This is sort of why I think the publishing world needs to catch up with the times finally. I've had a vision called the Now Edition for some time. Why do we call things ebooks when they are just replicas of dead tree books stuck in an old fashioned publishing system? Why aren't they as fresh as today and as deep as the best literature, with social interaction with the authors and each other one click away? I guess the logistics, legalities and monetization are challenges but certainly not unable to be overcome. I even mocked one up in Floot.
What i'm waiting on is digital paper to the level i could grab a digital book off my shelf and take it with me on a plane. It would connect on its own, do what you're talking about, but also permit me to write on it like i'd write in a book with my pen. Fold it up and put it in my bag, comes hardback or paperback.
Oh a lot of us have given up on digital paper, but maybe it will eventually finally happen, just like in the movie Minority report, the subway scene where the USA Today headline was changing as the person perused their paper. But the foundation of the now Edition could work with that as well.
I helped build an AI company, and the strangest thing about this acceleration is how little it has clarified what I actually want to make. The machine can now turn almost any passing idea into a plausible business, film, or future. I've begun to suspect that the real scarcity isn't intelligence anymore as much as it's simply CONVICTION. Why is the willingness to choose one possibility and let the others die so difficult?
Congratulations on the book, and running small agent teams every day has taught me that the control comes back in knowing which results to trust and which to actually read.
I've been doing work with AI. As an experiment, I've been vibe-coding both a Baldur's Gate 3 mod (with some difficulties, as I'm translating a 5.5e homebrew class into a 5e-based system) and seeing if it can cook up an SNES shmup with me doing design work. Not SNES style, to be clear, it currently runs on an emulator and could in theory be written to a cartridge one would put into 35 year old hardware. (I wanted to see how well the AI could work under stricter hardware constraints than most of the modern projects it would have learned from.)
So I've got some idea of what these tools can do. I've built something in spare hours that would have taken a 90s dev team months on the nine-to-five. I can decide to make massive changes and it just happens. It's annoyingly "lazy", I find, but when I am clear in my commands, it executes. Frankly: AI is the biggest thing to happen to programming and video gaming alike in decades.
But what's this exponential really mean for the rest of humanity? When's it gonna directly affect, e.g., the price of soybeans? That's what's got me really wondering.
The efficiency of the system is amazing but the economics are risky. One of our senior economists has compared the investment in AI to that of the railway industry during the 19th century which finally bankrupted itself through stupidity and the belief that it was infallible. You mention over 100 systems, all with databases of the same data using Large Language modules. Is it not time for collaboration, merging, agreements of areas of interest. There is a need for control, for preventing one AI system penetrating other AI systems, either accidentally or because it has been given a task and has no moral sense to restrict itself. The British might get a codified constitution out of this. Currently AI is scaring the pants off mathematicians by solving maths problems that have existed since maths was defined. Then there are the Chinese who, having had an advanced civilisation when the West lived in mud huts and used butter to control their hair, want to exert dominance in their sphere at the present time, and maybe world rule eventually. How are we going to control their AI systems? It took decades to develop Firewalls for the internet.
I think there's a good reason why the job apocalypse has not yet arrived and is probably a long way into the future. But I don't think the lack of AI capability is why. There's just too many human and organizational obstacles to leveraging the full capability. The AI companies claim to be massively leveraging it already, at least for coding, but it's hard to know how much trust to put in their claims because of their vested interests on the other hand, there are definitely individual entrepreneurs who appear to be making massive gains in productivity in their own solo or very small businesses. For example, I take Nate Silver at his word that the use of AI has dramatically altered his workflow.
That being said, I do not see a job apocalypse on the horizon, because the friction to adoption leading to significant gains in revenue are very high and are likely going to take a very long time to overcome. However, it may turn out in the interim that smaller startup companies that are able to build AI into their workflows from the ground up without having all of the existing human and organizational obstacles may end up displacing many companies that have no way of moving any faster than they currently are. And I think obviously companies that are able to displace existing organizations will be limited for the foreseeable future to industries where the probabilistic output of LLMs is acceptable. Some industries do require close to perfection. However, many other industries might prefer perfection, but they generally don't get anywhere near it from their humans.
I worked in the area of instructional design and technology for online education for many years before I retired and moved to consulting work. I am completely convinced at this point that current LLMs can do equal or better work than most of the humans I ever had working under me in this area. I have been able to complete jobs on my own that formerly would've required a full-time team of four or five people. Part of that team might be doing the same or similar work to me, but other members would've been doing specialized work that I was incapable of doing myself now, I can task an LLM to complete certain technical aspects of the job that were beyond my skill set. Historically, I had to spend a lot of time communicating my needs as well as project managing a completion of certain tasks. Now, I can spend less time communicating and project managing to produce an equivalent product.
Pre ordered the ebook (which for some reason is not available today but in two days in The Netherlands)
This is sort of why I think the publishing world needs to catch up with the times finally. I've had a vision called the Now Edition for some time. Why do we call things ebooks when they are just replicas of dead tree books stuck in an old fashioned publishing system? Why aren't they as fresh as today and as deep as the best literature, with social interaction with the authors and each other one click away? I guess the logistics, legalities and monetization are challenges but certainly not unable to be overcome. I even mocked one up in Floot.
Oh, and here it is - no plumbing, but look and feel. https://thenowedition.floot.app/
What i'm waiting on is digital paper to the level i could grab a digital book off my shelf and take it with me on a plane. It would connect on its own, do what you're talking about, but also permit me to write on it like i'd write in a book with my pen. Fold it up and put it in my bag, comes hardback or paperback.
Oh a lot of us have given up on digital paper, but maybe it will eventually finally happen, just like in the movie Minority report, the subway scene where the USA Today headline was changing as the person perused their paper. But the foundation of the now Edition could work with that as well.
I helped build an AI company, and the strangest thing about this acceleration is how little it has clarified what I actually want to make. The machine can now turn almost any passing idea into a plausible business, film, or future. I've begun to suspect that the real scarcity isn't intelligence anymore as much as it's simply CONVICTION. Why is the willingness to choose one possibility and let the others die so difficult?
This distillation is stunning.
Congratulations on the book, and running small agent teams every day has taught me that the control comes back in knowing which results to trust and which to actually read.
Thx Ethan. Definitely gonna up the ante with opus. Your otter video blows past the simple explainers I’ve been asking it to make…
Bought the Kindle this morning!
I've been doing work with AI. As an experiment, I've been vibe-coding both a Baldur's Gate 3 mod (with some difficulties, as I'm translating a 5.5e homebrew class into a 5e-based system) and seeing if it can cook up an SNES shmup with me doing design work. Not SNES style, to be clear, it currently runs on an emulator and could in theory be written to a cartridge one would put into 35 year old hardware. (I wanted to see how well the AI could work under stricter hardware constraints than most of the modern projects it would have learned from.)
So I've got some idea of what these tools can do. I've built something in spare hours that would have taken a 90s dev team months on the nine-to-five. I can decide to make massive changes and it just happens. It's annoyingly "lazy", I find, but when I am clear in my commands, it executes. Frankly: AI is the biggest thing to happen to programming and video gaming alike in decades.
But what's this exponential really mean for the rest of humanity? When's it gonna directly affect, e.g., the price of soybeans? That's what's got me really wondering.
Right. Makes you think Cory Doctorow has something in his critique.
The efficiency of the system is amazing but the economics are risky. One of our senior economists has compared the investment in AI to that of the railway industry during the 19th century which finally bankrupted itself through stupidity and the belief that it was infallible. You mention over 100 systems, all with databases of the same data using Large Language modules. Is it not time for collaboration, merging, agreements of areas of interest. There is a need for control, for preventing one AI system penetrating other AI systems, either accidentally or because it has been given a task and has no moral sense to restrict itself. The British might get a codified constitution out of this. Currently AI is scaring the pants off mathematicians by solving maths problems that have existed since maths was defined. Then there are the Chinese who, having had an advanced civilisation when the West lived in mud huts and used butter to control their hair, want to exert dominance in their sphere at the present time, and maybe world rule eventually. How are we going to control their AI systems? It took decades to develop Firewalls for the internet.
I wish I had had this article to cite when I wrote my essay about future shock a couple weeks ago.
https://grantwp.substack.com/p/future-shock-is-here
I think there's a good reason why the job apocalypse has not yet arrived and is probably a long way into the future. But I don't think the lack of AI capability is why. There's just too many human and organizational obstacles to leveraging the full capability. The AI companies claim to be massively leveraging it already, at least for coding, but it's hard to know how much trust to put in their claims because of their vested interests on the other hand, there are definitely individual entrepreneurs who appear to be making massive gains in productivity in their own solo or very small businesses. For example, I take Nate Silver at his word that the use of AI has dramatically altered his workflow.
That being said, I do not see a job apocalypse on the horizon, because the friction to adoption leading to significant gains in revenue are very high and are likely going to take a very long time to overcome. However, it may turn out in the interim that smaller startup companies that are able to build AI into their workflows from the ground up without having all of the existing human and organizational obstacles may end up displacing many companies that have no way of moving any faster than they currently are. And I think obviously companies that are able to displace existing organizations will be limited for the foreseeable future to industries where the probabilistic output of LLMs is acceptable. Some industries do require close to perfection. However, many other industries might prefer perfection, but they generally don't get anywhere near it from their humans.
I worked in the area of instructional design and technology for online education for many years before I retired and moved to consulting work. I am completely convinced at this point that current LLMs can do equal or better work than most of the humans I ever had working under me in this area. I have been able to complete jobs on my own that formerly would've required a full-time team of four or five people. Part of that team might be doing the same or similar work to me, but other members would've been doing specialized work that I was incapable of doing myself now, I can task an LLM to complete certain technical aspects of the job that were beyond my skill set. Historically, I had to spend a lot of time communicating my needs as well as project managing a completion of certain tasks. Now, I can spend less time communicating and project managing to produce an equivalent product.