I like so much of this. And agree with the overall argument. But a couple of quibbles, in good faith.
1) I don't think any future ever "belongs to anyone." There may be winners and losers, but no one owns the future. And everyone, even the invading genocide wielding settlers, struggles.
2) You suggest: "The question isn't whether AI will change everything. It already is." Already is changing everything? People's long term relationships seem to be not changing at all, at least among those I know. And for many of us, our long term relationships are the most important part of life.
Most jobs haven't much changed either, and those that require handy hands aren't likely to change for quite some time (fix your fridge? trim your trees?). The lives of let's say about 3 billion people on Earth are not much changing.
So yes, we who read you and write here are changing mightily, I agree, and we suspect we are in the vanguard. But most of the people I know don't really want me to talk about this stuff with them partly because they have not seen any substantial changes in their lives as yet. And some are resisting what changes seem to be offered to them, in principle, as they detest AI. (e.g. my 22 year old daughter!)
I guess I'm just suggesting we should slow down in our use of "everyone," "we" and "all" until we make sure we're not just talking about privileged Global North and educated people with leisure time to read and write about this stuff.
And, as I said, these are quibbles only! I endorse your exhortations for us in this discourse world.
I think what I was reaching for (and failed to articulate clearly) is that change is happening in the infrastructure layer that will eventually touch most lives, even if indirectly. Like how most people in 1995 weren't thinking about the internet, but the decisions being made then shaped the world their kids grew up in. But you're right - that's very different from "changing everything."
The "belongs to" phrasing was sloppy. Nobody owns tomorrow. I meant something more like "those who adapt consciously have more agency in shaping their own path through whatever comes." But even that assumes the privilege of having choices to make.
Ouch--I think you are right that my rambling was really just "precision" checking! I felt uneasy as I was writing it, and that is why. I now understand; I didn't understand why I was uneasy as I wrote.
I come from an academic background, decades of teaching college and writing monographs and textbooks. I am trying to learn how to blog effectively and, in general, I think impact is more important than precision in this context. And the impact of what you were saying was what I appreciated. And applaud.
I think I was out of line, and I appreciate your graciousness in not directly calling me out on it. I cheer you on in the work you doing as I very much agree that the decisions being made are gravely consequential. My whole aim on Substack is to argue these decisions will be more consequential for our inner than for our outer lives, and we need to prepare.
This had a beautiful grace to it. The pacing was wonderful. As someone who saw that example of change in 1995, while watching its impacts and working on the preparations for its infrastructural influences, I too hear the echoes of that decade. And I support your idea that the changes are indirect.
The federal workers laid off here in Virginia and Maryland will not get their jobs back based upon a need for them in their functions. Unfortunately, their jobs can be replaced by better AI that exists this year currently. The river flows, and even if you try to return to the shore you left... you can't reach the same shore twice.
yes, and then, yes again and then yes again and then yes again… All the way to the end and your choice of image to close this post with: all of those OLD chairs.
Thanks for the well-written and thought-provoking piece. I've been teaching coding, data science, AI, and math for years, and I haven't actually noticed anything really changing with modern LLMs. The only major change is that I hear everyone talking about how much things will change, but I haven't actually personally witnessed much change, at least not yet (excluding us witnessing the dead internet theory coming to fruition). People still seek out expert education, they still learn to code, and they still have jobs. Maybe there really will be a revolution for better or for worse, but all I've seen is 2 years of hype that seems to be finally cooling off.
Thanks for sharing your perspective, Mike. Your experience teaching these subjects is valuable, and I think you’re capturing something important - the gap between discourse and daily reality. You’re right that the fundamentals haven’t changed much. Students still need to understand logic, still struggle with pointers, still need human guidance. But I’d suggest the changes are happening in spaces you might not be watching closely.
Your students are likely using LLMs extensively outside your classroom - for debugging, for understanding concepts, for completing assignments. They’re just not telling you about it. The “dead internet theory” you mention is partly about this shadow usage that’s invisible to instructors.
The fact that “people still have jobs” is true but masks significant shifts. Junior developers are finding fewer entry points. Technical writers are seeing roles evaporate. The changes aren’t evenly distributed, which makes them easy to miss from certain vantage points.
I find it interesting that you frame this as “hype finally cooling off” when usage metrics show the opposite - acceleration and integration becoming so normal it’s invisible. That’s not cooling off; that’s absorption into the everyday. The revolution isn’t coming. It’s here. It’s just unevenly distributed and easy to miss if you’re looking for dramatic moments rather than gradual shifts in how work gets done.
But genuinely curious - what would constitute “real change” in your view? What would you need to see to think something fundamental had shifted?
Thanks for the reply! My students definitely use LLMs, and I encourage that. I also now include LLMs in my courses, e.g., prompting to get help with debugging and writing boiler-plate code. I guess what I meant is the hype about how people don't need to learn coding or math, because LLMs will do those instead. I still can't get ChatGPT to write matplotlib code, or solve my latex issues. And it makes obvious rookie mistakes when I ask it to simulate a signal and calculate a power spectrum.
What I see (again, just my anecdotal experience) is people using LLMs to do things they were already doing. It's just another tool. I'd consider it revolutionary when it makes qualitative changes, not quantitative. Maybe it is already here but unevenly distributed and hasn't hit me yet (though I consider myself in the field, e.g., a 90-hour course on LLM mechanisms and interpretability).
As for shifts in the job market: Sure, but job markets and the economy fluctuate regularly. There were no LLMs in the dot-com bubble, or in the 2008/9 crash, etc. I think it's still too soon to separate hype and unrelated fluctuations from long-term consequences. Society is, after all, a highly complex dynamical system, and causality is difficult to establish on short time scales.
I couldn't agree more that it's just another tool! After reading this very thoughtful and well written article and the extremely well thought replies by two thinkers in position of teaching, allow me to add a couple of anecdotes from working-in-a-potentially-repleaceable-job person in her late 40's (meaning, I was there in 1995 and saw it happening), who has daily dealings with LLM.
My career started in those mid 90's, I started by taking dictations and typing in multiple copies, one year later I was editing mistakes on a screen. First one computer in the whole office, eventually everyone with one on their desks, soon taking them even home in a bag. We all know the rest of that history.
Now, LLM... helps with repetitive tasks. Helps me draft an email when typing it all out may take 30 minutes, with LLM (previously prompted and well honed into something close to perfection) can spit out the first draft in seconds, then some more prompting, adjustments and voilà. Next one is smoother.
But, and here's the pitfall, relying on it to come out with anything creative and unique is simply silly. It can't. It's using strings of words others came up with or you fed it.
And it lies. Just lies.
Situation: Fed it a large (27 pages) doc, instructed it to extract the text (the doc is an instructional guide with images) and translate the text. "Sure, I can do that", when initially questioned on the details of the task. "I'll just need extra time, probably 40 minutes". 2 hrs later, where's the work? Lots of BS, plenty of running in circles, no result. Even step by step. Even to the level of just feeding it the manually extracted text, couldn't handle slightly over 2000 words of translation. So, tool. Very limited tool. And replacing technical writers? Hahaha, that's who I am. It can summarise, sure. It can make something sound better when my SME and I get caught in technicalities and detail. But it cannot create. It will become better, sure. But it will remain a machine and for a long while, an inadequate one. (And let's not even go into the issue of cognitive bias.)
And btw, those bloggers who use LLMs for their writing are easily discernible from very evident tell-tell signs. (Not saying the author did.)
This article is fundamentally flawed. It adheres to the dogmatic belief that all innovation is a supernatural force that is inherently good and cannot be resisted. That progress is a linear path where reversal is not only unthinkable, but impossible.
You speak of the very real fears that people have for AI. That they will lose their sense of purpose, their identities, their meaning to their own lives. That AI will reshape everything. All of this is true and yet you never consider whether these changes are beneficial. The answer is a forgone conclusion, or perhaps irrelevant. It doesn’t matter if this benefits people, it’s going to happen anyway.
Instead people have to learn to live with a constant sense of uncertainty. This is utter madness and drivel. For what end are people living in uncertainty? It acts as if the desire to have a stable life is wrong. What is the point of innovation if not to improve upon our lives?
There is no discussion about what kinds of effects this will have on society. There is no weighing the pros and the cons. No philosophical discussion about what it means to be human. No consensus on whether or not people actually want their lives to be reshaped or if they’re content where they are. Instead the drums of innovation beat on.
Innovation becomes the end in and of itself. We need to innovate because we need to innovate, and if millions of people suffer, then so be it. That’s the price that needs to be paid. It’s absurd.
The truth is nothing is stopping us from dismantling all AI infrastructure and moving on. It isn’t some unstoppable force. We can simply just say no. Of course, that isn’t going to happen. Humanity is mostly an unorganized mass that has no ability to force its will upon the world. Instead the highly centralized elites who wield immense power will force AI upon the rest of us and use it solely to enrich themselves. It’s already happening.
“In a world of infinite digital imagery and audio content generation, the scarcest resource isn't information or even intelligence,” it’s the ability to know how to work with one’s hands to make and maintain physical things—and the insight to recognize what to do and when.
The problem with the adapt or die mantra is that AI will adapt faster than we ever could. The AI bears list all of its limitations, but ChatGPT launched just three short years ago.
That degree you worked hard on for years? In another 2 or 3 will it be totally irrelevant? Maybe get a law degree…nevermind the newest LLM can seamlessly consume and learn all legal documents ever published in any state. A medical degree? Took a month to train a model what would take you years.
This isn’t an opportunity to reinvent oneself, it’s like trying to outrun a shotgun blast. Maybe you guess right, dodge the right way, and buy yourself a couple more years of employment, but we’re entering an era of universal insecurity.
I like so much of this. And agree with the overall argument. But a couple of quibbles, in good faith.
1) I don't think any future ever "belongs to anyone." There may be winners and losers, but no one owns the future. And everyone, even the invading genocide wielding settlers, struggles.
2) You suggest: "The question isn't whether AI will change everything. It already is." Already is changing everything? People's long term relationships seem to be not changing at all, at least among those I know. And for many of us, our long term relationships are the most important part of life.
Most jobs haven't much changed either, and those that require handy hands aren't likely to change for quite some time (fix your fridge? trim your trees?). The lives of let's say about 3 billion people on Earth are not much changing.
So yes, we who read you and write here are changing mightily, I agree, and we suspect we are in the vanguard. But most of the people I know don't really want me to talk about this stuff with them partly because they have not seen any substantial changes in their lives as yet. And some are resisting what changes seem to be offered to them, in principle, as they detest AI. (e.g. my 22 year old daughter!)
I guess I'm just suggesting we should slow down in our use of "everyone," "we" and "all" until we make sure we're not just talking about privileged Global North and educated people with leisure time to read and write about this stuff.
And, as I said, these are quibbles only! I endorse your exhortations for us in this discourse world.
I think what I was reaching for (and failed to articulate clearly) is that change is happening in the infrastructure layer that will eventually touch most lives, even if indirectly. Like how most people in 1995 weren't thinking about the internet, but the decisions being made then shaped the world their kids grew up in. But you're right - that's very different from "changing everything."
The "belongs to" phrasing was sloppy. Nobody owns tomorrow. I meant something more like "those who adapt consciously have more agency in shaping their own path through whatever comes." But even that assumes the privilege of having choices to make.
Thanks for the precision check.
Ouch--I think you are right that my rambling was really just "precision" checking! I felt uneasy as I was writing it, and that is why. I now understand; I didn't understand why I was uneasy as I wrote.
I come from an academic background, decades of teaching college and writing monographs and textbooks. I am trying to learn how to blog effectively and, in general, I think impact is more important than precision in this context. And the impact of what you were saying was what I appreciated. And applaud.
I think I was out of line, and I appreciate your graciousness in not directly calling me out on it. I cheer you on in the work you doing as I very much agree that the decisions being made are gravely consequential. My whole aim on Substack is to argue these decisions will be more consequential for our inner than for our outer lives, and we need to prepare.
Write on.
No need to worry - I appreciated it - i do enjoy the freedom of writing here being somewhat ‘looser’
This had a beautiful grace to it. The pacing was wonderful. As someone who saw that example of change in 1995, while watching its impacts and working on the preparations for its infrastructural influences, I too hear the echoes of that decade. And I support your idea that the changes are indirect.
The federal workers laid off here in Virginia and Maryland will not get their jobs back based upon a need for them in their functions. Unfortunately, their jobs can be replaced by better AI that exists this year currently. The river flows, and even if you try to return to the shore you left... you can't reach the same shore twice.
“..Learning to be comfortable with permanent uncertainty..” I love this! Absolutely!
Loved reading it
Amazing read 👏🏻
yes, and then, yes again and then yes again and then yes again… All the way to the end and your choice of image to close this post with: all of those OLD chairs.
Chairs that don’t fit anymore.
Thanks for the well-written and thought-provoking piece. I've been teaching coding, data science, AI, and math for years, and I haven't actually noticed anything really changing with modern LLMs. The only major change is that I hear everyone talking about how much things will change, but I haven't actually personally witnessed much change, at least not yet (excluding us witnessing the dead internet theory coming to fruition). People still seek out expert education, they still learn to code, and they still have jobs. Maybe there really will be a revolution for better or for worse, but all I've seen is 2 years of hype that seems to be finally cooling off.
Thanks for sharing your perspective, Mike. Your experience teaching these subjects is valuable, and I think you’re capturing something important - the gap between discourse and daily reality. You’re right that the fundamentals haven’t changed much. Students still need to understand logic, still struggle with pointers, still need human guidance. But I’d suggest the changes are happening in spaces you might not be watching closely.
Your students are likely using LLMs extensively outside your classroom - for debugging, for understanding concepts, for completing assignments. They’re just not telling you about it. The “dead internet theory” you mention is partly about this shadow usage that’s invisible to instructors.
The fact that “people still have jobs” is true but masks significant shifts. Junior developers are finding fewer entry points. Technical writers are seeing roles evaporate. The changes aren’t evenly distributed, which makes them easy to miss from certain vantage points.
I find it interesting that you frame this as “hype finally cooling off” when usage metrics show the opposite - acceleration and integration becoming so normal it’s invisible. That’s not cooling off; that’s absorption into the everyday. The revolution isn’t coming. It’s here. It’s just unevenly distributed and easy to miss if you’re looking for dramatic moments rather than gradual shifts in how work gets done.
But genuinely curious - what would constitute “real change” in your view? What would you need to see to think something fundamental had shifted?
Thanks for the reply! My students definitely use LLMs, and I encourage that. I also now include LLMs in my courses, e.g., prompting to get help with debugging and writing boiler-plate code. I guess what I meant is the hype about how people don't need to learn coding or math, because LLMs will do those instead. I still can't get ChatGPT to write matplotlib code, or solve my latex issues. And it makes obvious rookie mistakes when I ask it to simulate a signal and calculate a power spectrum.
What I see (again, just my anecdotal experience) is people using LLMs to do things they were already doing. It's just another tool. I'd consider it revolutionary when it makes qualitative changes, not quantitative. Maybe it is already here but unevenly distributed and hasn't hit me yet (though I consider myself in the field, e.g., a 90-hour course on LLM mechanisms and interpretability).
As for shifts in the job market: Sure, but job markets and the economy fluctuate regularly. There were no LLMs in the dot-com bubble, or in the 2008/9 crash, etc. I think it's still too soon to separate hype and unrelated fluctuations from long-term consequences. Society is, after all, a highly complex dynamical system, and causality is difficult to establish on short time scales.
I couldn't agree more that it's just another tool! After reading this very thoughtful and well written article and the extremely well thought replies by two thinkers in position of teaching, allow me to add a couple of anecdotes from working-in-a-potentially-repleaceable-job person in her late 40's (meaning, I was there in 1995 and saw it happening), who has daily dealings with LLM.
My career started in those mid 90's, I started by taking dictations and typing in multiple copies, one year later I was editing mistakes on a screen. First one computer in the whole office, eventually everyone with one on their desks, soon taking them even home in a bag. We all know the rest of that history.
Now, LLM... helps with repetitive tasks. Helps me draft an email when typing it all out may take 30 minutes, with LLM (previously prompted and well honed into something close to perfection) can spit out the first draft in seconds, then some more prompting, adjustments and voilà. Next one is smoother.
But, and here's the pitfall, relying on it to come out with anything creative and unique is simply silly. It can't. It's using strings of words others came up with or you fed it.
And it lies. Just lies.
Situation: Fed it a large (27 pages) doc, instructed it to extract the text (the doc is an instructional guide with images) and translate the text. "Sure, I can do that", when initially questioned on the details of the task. "I'll just need extra time, probably 40 minutes". 2 hrs later, where's the work? Lots of BS, plenty of running in circles, no result. Even step by step. Even to the level of just feeding it the manually extracted text, couldn't handle slightly over 2000 words of translation. So, tool. Very limited tool. And replacing technical writers? Hahaha, that's who I am. It can summarise, sure. It can make something sound better when my SME and I get caught in technicalities and detail. But it cannot create. It will become better, sure. But it will remain a machine and for a long while, an inadequate one. (And let's not even go into the issue of cognitive bias.)
And btw, those bloggers who use LLMs for their writing are easily discernible from very evident tell-tell signs. (Not saying the author did.)
Apologies for long reply, couldn't help myself.
This article is fundamentally flawed. It adheres to the dogmatic belief that all innovation is a supernatural force that is inherently good and cannot be resisted. That progress is a linear path where reversal is not only unthinkable, but impossible.
You speak of the very real fears that people have for AI. That they will lose their sense of purpose, their identities, their meaning to their own lives. That AI will reshape everything. All of this is true and yet you never consider whether these changes are beneficial. The answer is a forgone conclusion, or perhaps irrelevant. It doesn’t matter if this benefits people, it’s going to happen anyway.
Instead people have to learn to live with a constant sense of uncertainty. This is utter madness and drivel. For what end are people living in uncertainty? It acts as if the desire to have a stable life is wrong. What is the point of innovation if not to improve upon our lives?
There is no discussion about what kinds of effects this will have on society. There is no weighing the pros and the cons. No philosophical discussion about what it means to be human. No consensus on whether or not people actually want their lives to be reshaped or if they’re content where they are. Instead the drums of innovation beat on.
Innovation becomes the end in and of itself. We need to innovate because we need to innovate, and if millions of people suffer, then so be it. That’s the price that needs to be paid. It’s absurd.
The truth is nothing is stopping us from dismantling all AI infrastructure and moving on. It isn’t some unstoppable force. We can simply just say no. Of course, that isn’t going to happen. Humanity is mostly an unorganized mass that has no ability to force its will upon the world. Instead the highly centralized elites who wield immense power will force AI upon the rest of us and use it solely to enrich themselves. It’s already happening.
Constant uncertainty is a definition of anxiety. Who wants that?! And to what end?
(I am agreeing with you)
Still sounds like a thinkpiece pretending ai isn't changing everything.
I like much of your argument, but reading GPT writing is exhausting and emptying.
“In a world of infinite digital imagery and audio content generation, the scarcest resource isn't information or even intelligence,” it’s the ability to know how to work with one’s hands to make and maintain physical things—and the insight to recognize what to do and when.
The problem with the adapt or die mantra is that AI will adapt faster than we ever could. The AI bears list all of its limitations, but ChatGPT launched just three short years ago.
That degree you worked hard on for years? In another 2 or 3 will it be totally irrelevant? Maybe get a law degree…nevermind the newest LLM can seamlessly consume and learn all legal documents ever published in any state. A medical degree? Took a month to train a model what would take you years.
This isn’t an opportunity to reinvent oneself, it’s like trying to outrun a shotgun blast. Maybe you guess right, dodge the right way, and buy yourself a couple more years of employment, but we’re entering an era of universal insecurity.
Interesting read just wondering to what extent did you use an LLM to write this..
Generative AI is not doing any of these things:
The creative destruction of the next 10 plus years is going to be unbelievable and unbearable. Good luck everyone 😅
Have you posted this on LinkedIn? Please cross-post of this.