Users are likely to gain massively from the buildout of MAMLMs—natural-language interfaces, very big-data very high-dimension very flexible-functional-form classification and regression analysis...
That's not surprising at all. It has become increasingly difficult to find websites beyond a handful of well established players. Now I'm wishing for the return of something like the old Yahoo or Yellow Pages so I can actually find things.
As an ex-software developer, I was sort of jolted this week reading Andrew Ng's latest email missive on AI "vibe coding" and the skills needed. He had also mentioned 2 AI-centric IDEs. I had previously looked at videos of Google's Firebase Studio, which had mixed reviews. Ng mentioned Cursor and Windsurf. So I thought I should see what they are about. They are similar to Firebase in that they completely overturn what I did as a software engineer. The IDEs both code applications with prompts, fix bugs, and produce [alpha/beta level] code. But all the design work, architecture, and making sure each module works correctly, was gone. I feel dispossessed from such a software development world. It is almost incomprehensible. It seems developers would be creating AI hacks and then debugging the code that the AI was unable to fix. Ugh!
As for overturning the current tech giants, check out this vision for AI-based computers.
Shades on 1990s Netscape saying the browser will usurp the Windows OS. We saw what happened next. If his vision ever comes to pass, wouldn't this displace Microsoft's business and Apple's software development team? An infinitely malleable interface that responds to your [voice] prompts. I expect that some traditional applications would still be needed, although how they would look and even work might be very different. From my traditional perspective, where computer applications are deterministic, it feels more like telling an assistant what you want, rather than doing it yourself. [Will we all act like pre-computer age bosses having assistants and secretaries to do all the technical tasks?] I shudder at the thought. If it does go in that direction, I certainly do not want to be reliant on monthly subscriptions to an AI company to use a computer, and where that use could be cut off at a whim, or as little as an internet connection disruption. I would rather have the AI be local and running on my computer, even if it is not as powerful as the latest version running on a giant server farm.
I believe the possible scope of what AI will bring is too broad and too unpredictable to make any predictions with confidence. The current situation seems to me to call for scenario planning as the best applicable technique to describe and perhaps prepare for such a range of possible futures.
I don't remember the largest firms having steady 25% profit margins in 2000. Such large, sustained profit margins are a market failure which would normally be corrected with antitrust action. The precedent from 2000 isn't NASDAQ, it is Putin.
"What would Putin do?" is the question, if presented with six incredibly profitable firms whose enormous rents exist because of the state's protection? We should assume the same will happen in the US. Campaign contributions and inaugural donations won't be sufficient. Elon was an example.
As far as froth in the market goes, I was flown to India to consult with a tech company in fall of 2000 and spring 2008. I will let you know if the request comes again. Then, sell.
> Jeff Luth: As far as froth in the market goes, I was flown to India to consult with a tech company in fall of 2000 and spring 2008. I will let you know if the request comes again. Then, sell...
As with the atom bomb, the real secret of MAMLMs was that it could be done. Once that was known, the proliferation of models showed that it could be done for a few billion dollars, and DeepSeek showed that it could be done for peanuts.
And, so far the value perceived by users is similarly modest. At least for now, most people use AI as a Google replacement with a natural language interface, an improvement (pretty modest considering Google came out 25 years ago) for which they aren't willing to pay very much. Unless things change, this is very bad for Google. The only implication for the rest of the tech titans is that they won't recover the money they spent on the initial round of AI.
Your analysis assumes that the outcomes of the massively increased spending we are now observing will be different: both impossible to reproduce cheaply, and so valuable that people will pay a lot for it. What outcomes do you have in mind?
Let me stick my neck out: MAMLMs as natural-language interfaces are not going to produce profits for the providers. They are going to be expensive to provide. They will provide enough utility to users that platform oligopolists who don't want to be disrupted will have to provide them, or lose their existing cash flows. But they will not provide enough utility that platform oligopolists will be able to charge for them. So this form of "AI" ought to be very bad news for the stock market.
By contrast, MAMLMs as generalized high-dimensional flexible-functional-form huge-data classification and curve-fitting is a true wild card: it may have very few uses, and it may have enormous uses:
> John Quiggin: As with the atom bomb, the real secret of MAMLMs was that it could be done. Once that was known, the proliferation of models showed that it could be done for a few billion dollars, and DeepSeek showed that it could be done for peanuts. And, so far the value perceived by users is similarly modest. At least for now, most people use AI as a Google replacement with a natural language interface, an improvement (pretty modest considering Google came out 25 years ago) for which they aren't willing to pay very much. Unless things change, this is very bad for Google. The only implication for the rest of the tech titans is that they won't recover the money they spent on the initial round of AI. Your analysis assumes that the outcomes of the massively increased spending we are now observing will be different: both impossible to reproduce cheaply, and so valuable that people will pay a lot for it. What outcomes do you have in mind?...
Google Generative AI is usually the first thing that shows up on my IPad search. Hope I got that right. I like the overview that comes first. It saves a lot of time when a quick answer is needed.
Plus, so many of the searches give links to news media that you need to sign up for, and that is frustrating.
Although lately a new item has shown up on the bottom of the search somewhat bothers me. There is a link that says “ Dive deeper in AI mode.” There is so much detail in there, it is hard to see a need for it unless someone wants to do a lot more research.
The link sounds like “borderline cute”. Also, does that extensive programming put too much load on your computer?
There is something about generative AI that it gets boring after a few paragraphs. There is no sense of a person writing it. That is okay, I don’t expect a computer to do so much it gets stressed out. If the computer is not cleaned regularly and runs too hot, it will start to give wrong answers.
This IPad does not get too hot so far, but my older Kindle does run hot after awhile.
Is the market's response really " irrational exuberance"? This touches on a long running issue in Industrial Organization, the nature of barriers to entry. Both structuralist economists and Chicago/Austrian agree that unlike a diamond, barriers to entry are not forever. However, Chicago would severely limit the duration of barriers. In fact, the only persistent barrier they would recognize is where economies of scale are exhausted at a substantial share of market demand.
In contrast, structuralists note that the gains from accruing market power, however transitory, justifies considerable expenditures to shield your firm from competitive pressures. First mover advantages confer such benefits that a completive arms race is almost inevitable. In the AI space, the massive capital investments are easier for incumbents than potential entrants. Beyond that, investments that do not yield a marketable product become sunk which will discourage late comers competitive entry.
I seriously doubt that most users are going to benefit at all from the rollout of LLMs, especially not the LLMs that the big players are going to make available for the lower tier users of their services. It doesn't make any business sense. Google's goal is to sell advertisements, so there is no reason for them to make their LLM's answers any better than their search's answers, and they intentionally made search's answers worse so they could sell more advertising. Facebook has no reason to do any better, and their announced plan is to provide people with LLM generated "friends" to generate traffic they can sell advertising against. Spotify is already using LLMs to generate royalty free playlists in various styles to sell advertising against. Netflix would love to do something similar, but LLMs can't generate video well enough to sell advertising against, at least not yet.
Apple, with its alternate strategy of providing things that their users find useful to sell hardware against, is grappling with turning LLMs into something useful. You'll notice that Google isn't offering an LLM assistant linking its various productivity apps. Facebook has no similar plans either. It's goal is entertainment. At the 2024 WWDC, Apple made some claims about the future capabilities of their LLM which as of the 2025 WWDC have not been delivered. They set themselves up to be scooped, but where is one of the other players showing off their system doing what Apple has failed to do? Apple is excellent at marketing, so it's surprising that no other company's marketing department has gone for this gimme.
Speaking of the slop revolution -
https://housefresh.com/how-google-decimated-housefresh/
That's not surprising at all. It has become increasingly difficult to find websites beyond a handful of well established players. Now I'm wishing for the return of something like the old Yahoo or Yellow Pages so I can actually find things.
As an ex-software developer, I was sort of jolted this week reading Andrew Ng's latest email missive on AI "vibe coding" and the skills needed. He had also mentioned 2 AI-centric IDEs. I had previously looked at videos of Google's Firebase Studio, which had mixed reviews. Ng mentioned Cursor and Windsurf. So I thought I should see what they are about. They are similar to Firebase in that they completely overturn what I did as a software engineer. The IDEs both code applications with prompts, fix bugs, and produce [alpha/beta level] code. But all the design work, architecture, and making sure each module works correctly, was gone. I feel dispossessed from such a software development world. It is almost incomprehensible. It seems developers would be creating AI hacks and then debugging the code that the AI was unable to fix. Ugh!
As for overturning the current tech giants, check out this vision for AI-based computers.
https://www.youtube.com/watch?v=Z-XYhq0190E
Shades on 1990s Netscape saying the browser will usurp the Windows OS. We saw what happened next. If his vision ever comes to pass, wouldn't this displace Microsoft's business and Apple's software development team? An infinitely malleable interface that responds to your [voice] prompts. I expect that some traditional applications would still be needed, although how they would look and even work might be very different. From my traditional perspective, where computer applications are deterministic, it feels more like telling an assistant what you want, rather than doing it yourself. [Will we all act like pre-computer age bosses having assistants and secretaries to do all the technical tasks?] I shudder at the thought. If it does go in that direction, I certainly do not want to be reliant on monthly subscriptions to an AI company to use a computer, and where that use could be cut off at a whim, or as little as an internet connection disruption. I would rather have the AI be local and running on my computer, even if it is not as powerful as the latest version running on a giant server farm.
I believe the possible scope of what AI will bring is too broad and too unpredictable to make any predictions with confidence. The current situation seems to me to call for scenario planning as the best applicable technique to describe and perhaps prepare for such a range of possible futures.
I don't remember the largest firms having steady 25% profit margins in 2000. Such large, sustained profit margins are a market failure which would normally be corrected with antitrust action. The precedent from 2000 isn't NASDAQ, it is Putin.
"What would Putin do?" is the question, if presented with six incredibly profitable firms whose enormous rents exist because of the state's protection? We should assume the same will happen in the US. Campaign contributions and inaugural donations won't be sufficient. Elon was an example.
As far as froth in the market goes, I was flown to India to consult with a tech company in fall of 2000 and spring 2008. I will let you know if the request comes again. Then, sell.
:-)
> Jeff Luth: As far as froth in the market goes, I was flown to India to consult with a tech company in fall of 2000 and spring 2008. I will let you know if the request comes again. Then, sell...
As with the atom bomb, the real secret of MAMLMs was that it could be done. Once that was known, the proliferation of models showed that it could be done for a few billion dollars, and DeepSeek showed that it could be done for peanuts.
And, so far the value perceived by users is similarly modest. At least for now, most people use AI as a Google replacement with a natural language interface, an improvement (pretty modest considering Google came out 25 years ago) for which they aren't willing to pay very much. Unless things change, this is very bad for Google. The only implication for the rest of the tech titans is that they won't recover the money they spent on the initial round of AI.
Your analysis assumes that the outcomes of the massively increased spending we are now observing will be different: both impossible to reproduce cheaply, and so valuable that people will pay a lot for it. What outcomes do you have in mind?
Let me stick my neck out: MAMLMs as natural-language interfaces are not going to produce profits for the providers. They are going to be expensive to provide. They will provide enough utility to users that platform oligopolists who don't want to be disrupted will have to provide them, or lose their existing cash flows. But they will not provide enough utility that platform oligopolists will be able to charge for them. So this form of "AI" ought to be very bad news for the stock market.
By contrast, MAMLMs as generalized high-dimensional flexible-functional-form huge-data classification and curve-fitting is a true wild card: it may have very few uses, and it may have enormous uses:
> John Quiggin: As with the atom bomb, the real secret of MAMLMs was that it could be done. Once that was known, the proliferation of models showed that it could be done for a few billion dollars, and DeepSeek showed that it could be done for peanuts. And, so far the value perceived by users is similarly modest. At least for now, most people use AI as a Google replacement with a natural language interface, an improvement (pretty modest considering Google came out 25 years ago) for which they aren't willing to pay very much. Unless things change, this is very bad for Google. The only implication for the rest of the tech titans is that they won't recover the money they spent on the initial round of AI. Your analysis assumes that the outcomes of the massively increased spending we are now observing will be different: both impossible to reproduce cheaply, and so valuable that people will pay a lot for it. What outcomes do you have in mind?...
Looks like Bezos, Cook, Zuck et al read "Equipment Investment and Economic Growth"
Google Generative AI is usually the first thing that shows up on my IPad search. Hope I got that right. I like the overview that comes first. It saves a lot of time when a quick answer is needed.
Plus, so many of the searches give links to news media that you need to sign up for, and that is frustrating.
Although lately a new item has shown up on the bottom of the search somewhat bothers me. There is a link that says “ Dive deeper in AI mode.” There is so much detail in there, it is hard to see a need for it unless someone wants to do a lot more research.
The link sounds like “borderline cute”. Also, does that extensive programming put too much load on your computer?
There is something about generative AI that it gets boring after a few paragraphs. There is no sense of a person writing it. That is okay, I don’t expect a computer to do so much it gets stressed out. If the computer is not cleaned regularly and runs too hot, it will start to give wrong answers.
This IPad does not get too hot so far, but my older Kindle does run hot after awhile.
Is the market's response really " irrational exuberance"? This touches on a long running issue in Industrial Organization, the nature of barriers to entry. Both structuralist economists and Chicago/Austrian agree that unlike a diamond, barriers to entry are not forever. However, Chicago would severely limit the duration of barriers. In fact, the only persistent barrier they would recognize is where economies of scale are exhausted at a substantial share of market demand.
In contrast, structuralists note that the gains from accruing market power, however transitory, justifies considerable expenditures to shield your firm from competitive pressures. First mover advantages confer such benefits that a completive arms race is almost inevitable. In the AI space, the massive capital investments are easier for incumbents than potential entrants. Beyond that, investments that do not yield a marketable product become sunk which will discourage late comers competitive entry.
I seriously doubt that most users are going to benefit at all from the rollout of LLMs, especially not the LLMs that the big players are going to make available for the lower tier users of their services. It doesn't make any business sense. Google's goal is to sell advertisements, so there is no reason for them to make their LLM's answers any better than their search's answers, and they intentionally made search's answers worse so they could sell more advertising. Facebook has no reason to do any better, and their announced plan is to provide people with LLM generated "friends" to generate traffic they can sell advertising against. Spotify is already using LLMs to generate royalty free playlists in various styles to sell advertising against. Netflix would love to do something similar, but LLMs can't generate video well enough to sell advertising against, at least not yet.
Apple, with its alternate strategy of providing things that their users find useful to sell hardware against, is grappling with turning LLMs into something useful. You'll notice that Google isn't offering an LLM assistant linking its various productivity apps. Facebook has no similar plans either. It's goal is entertainment. At the 2024 WWDC, Apple made some claims about the future capabilities of their LLM which as of the 2025 WWDC have not been delivered. They set themselves up to be scooped, but where is one of the other players showing off their system doing what Apple has failed to do? Apple is excellent at marketing, so it's surprising that no other company's marketing department has gone for this gimme.