Humans cannot generate flawless output on one pass either. That’s why we have the concept of the draft. The lawyers who submitted LLM output without reviewing it are likely the type of bad lawyer who would submit an articling student’s first draft without reading it either.
The revenue situation is fairly obvious: embedded advertising. The people who cannot afford YouTube premium will similarly put up with ads in their free ChatGPT.
Maybe. But that implies that the ad market needs to expand by at least 25% and possibly more like 50% because AI ads will take away revenue from existing ad channels that are the companies doing much of the AI buildout.
And ads anywhere still need to generate revenue that justifies ad rates. Last time I was deep in that subject, skepticism was rising fast regarding rates of conversion to a click, much less to a sale. I'm not up on the latest but difficult to imagine there's significantly more evidence demonstrating ROI is justifying current rates (related but different, see also: paid outreach / promoted posts). Maybe few care when times are fat.
My hunch is that there are some massive interdependencies that will make for seriously turbulent times in surprising places, once and to the extent that times get tight.
Generally I lean heavily on Grok and fool around with duckduck's free models. Grok is helping me learn R over Stata at light speed, including presentation of trade-offs / pros/cons of different code options. Almost any other time I used* it as Deep Search, which provided* a box that shows 1. its sources, 2. its search terms, and 3. its process of scanning results and revising its sources and searches. Where it goes off-track in any response is (was*) thus spelled out, valuable not only for course-correction in that episode but for building skills to direct future queries or tasks more usefully.
*I /used to/ use it always in Deep Search, dismayed to see that's apparently /just/ been disabled (8-10 Aug). Grok suggests its disappearance "seems tied to resource allocation" (more pictures, less usefulness? or tired of losing $20 for every buck spent?) but vanilla Grok gets most of its info from X hot takes and the like, so that's probably just X rando speculation. Too bad.
Maybe a smaller work force is a ̶g̶o̶o̶d̶ ̶ great thing. Perhaps the world were families included a breadwinner and a homemaker wasn't so bad.
I queried Grok concerning percentage of women in the work force;
"1925: Data from the U.S. Census Bureau indicates that approximately 20% of the U.S. workforce was female in 1920, with similar figures likely for 1925. Women were primarily employed in roles like domestic service, clerical work, and manufacturing, with limited access to professional occupations.
2025: According to the U.S. Bureau of Labor Statistics, in July 2025, women made up about 47% of the U.S. workforce. This reflects a significant increase over the century, driven by societal changes, increased education, and policy shifts."
Since this is trivial I didn't double check if Grock's lying but (Data sources noted in Grok's reply.) it'd be easy to do so.
No, the honorable trade of homemakeing did not mean a world of barefoot and pregnant drudges. A. D.... (After Da advent of fridges, vacuums and washing machines.) the lady of the house had time and means for an active and satisfying life, social and otherwise.
Maybe, just maybe less work and more play just might bring Jack and Jill a lot more joy.
BTW: Anchorage/lower forty eight; For may years many of us here in Alaska kind of thought of it that way, we'd often refer to it as north Seattle, a large city only twenty minutes away from Alaska. ;-)
The main reason wives are now, currently in the workforce is because the husband alone can no longer earn enough to support the family. Why this is, is rather ironic. Bedford women enter the workforce force in the US, enmasse, the husband generally could work a 9 to 5 job and support his family. Once women entered, the workforce doubled in size, basically over night, flooding the market with potential employees without increasing available jobs. Thus we now have the reality that few married couples, especially those with kids, can afford to live on a single paycheck. AI will not solve this problem.
The basic problem we have sorting through this stuff, is that it involves value estimation for types of labor, and we have been employing and trusting university faculty to have opinions and answers there.
There's a non-austrian and an austrian approach to this economic modeling, plus the bizarro totalitarian imaginationland. Non-austrian is that we can assign book values, and then aggregate those, and our understanding of the systemic error may or may not include a fraud adjustment term or a dozen. austrian is that we aggregate voluntary transactions.
Austrian models are not entirely forecastable, because the people doing the transactions do not know themselves yet. So economic forecasters (judging as a non-specialist), may be fucked unless they adopt a non-austrian model, and make some assignments or guesses about the book values. Book value estimation based economic forecasting is going to be wildly terrible where machines are concenred, and where significant work force changes are concerned.
My view is that engineers are not trained to have good economic models, and have to reinvent the technique for themselves. I also think that the correct view involves having a sense of currency, nutrional calories, and energy.
Food is the most fundamental economic activity.
A secondary activity of some importance is security, which is part of preventing randos for burning down fields of viable crops.
Right now, academics have raised doubts about how much they are screwing us on primary food production, and on security issues. Academic research, returns in the future, are a little bit less vital than food in the present, and security in the present, are. So we know that a lot of faculty and retired faculty running their mouths in public ATM are wrong, and maybe also harmful, when it comes to some of this estimation.
I don't think there are many people equipped by training or by experience to do a really excellent job of judging all of this.
My guess is that the tasks we are talking about are white collar, and are probably not touching the most extreme or important inefficiencies in the white collar world. Which are government regulation.
There are neural net approaches to food problems, or security problems, but I am not yet aware of any killer app that would drastically change the economics of those.
Of course, food can still be pretty cheap, so we could drastically change the economy and have wildly different distributions of work.
But, to me, this effort has smelled like technocrats trying to force a change, instead of trying to be open to causing a change, and also primarily focused on common sense business fundamentals. Of course, big tech has fundamentals that are maybe crazy making enough to inspire a lot of malinvestment. And the place they are trying to make the change did not feel to me like it must be value dense enough to be where a new machine changes the economy.
We don't entirely know what a machine can do, until we have the machine, and see.
There's a tribal logic that supposes that if the Obama lunatics hate it, it must actually be good. That then would be an argument for AI. Because the Obama organization seems to have gone against AI, as part of their opposition to free speech. I do not credit Obama with such an understanding of human behavior, or of machines, or of the future.
It all comes down to having made our best initial and early guesses, and then seeing what the machines do, and what people choose as a result.
We can be sure that I err. I'm not for the LLMs. But, I can freak out and nearly paralyze myself by trying to change to a new tool chain while I am too scared to change anything, simply because I frighten myself over the hypothetical of changes to my OS.
(I identify as being a Sanjay tier student of programming. So I find some of this concerning, before I recall that a) I am bad at estimating my skill levels b) I've mostly been gambling on a significantly different skillset that I am training up on.)
I think there are at least two additional points that need to be considered. Why are these different AI systems (is that the correct term) being given out free? To me, it looks very much like a drug dealer giving laced candy free to young children. Not that I think or don’t think that if the following is true, that it will be successful enough to make up for the short fall, but what happens once an institution becomes dependent upon AI for its operations and the free version ceases to be and the only alternative is to pay for an expensive license? Other programs and services have done this.
The other that comes immediately to my mind is where in the hell is the power going to come from? Many of the same dolts pushing for AI everything are the same that are actively working to destroy humanity’s ability to generate any energy unless it comes from the solar or wind. From what little I have read, even if the whole world imitated N. Korea, there still would not be enough energy generation capability to power AI’s voracious appetite for electricity.
Another just came to mind. Personal experiences this year alone indicate that all customer service inquiries will soon, like perhaps in months, be handled by AI alone. Even when you go in to the B&M customer service center.
While I have fun playing with AI to edit my photos, I see little good and a hell of a lot of extremely bad to come from AI.
Humans cannot generate flawless output on one pass either. That’s why we have the concept of the draft. The lawyers who submitted LLM output without reviewing it are likely the type of bad lawyer who would submit an articling student’s first draft without reading it either.
The revenue situation is fairly obvious: embedded advertising. The people who cannot afford YouTube premium will similarly put up with ads in their free ChatGPT.
Maybe. But that implies that the ad market needs to expand by at least 25% and possibly more like 50% because AI ads will take away revenue from existing ad channels that are the companies doing much of the AI buildout.
And ads anywhere still need to generate revenue that justifies ad rates. Last time I was deep in that subject, skepticism was rising fast regarding rates of conversion to a click, much less to a sale. I'm not up on the latest but difficult to imagine there's significantly more evidence demonstrating ROI is justifying current rates (related but different, see also: paid outreach / promoted posts). Maybe few care when times are fat.
My hunch is that there are some massive interdependencies that will make for seriously turbulent times in surprising places, once and to the extent that times get tight.
I use AI shortcuts quite similarly, and started seriously testing in large part coincidentally prompted by an earlier esr X post:
https://x.com/esrtweet/status/1910809356381413593
Generally I lean heavily on Grok and fool around with duckduck's free models. Grok is helping me learn R over Stata at light speed, including presentation of trade-offs / pros/cons of different code options. Almost any other time I used* it as Deep Search, which provided* a box that shows 1. its sources, 2. its search terms, and 3. its process of scanning results and revising its sources and searches. Where it goes off-track in any response is (was*) thus spelled out, valuable not only for course-correction in that episode but for building skills to direct future queries or tasks more usefully.
*I /used to/ use it always in Deep Search, dismayed to see that's apparently /just/ been disabled (8-10 Aug). Grok suggests its disappearance "seems tied to resource allocation" (more pictures, less usefulness? or tired of losing $20 for every buck spent?) but vanilla Grok gets most of its info from X hot takes and the like, so that's probably just X rando speculation. Too bad.
I saw that Xeet too and it made me more open to trying AI assisted sw dev
Maybe a smaller work force is a ̶g̶o̶o̶d̶ ̶ great thing. Perhaps the world were families included a breadwinner and a homemaker wasn't so bad.
I queried Grok concerning percentage of women in the work force;
"1925: Data from the U.S. Census Bureau indicates that approximately 20% of the U.S. workforce was female in 1920, with similar figures likely for 1925. Women were primarily employed in roles like domestic service, clerical work, and manufacturing, with limited access to professional occupations.
2025: According to the U.S. Bureau of Labor Statistics, in July 2025, women made up about 47% of the U.S. workforce. This reflects a significant increase over the century, driven by societal changes, increased education, and policy shifts."
Since this is trivial I didn't double check if Grock's lying but (Data sources noted in Grok's reply.) it'd be easy to do so.
No, the honorable trade of homemakeing did not mean a world of barefoot and pregnant drudges. A. D.... (After Da advent of fridges, vacuums and washing machines.) the lady of the house had time and means for an active and satisfying life, social and otherwise.
Maybe, just maybe less work and more play just might bring Jack and Jill a lot more joy.
BTW: Anchorage/lower forty eight; For may years many of us here in Alaska kind of thought of it that way, we'd often refer to it as north Seattle, a large city only twenty minutes away from Alaska. ;-)
Yes indeed, the SAHM should ride again. That will solve all kinds of problems
The main reason wives are now, currently in the workforce is because the husband alone can no longer earn enough to support the family. Why this is, is rather ironic. Bedford women enter the workforce force in the US, enmasse, the husband generally could work a 9 to 5 job and support his family. Once women entered, the workforce doubled in size, basically over night, flooding the market with potential employees without increasing available jobs. Thus we now have the reality that few married couples, especially those with kids, can afford to live on a single paycheck. AI will not solve this problem.
The basic problem we have sorting through this stuff, is that it involves value estimation for types of labor, and we have been employing and trusting university faculty to have opinions and answers there.
There's a non-austrian and an austrian approach to this economic modeling, plus the bizarro totalitarian imaginationland. Non-austrian is that we can assign book values, and then aggregate those, and our understanding of the systemic error may or may not include a fraud adjustment term or a dozen. austrian is that we aggregate voluntary transactions.
Austrian models are not entirely forecastable, because the people doing the transactions do not know themselves yet. So economic forecasters (judging as a non-specialist), may be fucked unless they adopt a non-austrian model, and make some assignments or guesses about the book values. Book value estimation based economic forecasting is going to be wildly terrible where machines are concenred, and where significant work force changes are concerned.
My view is that engineers are not trained to have good economic models, and have to reinvent the technique for themselves. I also think that the correct view involves having a sense of currency, nutrional calories, and energy.
Food is the most fundamental economic activity.
A secondary activity of some importance is security, which is part of preventing randos for burning down fields of viable crops.
Right now, academics have raised doubts about how much they are screwing us on primary food production, and on security issues. Academic research, returns in the future, are a little bit less vital than food in the present, and security in the present, are. So we know that a lot of faculty and retired faculty running their mouths in public ATM are wrong, and maybe also harmful, when it comes to some of this estimation.
I don't think there are many people equipped by training or by experience to do a really excellent job of judging all of this.
My guess is that the tasks we are talking about are white collar, and are probably not touching the most extreme or important inefficiencies in the white collar world. Which are government regulation.
There are neural net approaches to food problems, or security problems, but I am not yet aware of any killer app that would drastically change the economics of those.
Of course, food can still be pretty cheap, so we could drastically change the economy and have wildly different distributions of work.
But, to me, this effort has smelled like technocrats trying to force a change, instead of trying to be open to causing a change, and also primarily focused on common sense business fundamentals. Of course, big tech has fundamentals that are maybe crazy making enough to inspire a lot of malinvestment. And the place they are trying to make the change did not feel to me like it must be value dense enough to be where a new machine changes the economy.
We don't entirely know what a machine can do, until we have the machine, and see.
There's a tribal logic that supposes that if the Obama lunatics hate it, it must actually be good. That then would be an argument for AI. Because the Obama organization seems to have gone against AI, as part of their opposition to free speech. I do not credit Obama with such an understanding of human behavior, or of machines, or of the future.
It all comes down to having made our best initial and early guesses, and then seeing what the machines do, and what people choose as a result.
We can be sure that I err. I'm not for the LLMs. But, I can freak out and nearly paralyze myself by trying to change to a new tool chain while I am too scared to change anything, simply because I frighten myself over the hypothetical of changes to my OS.
(I identify as being a Sanjay tier student of programming. So I find some of this concerning, before I recall that a) I am bad at estimating my skill levels b) I've mostly been gambling on a significantly different skillset that I am training up on.)
I think there are at least two additional points that need to be considered. Why are these different AI systems (is that the correct term) being given out free? To me, it looks very much like a drug dealer giving laced candy free to young children. Not that I think or don’t think that if the following is true, that it will be successful enough to make up for the short fall, but what happens once an institution becomes dependent upon AI for its operations and the free version ceases to be and the only alternative is to pay for an expensive license? Other programs and services have done this.
The other that comes immediately to my mind is where in the hell is the power going to come from? Many of the same dolts pushing for AI everything are the same that are actively working to destroy humanity’s ability to generate any energy unless it comes from the solar or wind. From what little I have read, even if the whole world imitated N. Korea, there still would not be enough energy generation capability to power AI’s voracious appetite for electricity.
Another just came to mind. Personal experiences this year alone indicate that all customer service inquiries will soon, like perhaps in months, be handled by AI alone. Even when you go in to the B&M customer service center.
While I have fun playing with AI to edit my photos, I see little good and a hell of a lot of extremely bad to come from AI.
Alaska is a very cold place. It also has lots of water. It sounds like the ideal place to site data centers.