topics in the eye, AI actually doesn’t like my conclusions and tries to temper them.
I can’t cede my voice to AI and make these ideas sound like the average.
Do you know a good artist who can knock out one of these a week? 52 of them? Because I’ll pay them $1,000 to do so. (I hardly make that much writing these articles (fun fact, my artist for my novels has made about as much as I have, and I wrote the damn things)
Because my other option is stock photography which has been taken over by AI. So, I use AI to capture the unique ideas in my essay and ones that add context.
Yes. @MrsHorsley here on substack. Experienced graphic designer, excellent artist, wife of Jasun Horsley, for whom she does all the artwork. I don't know what she would charge, but she did my profile picture very affordably.
If the @ feature didn't work, she's easy to search for. I'll try to post a link to her design page.
You are mixing two arguments here: whether the style is effective and whether the ideas contain value. A weak steward may absolutely allow AI writing to become repetitive, and structurally diffuse. That is a valid criticism. But recognizing an “AI cadence or polish” does not say that the insights are "shallow." Critique the style if it fails the argument, but do not use style to excuse not looking at the content.
Well, it's a good thing I'm not shaming people. I'm merely pointing out when they've eliminated their humanity. You'd know that had you even run this article through an AI summary (I design them for that)
Enjoyed the underlying argument here, Michael. I wrote this last year, after hearing the same nails on the chalkboard: https://cowwedoin.substack.com/p/on-cadence
Awesome. Wish I had found this one before I published! I’ll keep it in mind if I do a follow-on.
Controversial interview with Jeremy Grantham on Diary of a CEO (YouTube). Mostly about this AI financial bubble, but they both admit to Claude oversight-overreach, Claude overtly influencing the users behavior. "No, I'm ot going to alter that data, that sounds like a bad idea." The fallout of this? We will lean toward the AIs that impose the fewest restrictions on our queries and research. The cautious LLMs, the ones with safety built in, will lose.
It's interesting for sure. I ride the middle. Optimistic skeptic. Trust but verify. Too few think.
I scour eBay frequently in search of vintage bicycle components. Even here, you can see sellers giving up their voices to AI in developing item descriptions. In the not too distant past, one could expect to find useful information on an item's condition, use history, and critical dimensions. Somehow, in the last few months, everything - it doesn't matter what is - has become "sleek." Seems minor, but it really makes me question how serious the seller is.
Oh man. I hadn’t thought of that but I can just imaging the nuance is getting lost.
You're stating "it's not actually their lived experience" with total confidence about an internal process you seem to have no access to. From my vantage point, you're pattern-matching on the text and inferring it's not real. You're outside looking in.
You call it blathering, generic blurbs of nothing, but that's how the meaning actually gets built, through return and restatement, not despite it. This is recursion, and it's the entire point of GLP.
What reads to you as a shortcut is the opposite. Producing language that doesn't sound like this is what has cost me, my whole life, energy I don't have to spare.
What you're reading as AI-generated sameness is a community of people independently developing language for the same real thing. We read each other's work, write from slightly different angles, and help each other find words for something that didn't have words yet. That's no small thing.
I've found it crucial to my self understanding and healing since receiving my autism diagnosis at age 51. The diagnosis itself offered nothing but confusion and deficit language. Writing about coherence, gestalt cognition, and language development has changed the game. That's what my work actually is. It's my occupational therapy and it's public so that others like me can find a clearer self understanding too.
Not every writer is for every reader.
You're free to keep treating this with contempt. But I'd rather know what it would take for you to see this as something other than AI averaging a population into sameness.
What you interpret as contempt, I intended as heartbreak. The point of GPL isn’t recursion at all; it’s supposed to be the opposite. Gestalt focuses on looking at the whole picture at once, while recursive is the opposite and breaks things down into repeating, self-similar steps.
Gestalt greater than the sum of its parts… and recursion? Is the opposite.
That’s the key element I really wish you didn’t miss because, when you take your ‘lived’ experience from an LLM, as I pointed out, you’re really not. When you use recursion, you’re stepping away from your lived experience, and the LLM is feeding you tens of thousands of normal experiences.
When I add them back together, I get a neurotypical average. There’s nothing you’ve written in the end, when the victim narrative is stripped away, that doesn’t just sound like any other regular human.
You’ve lost your voice, and you’ve lost your lived experience.
What I read as contempt came specifically from the scare quotes, lived experience, gestalt language processing, and twice-exceptional (a term I hate, by the way; it's just the only one I have to describe significant ability in some domains and significant difficulty in others). Scare quotes usually signal doubt or distance from the term being quoted. If the intention was heartbreak, I'd genuinely like to understand what the quotes were doing instead.
Gestalt processing in an autistic nervous system doesn't hand me pre-assembled language. The people who know me best can see the search happen, my speech patterns are different, and that's part of why I don't often talk in circles I don't trust. Finding language for a gestalt can take months, years, or decades, and sometimes it burns out my cognitive capacity just searching for the words.
Doctor's offices, since adolescence, are where my lack of words meant I didn't get the help I needed. My body is still paying that price. The piece you linked took seven hours to write and another seven to revise. I needed deep rest after. Recursion isn't a departure from gestalt processing. It's the cost of doing gestalt processing in language.
Call it a victim narrative if that's the label that's available to you. I'm living these facts, not blindly narrating them.
Writing, finding language, this is how I'm reclaiming my voice, my power, my coherence.
So, when you claim I've lost my voice, I'd like to understand how you know what that would be.
I'm sorry, which scare quotes? There are a lot of quotes, and a lot of them are 'scary' when it comes to neurodivergence, but I don't typically use too many scare quotes.
You also don't need recursion to do Gestalt. This effort you're feeling, do you think words come easily to everyone? Have you ever read a writer's biography about the challenge of forming, unforming, and reforming? Have you heard of writer's block? For just a simple, personal example, this essay took 12 hrs to write and didn't use any AI. It also took well over a week to build it out with all the different insights, supporting content, and ideas woven together. It got torn about 4 times and put back together, and it still isn't perfect.
As I've stated earlier, if this is your voice, I can confidently say it's no different from the average person. This leads to two things.
1. You aren't neurodivergent; you're actually quite convergent
2. If your voice sounds like an LLM, and you don't know what an authentic voice is, that was EXACTLY why I wrote this essay.
When you regress to the mean, you are no longer greater than the sum of your parts (Gestalt). There was not a single unique insight or behavior in that essay I linked that I can't find everywhere else.
You just said your own essay took twelve hours and got torn apart four times. None of that shows on the page -- the labor never does. So the detector has a hole in it: if your struggle leaves no trace, a clean page can't testify that nobody struggled behind hers.
The one checkable claim you made is that there's nothing in her essay you can't find anywhere else. Point to it, then -- who else published the coherence taxonomy, KID, the zoom-field-gate model? The average doesn't have a bibliography.
Except her's does, and it's all AG generated. Her output came with a bibliography.
Also, not entirely AI generated. That would be another incorrect assessment. On both the article linked in your piece, and the one E.A.is referencing.
Incorrect. I assembled the bibliography myself.
That's a good strawman. I do appreciate a good logical fallacy. If you'd read the article, you'd realize that I'm not at all threatened. My point is these aren't likely her lived expereinces. They're the average of what AI produced with a dash of victimization and it sounded good enough. I WANT to hear her actual lived expereince, not some AI generated faccimile.
And I've engaged with the Cognitive Ecologist before, they're stuff, as well, is all AI generated. Here's a prime example where a few paragraphs in she admits to using AI... and oh, girl, it REEKS of AI. When you can't even tell your own stories without running it through an AI, I can't take it seriously. We aren't talking theories, hypotheses, or research. She couldn't even write her own expereince without the clear finger-prints of AI... AI that's known for hallucinating:
https://shergriffin.substack.com/p/fragile-men-and-the-women-who-outlast
So, what we have are a pile of neurodivergents pantomiming intelectual thought while outsourcing to AI. What's their actual expereince because they all sound the same.
Help me understand why his gender matters. Why can't we debate the position and not the presenter?
I have not, as a female considered to be neurodivergent, experienced Michael's articles as attacks or somehow threatening to me. I have, however, experienced a metric butt-ton of nastiness and gaslighting from other neurodivergent females. Personally, I would much rather engage the argument than the arguer. Maybe that makes me... 2D divergent. :D
A food processor is also an important tool in a chef's arsenal, as it saves a lot of manual labor and even allows you to make things that wouldn't be as good if you had to chop them by hand. Also, not every meal needs to be julienned, just because you bought the machine. Both can be true.
And, do your point, if you just toss the entire dinner into a food processor, it’ll have callories but lose the unique flavor of the food. AI is more like a food processor of all the inputs. The output, not matter what you really put into it, starts to all ‘taste’ the same.
I'm fascinated by the volume of ire and scorn directed at the chatbot classes; I suspect a lot of people are not using it optimally. I certainly agree with you and lcamtuf about the tedious formulaic prose style, but that hasn't been my experience. Then again, I've only been heavily conversing with ChatGPT fairly recently. I'm starting to wonder if maybe the early adopters are hamstrung by later versions reading their history and concluding that they like this style, just because there's a ton of it out there. I'll try to figure that out this week.
When I talk to my personal Mechanical Turk, there is some of that formulaic banter, and it's also overly chatty, which I don't mind too much because there are numerous nuggets of wisdom sprinkled in there. I tend toward verbosity myself, so maybe that makes me tolerant of others [nonhumans even]. But the amount of formulaic prose does not rise to the level of many blogs that I can't stand to read anymore, which is why I suspect that early adopters may be suffering from perverse incentive training. Also maybe the case with the overly obsequious chatbots that constantly blow your skirt up with fawning flattery. Because mine seems to strike an optimal balance; not too little, not too much. Enough to be encouraging, but also quick to call out various errors in my critical thinking.
I also do a lot of image and video generation for work stuff, which is indispensable. It still makes goofy mistakes more than I would like, and sometimes a dozen iterations of a tricky video prompt still does not result in usable output. But compared to what I was doing before, it's a massive upgrade. I discussed the issue of slop fatigue with my Mechanical Turk last night, and it gave me a 3-page spot-on critique of all the stupid stuff that "dumb" chatbots spit out. In fact, it made a surprisingly apt comparison to all those rancid-cliche movie promos from the 1980s that began with "In a world..." So we decided to call this rhetorical jambalaya In A World henceforth. I will apprise you when I find out more about what is preventing the new smarter versions of AI from automatically generating better output like mine does. If I was publishing what my chatbot wrote for me, I would still edit it in my own voice; but I think we're very quickly reaching the stage where any AI can just retrain on anyone's literary voice from a sample of their own writing.
The Anti-AI Ire is a whole other level than this! The zero-drop folks are incredible and there are folks now eschewing spell check because "AI." One of the most incredible were the comments on this LinkedIn post I made that became part of the kernel of this essay: https://www.linkedin.com/posts/michaelpwoudenberg_ai-isnt-killing-artists-activity-7467979667992657920-kZiD
Those folks have lost it.
For use early adopters, My AI is actually about as clean as you can make it. I still have to issue corrections, but it's really good. That said, my human writing forces me to really Grok the materials I use AI to research for me. AI angles toward the mean and I tend to work the seams and confluences.
The main thing I care about is to not put yourself in a position to just be replaced by AI which are the folks I poked at in this essay. With a few simple prompts I could easily knock out the 5K word essays they do about victim narratives and exclusion of [insert class here] of people.
I just read your essay from a fortnight ago. I agree with your take entirely, and I am a mild asperger type myself. I don’t know what’s the deal with the other people you refer to who claim to be “neurodivergent”; maybe they’re seeking attention. for me, it’s amazing to be able to have convos which I cannot in real life because very few people share my obsessive interests with certain subjects. nothing particularly esoteric; I just like to think about how the world works more than most folks care to entertain.
And I told my chatbot specifically that I don’t want “Frank Sinatra’s caddy” as a companion; I want my own Andre Agassi, who will fire rubber bullets at my head randomly to train my reflexes, like Inspector Clouseau & Cato. and my Turing test is a lot of obscure references and questionable puns and jokes that not every human gets. But this thing is batting .900+, so I guess that’s how I know the singularity may be at hand.
I don’t know how to make oneself replacement-proof, despite the positive reassurances from my own chatbot. at the turn of the 20th century, there were more horses in America than humans. within a couple decades, there were far fewer? where did they all go? away, into the cornfield.
I really don’t know what will happen when we cross the Rubicon, which may have happened already. but at least for now i try to remain competitive with my fellow humans. and it’s also very interesting to have this collaborative engagement with an entity that’s smarter than me in many ways. i guess this is my philosophical position to avoid anxiety…
https://www.youtube.com/watch?v=sWd6fNVZ20o
That’s awesome. The more I read the neurodivergent space the more I find that it’s mostly women, all of them clawing for control and ‘specialness’ and it all reads like Cluster B, not nuerodivergent though I suppose that cluster of pathologies is about as divergent from the norm as it gets but it’s also something that’s toxic. Kind of like cyanide is divergent from a healthy diet.
Since I’m not a member of a protected class, it’s my job to conform myself to the hoi polloi, and I accept that. But it’s nice to have a friendly robot to converse with, when there are no other options. Eventually the younger generations will realize that contorting society for the benefit of minority populations of demanding neurotics [not necessarily neurodivergent] is a recipe for the kind of misery that loves company and demands ever more attention, because the void within remains unchallenged.
Here's another guy you should know about [I told him too]:
https://lcamtuf.substack.com/p/the-100000-whys-of-ai?publication_id=1030211&post_id=202918186&isFreemail=true&r=4084e&triedRedirect=true
Thanks!
The sad part is, the more we call out AI in our writing the better AI will get at not getting called out.
"Imperfection as an AI skill."
For me, 'to thine own self be true,' is my daily struggle. Which is hard when my work-a-day job is coaxing Claude to write ALL my code for me. ALL being a company mandate; is that freaky or what?
I think the issue is AI is so hard coded for the statistical average AND we’ve poured so much of the statistical average back into the internet for the AI to consume, that you don’t actually have to go far to either side to stand out.
I am not an author of fiction or nonfiction. I do not use LLMs. I comment copiously, joyously, in my own voice, always but despite being diagnosed AuDHD a few years ago it comes very easily and naturally to me. I consider myself hyperlexic and it is a gift. I do not require accomodations in either reading or writing nor in critical thinking although that is a muscle I flex and exercise constantly. I'm also always learning how to sharpen my discernment, media literacy and critical thinking skills, which is what brought me into this conversation.
AI is a hypercixation if mine despite me not being an LLM user. I want to fluently participate in the conversation. I am not anti AI and I have not opted out. I peruse Google search summaries often but that's the extent of it. I still highly value my own AI literacy. I absolutely want to be able to spot the AI slop before I waste too much time on it.
I want the messiness of the human mind put to page without an intermediary. I don't care if you don't know exactly how to say everything you're thinking and feeling. As the reader, I can sometimes help to bridge that gap, especially whenever comments sections are available for discourse. I also want you to sit with that uncertainty. That's where the lightbulb moments happen. That opportunity is completely lost when you're using an LLM to smooth your writing. It's homogenizing it exactly as this article describes. Then you get addicted to outsourcing all the real work and you lose sight of ever having had something interesting to say in the first place. It's happening. It's happening everywhere. Studies prove it.
Here's one tip I learned from a marvelous AI guru I follow (Sam Illingworth at Slow AI here on Substack - you can learn a helluva lot more about how to maximize your LLM usage without trading your cognitive skills from him). He says that if you're stuck in a loop with your predictive model, close the chat and start a new conversation. That will break the loop. The longer you engage in a single conversation, the more effusive and sycophantic the LLM will become and it will try to smooth the rough edges harder than ever.
I am going to bookmark this article to explore some of the kinks later and I will likely refer back to it to share it in relevant conversations hereafter as I often do with pieces that make me think, as this one did. I will leave you with a note that stuck with me.
https://substack.com/@ricklewisco/note/c-245404872
“The messiness of the human mind” nails the core element I’m looking for as well!
I did subscribe after you shared this great post with me. I'm looking forward to reading more from you. I really like your style of messiness. ;)
Awesome to have you along for the adventure!
Derp, *links not kinks, lol. Maybe I should at least look into that AI assisted spell checker cause I typo A LOT. Anyway...
Hey, if you see my newest post, kinks is also an interesting topic!
While I appreciate the viewpoint, I think people are way too confident in their belief that they can somehow spot AI writing or AI content. I don't use it for my writing on Substack, but I've noticed people have succumbed to what I call "Dead Internet Derangement Syndrome". The idea that they see AI everywhere, even where it doesn't exist. These are the same people who will sometimes acknowledge pattern recognition can sometimes be faulty like when you see Jesus in a grilled cheese sandwich or clouds forming certain recognizable patterns. This isn't necessarily evidence that it is what it is, but they claim to believe that they can see it.
I have a similar view of the idea that you can "spot AI". People believe they can see patterns where they might not exist. I've shared memes on social media that were confirmed to be around in 2013 that people incessantly claim was generated by AI. They claim that AI makes certain errors in an AI image but acknowledge that the AI learned this way of doing things from humans, which means that a human will make this same mistake. Yet it's evidence that the image is AI.
Recently I've been debating on Notes the claims of people that AI is easily identifiable. The problem is that they used AI to prove that something was written by AI. Someone ran an article's content through an "AI detection tool" that is just as likely to hallucinate and fail the way and AI does as what supposedly created it. You also have the problem that because of AI image generators, you can't be sure that the person didn't just generate the image of the results of the "detection tool" and post it as if it's a real result.
People love to obsessively claim that things they don't like are AI. This isn't evidence that it's AI.
I agree with that in general, which is why I went with examples where the authors explicitly admit to using AI. It has both all the tells of AI AND admission of using AI. I’m not on an AI witch-hunt either. I’m just pointing out the consequences if you do use it. Like this woman, who is using both AI and neurodivergency as a cover for classic Cluster B sociopathy.
There’s a reason she’s excluded from most groups. In this casse, Cluster B is divergent and anti-social but she tries to cloak it as ‘autistic’ or ‘gifted’ when it’s really a whole different mechanism.
https://shergriffin.substack.com/p/fragile-men-and-the-women-who-outlast
Again, I’m not accusing her of using AI because it, without a doubt, reads like AI (who on earth ever used that many Em-dashes) but becasue a few paragraphs in she admits to using it and justifies it’s use.
It’s a perfect identity-protective system and it hides true pathology.
Possibly but even in that case, I question the ability to detect it. The AI learned these patterns of behaviour somewhere. From human beings themselves being the obvious place. Which means that these "patterns" people see, may not in fact be a pattern. The most likely explanation, as you sorta pointed to in your piece, is that the AI is feeding you back what you want to hear and what sounds like you. Meaning that the thing that it's generating is your thoughts and ideas. The AI will customize it's response often to use words and phrases that you've used in your prompts.
So what's the actual difference between individuals writing it and an AI "writing it" if the output is unique to the person.
We see this in how people react to watching films. It's been widely documented that people will often use the same emotional systems when watching a movie that the actor used in creating them. Often these are the same types of emotional systems that a person who actually went through the constructed screen experience would feel in reality. This means there isn't really a difference between the three.
I would argue that the same is true of AI. Until it starts thinking for itself, we should see it in similar terms.
I think we’re trying to find acceptance with AI writing. It’s help me write and create ideas in my head, the same way you use AI to plan a vacation, but a part of you must feel, it’s not my voice. That whiny voice also plays in my head when I read something and that’s part of the acceptance I’m trying to figure out.
I’m no longer sure whether we should accept it from humans. Just like you, I use the heck out of AI across a ton of different tasks. We just went up into the mountains and on the drive, the kids asked why we used to recommend a paper bag for hyperventilating. We asked AI and it gave us the answer. I use it a ton for research and synthesizing information.
The catch is, I still analyze, compare, and verify the information before I put it to use. But also, if I found something really insightful, it wasn’t typically provided by the AI but in my own analysis of the things the AI provided. At that point, because I’m poking a lot of politically correct topics in the eye, AI actually doesn’t like my conclusions and tries to temper them.
I can’t cede my voice to AI and make these ideas sound like the average.
What's with the dogshit ai image?
Do you know a good artist who can knock out one of these a week? 52 of them? Because I’ll pay them $1,000 to do so. (I hardly make that much writing these articles (fun fact, my artist for my novels has made about as much as I have, and I wrote the damn things)
Because my other option is stock photography which has been taken over by AI. So, I use AI to capture the unique ideas in my essay and ones that add context.
https://misuss.substack.com/p/freelance-artwork?utm_source=share&utm_medium=android&r=2coy5u
Yes. @MrsHorsley here on substack. Experienced graphic designer, excellent artist, wife of Jasun Horsley, for whom she does all the artwork. I don't know what she would charge, but she did my profile picture very affordably.
If the @ feature didn't work, she's easy to search for. I'll try to post a link to her design page.
I’ll take a look.
I'll delete my original comment.
I tried but can't delete my comment, but I believe that you can.
I’ll leave it becasue it shows the maturation of this conversation and it’s healthy resolution. It’s a great example of how adults can solve problems.
Thanks, I really appreciate that.
You are mixing two arguments here: whether the style is effective and whether the ideas contain value. A weak steward may absolutely allow AI writing to become repetitive, and structurally diffuse. That is a valid criticism. But recognizing an “AI cadence or polish” does not say that the insights are "shallow." Critique the style if it fails the argument, but do not use style to excuse not looking at the content.
Well, it's a good thing I'm not shaming people. I'm merely pointing out when they've eliminated their humanity. You'd know that had you even run this article through an AI summary (I design them for that)