Ethan Mollick on misunderstanding of LLMs; David Deming on being forced to create better assignments; Joanne Jacobs on using ChatGPT to write a high school essay; Sam Hammond projects the future
An office drone, whose job is preparing reports for other office drones, uses an LLM to write the reports in 5 minutes that used to take 8 hours. Who benefits economically? Of course, the office drone #1 benefits- he is earning the same salary as before for approximately 1/100th of the previous effort. What would such an office drone do with that extra free time at work? It likely would be detrimental for him to let his employer know that he was getting the same work done in 1% of the time as before as the employer/manager might well just lay him off and let the LLM do all the work.
What I am trying to figure out is where exactly LLMs will offer the greatest economic benefits/material wealth to the overall population generally in the near term (like next 20 years). I see medical diagnoses/care plans, legal work of all kinds, perhaps teaching, but I am having a hard time coming up with others. This is different than automation for production of real materials and goods.
I was thinking about the writers' strike in Hollywood yesterday or the day before- I can see LLMs putting those people out of work inside of 10 years, along with almost all "journalists" and novelists. Does ChatGTP have its own Substacks yet? What if Arnold or The Zvi is producing every single one of their posts these days using an automated LLM- how could we tell; do we benefit economically? What would the world do with a new novel written in the style of Charles Dickens every single day, or every report required by the Department of Labor written in a grand total of 1 workerday for the entire US corporate landscape?
It is impossible to exaggerate the truly incredible amount of low hanging fruit out there, which could have radically disruptive impacts on prices and employment in entire sectors, so long as someone figures out some creative way of circumventing the currently impenetrable armor of the only thing keeping the fruit on the trees: a robust system constructed out of rent-seeking relationships.
A lot of what the original disruption unleashed by new network tech was really the consequence of this kind of circumvention. E.g., uber. My prediction is that this is what the first wave of AI-based disruption is going to look like. When people say things like, "there's no good reasons but bad guild-boosting regs for why Americans pay so much more for ... " well, watch out.
I can definitely see LLMs putting a lot of people out of work, and that there might be a large net benefit in those people finding different productive work that wasn't possible to man previously, but what if it is just more make-work?
Just as it has tended to be across many different societies for many generations, make-work will be used as a shock-absorber for the slow-to-adjust labor impacts which are the consequence of an automation-explosion-based economic disruption which is happening more rapidly than the process of entrepreneurial discovery of opportunities for re-allocation.
Governments will throttle the legal (or quasi-legal) requirements for make-work up and down as necessary, they just need the most acceptable fairy tale for why those jobs are legitimate and needed and only """market failure""" and greed is why evil corporations are hiring below the socially optimal requirement. There is infinite demand for any service as the price goes to zero, but you really looking for a kind of service that can elastically and quickly absorb an almost unlimited influx of additional labor and the 'right' price of which is inherently hard to value. That's why my hunch is that "security" and "compliance" will be the services that take the brunt of the blast wave from the coming disruption. Some security jobs take a lot of talent and lengthy training. But not all. In Latin America I saw several places where it seemed as though every parking lot had at least one security-focused attendant for all open hours. The US has lots of parking lots. A legal requirement and/or a breakdown in security and order could keep millions employed, and people will all just tend not to question whether or not they are employed productively.
The legal profession was one I was looking at- legal documents are a necessary thing for an ownership society like ours. Lawyers are a necessary evil, but expensive. If LLMs could write valid legal documents with just a few input data, then a lot of value would be created by cutting out most of the legal profession altogether. If you could have an LLM represent you competently in court or help you defend yourself, even better.
Then my worries would be realized- we will be using LLMs to do today's make-work, and creating more make-work for the displaced human employees. Where is there going to be a net benefit to true productivity? That was what I was trying to figure out- where do the LLMs do work that was truly productive before the LLMs?
I don't disagree, but the point of the first part of my comment, perhaps too obscure, is what was the actual material benefit of the office drone's written reports before LLMs? It almost seems to me that the largest net benefits to all of us would be automating work that was never materially beneficial in the first place, in which case the savings derived from using LLMs is the equivalent of not having office drones do so much make work in the first place. If all we do is replace the office drones' work with LLMs and then assign the office drones more make work that can't be done yet by LLMs, where is the benefit?
Additionally, what if the response of the Department of Labor, for example, to companies suddenly using LLMs to automate all the required paperwork is to simply require new magnitude of paperwork? LLMs will be great for producing additional worthless paper pushing.
"Schools aren’t teaching students how to write in cursive any more. Is that such a loss?"
I could not disagree more! I used to be a teacher. Educational theorists coming up with the notion (1980's if I remember right) that ticking multiple choice tick boxes was a valid substitute for kids actually having to express their knowledge in sentence form. This must rank as one of the stupidest ideas in the history of schooling.
My thought about cursive comes from an interest in genealogy and family history. Without being able to read cursive, a lot of past information is closed off to you.
We may not be on the same page here ('scuse the pun). I took Arnold Kling's use of 'cursive' as referring to 'writing in sentence form' which I think was reasonable given the context (the "Writing clearly requires thinking clearly." in Joanne Jacob's piece). I was not referring to joined up handwriting (or even any kind of handwriting) which I imagine largely disappeared from most schools long ago with the advent of the digital age.
Being able to construct written sentences (whether handwritten or on a digital screen) is - as Joanne Jacobs rightly observes - fundamental to the business of learning to think. The fact that educational theory 'experts' sought to by pass it (which will have been all about 'equity'-massaging the difference between bright children and under achieving ones) is a measure of just how dismally thick-witted and unproductive the academic, sinecure-creating 'Educational Theory' racket has always been. End of rant.
Nevertheless your "Is that such a loss?" seemed to have been in response to Joanne Jacobs' "Writing clearly requires thinking clearly." I don't see how it could have been understood any other way.
Sam Hammond's arguments seem more influenced by Balaji Srinivasan's notion of a "network state" than either Kurzweil or Stephenson. Or maybe a more accurate statement is that Srinivasan's arguments are influenced in part from Kurzweil & Stephenson (among others), and Hammond ran with that (fanciful?) argument.
Arnold: “Schools aren’t teaching students how to write in cursive any more.” This sentence seems to prove that we can ditch any particular thing that schools teach now.
A hypothetical:
An office drone, whose job is preparing reports for other office drones, uses an LLM to write the reports in 5 minutes that used to take 8 hours. Who benefits economically? Of course, the office drone #1 benefits- he is earning the same salary as before for approximately 1/100th of the previous effort. What would such an office drone do with that extra free time at work? It likely would be detrimental for him to let his employer know that he was getting the same work done in 1% of the time as before as the employer/manager might well just lay him off and let the LLM do all the work.
What I am trying to figure out is where exactly LLMs will offer the greatest economic benefits/material wealth to the overall population generally in the near term (like next 20 years). I see medical diagnoses/care plans, legal work of all kinds, perhaps teaching, but I am having a hard time coming up with others. This is different than automation for production of real materials and goods.
I was thinking about the writers' strike in Hollywood yesterday or the day before- I can see LLMs putting those people out of work inside of 10 years, along with almost all "journalists" and novelists. Does ChatGTP have its own Substacks yet? What if Arnold or The Zvi is producing every single one of their posts these days using an automated LLM- how could we tell; do we benefit economically? What would the world do with a new novel written in the style of Charles Dickens every single day, or every report required by the Department of Labor written in a grand total of 1 workerday for the entire US corporate landscape?
It is impossible to exaggerate the truly incredible amount of low hanging fruit out there, which could have radically disruptive impacts on prices and employment in entire sectors, so long as someone figures out some creative way of circumventing the currently impenetrable armor of the only thing keeping the fruit on the trees: a robust system constructed out of rent-seeking relationships.
A lot of what the original disruption unleashed by new network tech was really the consequence of this kind of circumvention. E.g., uber. My prediction is that this is what the first wave of AI-based disruption is going to look like. When people say things like, "there's no good reasons but bad guild-boosting regs for why Americans pay so much more for ... " well, watch out.
I can definitely see LLMs putting a lot of people out of work, and that there might be a large net benefit in those people finding different productive work that wasn't possible to man previously, but what if it is just more make-work?
Just as it has tended to be across many different societies for many generations, make-work will be used as a shock-absorber for the slow-to-adjust labor impacts which are the consequence of an automation-explosion-based economic disruption which is happening more rapidly than the process of entrepreneurial discovery of opportunities for re-allocation.
Governments will throttle the legal (or quasi-legal) requirements for make-work up and down as necessary, they just need the most acceptable fairy tale for why those jobs are legitimate and needed and only """market failure""" and greed is why evil corporations are hiring below the socially optimal requirement. There is infinite demand for any service as the price goes to zero, but you really looking for a kind of service that can elastically and quickly absorb an almost unlimited influx of additional labor and the 'right' price of which is inherently hard to value. That's why my hunch is that "security" and "compliance" will be the services that take the brunt of the blast wave from the coming disruption. Some security jobs take a lot of talent and lengthy training. But not all. In Latin America I saw several places where it seemed as though every parking lot had at least one security-focused attendant for all open hours. The US has lots of parking lots. A legal requirement and/or a breakdown in security and order could keep millions employed, and people will all just tend not to question whether or not they are employed productively.
The legal profession was one I was looking at- legal documents are a necessary thing for an ownership society like ours. Lawyers are a necessary evil, but expensive. If LLMs could write valid legal documents with just a few input data, then a lot of value would be created by cutting out most of the legal profession altogether. If you could have an LLM represent you competently in court or help you defend yourself, even better.
Then my worries would be realized- we will be using LLMs to do today's make-work, and creating more make-work for the displaced human employees. Where is there going to be a net benefit to true productivity? That was what I was trying to figure out- where do the LLMs do work that was truly productive before the LLMs?
I don't disagree, but the point of the first part of my comment, perhaps too obscure, is what was the actual material benefit of the office drone's written reports before LLMs? It almost seems to me that the largest net benefits to all of us would be automating work that was never materially beneficial in the first place, in which case the savings derived from using LLMs is the equivalent of not having office drones do so much make work in the first place. If all we do is replace the office drones' work with LLMs and then assign the office drones more make work that can't be done yet by LLMs, where is the benefit?
Additionally, what if the response of the Department of Labor, for example, to companies suddenly using LLMs to automate all the required paperwork is to simply require new magnitude of paperwork? LLMs will be great for producing additional worthless paper pushing.
"Schools aren’t teaching students how to write in cursive any more. Is that such a loss?"
I could not disagree more! I used to be a teacher. Educational theorists coming up with the notion (1980's if I remember right) that ticking multiple choice tick boxes was a valid substitute for kids actually having to express their knowledge in sentence form. This must rank as one of the stupidest ideas in the history of schooling.
My thought about cursive comes from an interest in genealogy and family history. Without being able to read cursive, a lot of past information is closed off to you.
We may not be on the same page here ('scuse the pun). I took Arnold Kling's use of 'cursive' as referring to 'writing in sentence form' which I think was reasonable given the context (the "Writing clearly requires thinking clearly." in Joanne Jacob's piece). I was not referring to joined up handwriting (or even any kind of handwriting) which I imagine largely disappeared from most schools long ago with the advent of the digital age.
Being able to construct written sentences (whether handwritten or on a digital screen) is - as Joanne Jacobs rightly observes - fundamental to the business of learning to think. The fact that educational theory 'experts' sought to by pass it (which will have been all about 'equity'-massaging the difference between bright children and under achieving ones) is a measure of just how dismally thick-witted and unproductive the academic, sinecure-creating 'Educational Theory' racket has always been. End of rant.
I meant cursive as a form of handwriting.
Nevertheless your "Is that such a loss?" seemed to have been in response to Joanne Jacobs' "Writing clearly requires thinking clearly." I don't see how it could have been understood any other way.
Sam Hammond's arguments seem more influenced by Balaji Srinivasan's notion of a "network state" than either Kurzweil or Stephenson. Or maybe a more accurate statement is that Srinivasan's arguments are influenced in part from Kurzweil & Stephenson (among others), and Hammond ran with that (fanciful?) argument.
Arnold: “Schools aren’t teaching students how to write in cursive any more.” This sentence seems to prove that we can ditch any particular thing that schools teach now.