The more logical explanation is that AI is not a wave like the Internet and Mobile, but it is instead a wave like cryptocurrencies, NFTs and tulip bulbs.
If there’s one thing that almost 3 decades at or near the forefront of Tech has taught me is that “novel” is not the same as “better”, and that of all the times a novel technology was pushed by insane amounts of hype, only a handful turned out to match the hype and the ratio of good-ones to bullshit has become much worse in the last 2 decades as the Tech Startup sector fully morphed from Techie-driven to Financeer-driven.
On hype alone “AI” (as in, what’s called now AI for the public, rather than the ML domain) stinks of greed-driven bullshit and the more one analyses the Technical details of LLMs and the Mathematics of it as well as of the improvements over time, the more painfully obvious it becomes that it’s not at all AGI or a path to it, rather it’s an overhyped attempt at it that turned out to be the wrong path. (All of which would’ve been absolutelly fine and a big Scientific step forward if it weren’t for the greedy financeer class and grifters pushing, purelly for their own personal enrichment, for people and companies to adopted it for doing things it’s not suitable for)
AI has an interesting economic trait in that it’s very, very expensive to deploy, and made very fast progress from 2022 to 2024. That caused investors with money to believe that:
Pushing the frontier was going to cost a lot of money. More than any other purported revolutionary tech.
Extrapolation of past improvement meant that whoever was on the cutting edge may end up with a product with a huge paying market.
So whoever wins this race would be rich, and the investment would have been worth it for them.
But since 2024, we’ve seen that the cutting edge got even more expensive much faster than expected, and much of the improvements in performance now come from inference rather than training, which represents a high ongoing cost.
Now, if we extrapolate from that trend line, we’ll see that the market will be much smaller for AI services at the cost it takes to provide that service, and the question then becomes whether the industry can make its operations cheaper, fast enough to profitably provide a service people will pay for.
I have my doubts they’ll succeed, and we might just be looking at the industry like supersonic flight: conceptually interesting, technically feasible, but just a commercial dead end because it’s too expensive.
The economics of it don’t add up and the growth rate of the curve of improvement over time has already significativelly fallen which looking at the historical curves for other technologies is a very strong indication that it’s approaching the limits of how far it will go even though it’s nowhere close to the hype.
So at both levels it all looks like a massive bet in the wrong horse that’s turning out not to be a winner but it keeps getting pushed by those who bet on it in the hope of making enough people and companies dependent that its sustained by nothing more than the unacceptable cost of it failing.
(In terms of strategy, it’s similar to how Uber started by using loopholes in the regulations for taxis, investing heavilly in becoming so big and established fast that when Authorities around the world got around to address those loopholes, they ended up accepting Uber and the like as something that could not be reversed and instead of regulating it out of existence, legitimized it. A very similar strategy was used by AirBNB: make the facts on the ground so big and reverting them so damaging that their low-value-adding business model with massive negative externalities and collateral damage ends up protected rather than made to pay for the societal costs of said collateral damage and negative externalities - essentially at some level Uber and especially AirBNB are being heavilly subsidized by society by being allowed to “polute” at will without paying for it).
So as I see it, the way Microsoft and other AI investors are going at it is to try and create a beachhead for it via hype, branding and lock-in in the expectation that something will come along at some point from the companies they invested in that is actually a genuine breakthrough that uses all the computing capacity created with their investment money.
I think that the reason why from the point of view of the public the AI adoption feels wrong is because it’s almost entirelly top-down, driven by marketing techniques and against the natural desires of people - it’s a novel form of entertainment being shoved down people’s throats as suitable for important responsabilities.
From my own experience, this feel a lot like the hype part of the cycle for the Segway, only with 100x or 1000x more investment money behind it.
The economics of it don’t add up and the growth rate of the curve of improvement over time has already significativelly fallen which looking at the historical curves for other technologies is a very strong indication that it’s approaching the limits of how far it will go even though it’s nowhere close to the hype.
Yeah, I’m convinced that they’ve maintained the illusion of continued exponential improvement from 2024-2026 by sneaking in exponential increase in resources (hardware complexity, power consumption), to prop things up past what should have been a plateau.
The way GDP is calculated you can in the short term create GDP “growth” by using debt to invest in things whose eventual return on investment is less than 1.
I think you’re right about there being entirely too much hype, and we’re definitely in a bubble right now, but I think this technology is here to stay. It definitely won’t have the current economic shape forever, but it’ll follow a similar trajectory to the “web 2.0”/social media tech. Is that a good thing? Probably not, but we may end up being surprised. I personally think running the models locally will end up being the best way to use AI.
Personal computers were a hype-bubble, until they weren’t.
The internet was a hype-bubble, until it wasn’t.
People having instant internet access to stock trading turned everything into hype-bubbles, and some of those - like BTC - are awfully stubbornly not popping - yet.
The Japanese commercial real-estate and US housing markets were supposed to be deflation-proof, but pump 'em up high enough and they will pop.
“AI” has “shown promise” since the 1960s. “Machine vision” was sorting and routing checks to banks for payment even back then, putting all kinds of clerks out of that job, into others. Same happened to telephone switchboard operators, “typing pools”, and later transcriptionists. Back in the 1990s I stood in the FDA presenting my company’s latest idea for a device, I stood next to a “computer vision for pap-smear screening” tool which was already, more than 25 years ago, out-performing standard human based pap-smear screening methods for false negatives by a rate of better than 2:1.
There are things LLMs can do today that they couldn’t do a year ago, and there are more people learning how to use LLMs effectively, just like there were people learning how to do more than just play Solitare on their personal computers in the early 1990s.
Agentic AI is mainly an entertainment technology being pushed as something that can take over professional responsabilities.
It’s being pushed like that because a lot of investors have been trying to get a new Web (1.0) Bubble running - the Internet was the last Tech that speculative investor could ride to infinity and beyond, ending up having an impact on everything (mobile also had an impact on everything but it wasn’t driven by such investors) and a lot of speculative investors in Tech have wanted their turn in the Get Stupidly Rich Quick wheel since 2000.
The social media bubble, even though it made a few people lots of money, was way smaller because its impact in businesses was much more limited than the Internet.
So for a lot of use cases where Agentic AI is being pushed, it’s kinda like pushing using Facebook or the Rubik Cube for all kinds of responsabilities business environments.
The funny bit is that without the insane hype from that kind of investors, Agentic AI would right now be finding the niches it’s well suited for, rather than being put in places were the kind of mistakes it makes once in a while can end lives, destroy careers and collapse companies.
My only complaint here is that there is a lot of very, very valid use cases for “AI” specifically “Agentic AI”.
We (including myself) may not like a lot of those uses because it devalues my fellow workers but it does not change the fact that it works.
The problem is everyone needs to be so goddamn polarizing and god forbid we have a mature honest discussion about the tools being built and how they are changing society as we know it.
We should be discussing and pushing for UBI across the world for decades now as youth unemployment is already at dangerous levels in continents like Africa (lol of course we don’t care because black people) but no instead we have asshats pushing a narrative of “AI bad”. It’s not. It has many purposes. Smarter people know this and it’s why it isn’t going away and the train is not going to stop if you don’t pull your head out of your ass.
/rant
I can’t wait to dip out of society and find somewhere in the middle of nowhere to live a quiet life with minimal technology in my life. I’m done with all of you. I stand by what I’ve said to my mum many times over the years. I hate people. I love persons.
Hey, the feeling is mutual, all of us would also like it if you fuck off to the middle of nowhere. Don’t forget to take all your “useful technology” with you while you’re at it.
The list of valid use cases for AI is bound by “what is the worst possible consequence of a mistake done here”, because the statistical distribution of mistakes in terms of severity of consequences of things like Agentic AI is uniform (meaning, they’re just as likely to do the worst mistakes with the nastiest consequences as they are doing the smallest mistakes), which it is not the case with humans who make more of an effort and give more attention to avoiding catastrophic mistakes and also have a “this is stupid” (i.e. don’t put glue in pizza, don’t tell a suicidal person to kill themselves) recognition capability which also stops a lot of the nastiest mistakes.
This is something which is not noticeable to most people because most people don’t have deep enough process experience in at least one expert domain and process analysis experience, to upfront recognized anything beyond the “in your face” elements of using AI (or using anything, really) in a process.
Very few people would think “what’s the risk profile for this business of giving this thing these responsabilities”.
So they seriously overestimate what are valid use cases for AI, something that the hype around it also pushes for: not a single AI vendor will ever mention just “error distribution” or anything close to it.
Obviously, when the thing blows up catastrophically by doing something which for a human is “obviously a bad idea”, THEN people recognized that AI is unsuitable for that, but by then its often too late.
(Easy example: lawyers using AI to make submissions to the Court and ending up disbarred because those submissions “quoted” invented case law).
So I don’t expect Agentic AI to fuck society up by taking a large fraction of the jobs, I expect Agentic AI to fuck society up by an accumulation over time of random catastrophic mistakes that kill people and collapse otherwise stable companies, mistakes that humans in such positions would never do or at least be way less likely to do.
It’s going to be akin to death by cummulative poisoning.
It’s going to be akin to death by cummulative poisoning.
Agree and despite what it may seem like it really is gradual right now. What people are avoiding (god the C level discussions I have been witness to is mindblowing) is the long term damage of their choices today.
The amount of times I have heard executive talks with “you know we both have kids around the same age what do you see this doing?” And they always wrap it with something positive. These fuckers most likely have their kids in private schools, not to mention their kids have all the connections these fucking parents can afford them.
In short the execs making the decisions have their heads equally shoved so far up their own asses they are ignoring the problems on the horizon.
My point is that for Agentic AI mistakes with catastrophic consequences are just as likelly as minor mistakes, which is not the case for people because humans can spot the “obviously stupid” or “obviously dangerous”, plus they make more of an effort to avoid mistakes that can have very bad consequences, so they tend to make catastrophic mistakes will less probability than minor mistakes.
People giving psychological advice are incredibly unlikely to tell suicidal people to “kill yourself”, those giving food recipes are incredibly unlikely to say that pizza should have glue on top or those deploying software in Production are incredibly unlikely to delete the whole fucking Production environment including backups.
So even if the total rate of mistakes of an an Agentic AI was less than a human, its rate of catastropic mistakes would still be much higher than a human.
This is however not obvious unless one actually analises the risk profile of using Agentic AI in a specific place in a specific process, a skill very few people have plus it requires information about and/or understanding of Agentic AI which itself very few people have and the AI vendors activelly do not want people to have.
So you end up with an e-mail fluffing and defluffing machine being used to summarize and store medical info about patients and then down the line somebody gets given something that kills them because the data on file had a critical mistake.
This is why I said that its “the worst possible consequence of a mistake done here” that limit Agentic AI suitability: because generally you’re going to have way more catastrophic mistakes with an AI that you will even with even an human with no domain experience.
When human makes a mistake, they learn, they continue to enrich humanity, they make a blueprint how not to make the same mistake again, if not for humanity, but at least for themselves. It also fuels some creativity so one mistake might lead to something good later.
When a mistake generator makes a mistake, it’s just another mistake in a pile of mistakes that only worsen our collective human experience.
The more logical explanation is that AI is not a wave like the Internet and Mobile, but it is instead a wave like cryptocurrencies, NFTs and tulip bulbs.
If there’s one thing that almost 3 decades at or near the forefront of Tech has taught me is that “novel” is not the same as “better”, and that of all the times a novel technology was pushed by insane amounts of hype, only a handful turned out to match the hype and the ratio of good-ones to bullshit has become much worse in the last 2 decades as the Tech Startup sector fully morphed from Techie-driven to Financeer-driven.
On hype alone “AI” (as in, what’s called now AI for the public, rather than the ML domain) stinks of greed-driven bullshit and the more one analyses the Technical details of LLMs and the Mathematics of it as well as of the improvements over time, the more painfully obvious it becomes that it’s not at all AGI or a path to it, rather it’s an overhyped attempt at it that turned out to be the wrong path. (All of which would’ve been absolutelly fine and a big Scientific step forward if it weren’t for the greedy financeer class and grifters pushing, purelly for their own personal enrichment, for people and companies to adopted it for doing things it’s not suitable for)
AI has an interesting economic trait in that it’s very, very expensive to deploy, and made very fast progress from 2022 to 2024. That caused investors with money to believe that:
But since 2024, we’ve seen that the cutting edge got even more expensive much faster than expected, and much of the improvements in performance now come from inference rather than training, which represents a high ongoing cost.
Now, if we extrapolate from that trend line, we’ll see that the market will be much smaller for AI services at the cost it takes to provide that service, and the question then becomes whether the industry can make its operations cheaper, fast enough to profitably provide a service people will pay for.
I have my doubts they’ll succeed, and we might just be looking at the industry like supersonic flight: conceptually interesting, technically feasible, but just a commercial dead end because it’s too expensive.
The economics of it don’t add up and the growth rate of the curve of improvement over time has already significativelly fallen which looking at the historical curves for other technologies is a very strong indication that it’s approaching the limits of how far it will go even though it’s nowhere close to the hype.
So at both levels it all looks like a massive bet in the wrong horse that’s turning out not to be a winner but it keeps getting pushed by those who bet on it in the hope of making enough people and companies dependent that its sustained by nothing more than the unacceptable cost of it failing.
(In terms of strategy, it’s similar to how Uber started by using loopholes in the regulations for taxis, investing heavilly in becoming so big and established fast that when Authorities around the world got around to address those loopholes, they ended up accepting Uber and the like as something that could not be reversed and instead of regulating it out of existence, legitimized it. A very similar strategy was used by AirBNB: make the facts on the ground so big and reverting them so damaging that their low-value-adding business model with massive negative externalities and collateral damage ends up protected rather than made to pay for the societal costs of said collateral damage and negative externalities - essentially at some level Uber and especially AirBNB are being heavilly subsidized by society by being allowed to “polute” at will without paying for it).
So as I see it, the way Microsoft and other AI investors are going at it is to try and create a beachhead for it via hype, branding and lock-in in the expectation that something will come along at some point from the companies they invested in that is actually a genuine breakthrough that uses all the computing capacity created with their investment money.
I think that the reason why from the point of view of the public the AI adoption feels wrong is because it’s almost entirelly top-down, driven by marketing techniques and against the natural desires of people - it’s a novel form of entertainment being shoved down people’s throats as suitable for important responsabilities.
From my own experience, this feel a lot like the hype part of the cycle for the Segway, only with 100x or 1000x more investment money behind it.
Yeah, I’m convinced that they’ve maintained the illusion of continued exponential improvement from 2024-2026 by sneaking in exponential increase in resources (hardware complexity, power consumption), to prop things up past what should have been a plateau.
The way GDP is calculated you can in the short term create GDP “growth” by using debt to invest in things whose eventual return on investment is less than 1.
I think you’re right about there being entirely too much hype, and we’re definitely in a bubble right now, but I think this technology is here to stay. It definitely won’t have the current economic shape forever, but it’ll follow a similar trajectory to the “web 2.0”/social media tech. Is that a good thing? Probably not, but we may end up being surprised. I personally think running the models locally will end up being the best way to use AI.
Personal computers were a hype-bubble, until they weren’t.
The internet was a hype-bubble, until it wasn’t.
People having instant internet access to stock trading turned everything into hype-bubbles, and some of those - like BTC - are awfully stubbornly not popping - yet.
The Japanese commercial real-estate and US housing markets were supposed to be deflation-proof, but pump 'em up high enough and they will pop.
“AI” has “shown promise” since the 1960s. “Machine vision” was sorting and routing checks to banks for payment even back then, putting all kinds of clerks out of that job, into others. Same happened to telephone switchboard operators, “typing pools”, and later transcriptionists. Back in the 1990s I stood in the FDA presenting my company’s latest idea for a device, I stood next to a “computer vision for pap-smear screening” tool which was already, more than 25 years ago, out-performing standard human based pap-smear screening methods for false negatives by a rate of better than 2:1.
There are things LLMs can do today that they couldn’t do a year ago, and there are more people learning how to use LLMs effectively, just like there were people learning how to do more than just play Solitare on their personal computers in the early 1990s.
Agentic AI is mainly an entertainment technology being pushed as something that can take over professional responsabilities.
It’s being pushed like that because a lot of investors have been trying to get a new Web (1.0) Bubble running - the Internet was the last Tech that speculative investor could ride to infinity and beyond, ending up having an impact on everything (mobile also had an impact on everything but it wasn’t driven by such investors) and a lot of speculative investors in Tech have wanted their turn in the Get Stupidly Rich Quick wheel since 2000.
The social media bubble, even though it made a few people lots of money, was way smaller because its impact in businesses was much more limited than the Internet.
So for a lot of use cases where Agentic AI is being pushed, it’s kinda like pushing using Facebook or the Rubik Cube for all kinds of responsabilities business environments.
The funny bit is that without the insane hype from that kind of investors, Agentic AI would right now be finding the niches it’s well suited for, rather than being put in places were the kind of mistakes it makes once in a while can end lives, destroy careers and collapse companies.
My only complaint here is that there is a lot of very, very valid use cases for “AI” specifically “Agentic AI”.
We (including myself) may not like a lot of those uses because it devalues my fellow workers but it does not change the fact that it works.
The problem is everyone needs to be so goddamn polarizing and god forbid we have a mature honest discussion about the tools being built and how they are changing society as we know it.
We should be discussing and pushing for UBI across the world for decades now as youth unemployment is already at dangerous levels in continents like Africa (lol of course we don’t care because black people) but no instead we have asshats pushing a narrative of “AI bad”. It’s not. It has many purposes. Smarter people know this and it’s why it isn’t going away and the train is not going to stop if you don’t pull your head out of your ass.
/rant
I can’t wait to dip out of society and find somewhere in the middle of nowhere to live a quiet life with minimal technology in my life. I’m done with all of you. I stand by what I’ve said to my mum many times over the years. I hate people. I love persons.
Hey, the feeling is mutual, all of us would also like it if you fuck off to the middle of nowhere. Don’t forget to take all your “useful technology” with you while you’re at it.
The list of valid use cases for AI is bound by “what is the worst possible consequence of a mistake done here”, because the statistical distribution of mistakes in terms of severity of consequences of things like Agentic AI is uniform (meaning, they’re just as likely to do the worst mistakes with the nastiest consequences as they are doing the smallest mistakes), which it is not the case with humans who make more of an effort and give more attention to avoiding catastrophic mistakes and also have a “this is stupid” (i.e. don’t put glue in pizza, don’t tell a suicidal person to kill themselves) recognition capability which also stops a lot of the nastiest mistakes.
This is something which is not noticeable to most people because most people don’t have deep enough process experience in at least one expert domain and process analysis experience, to upfront recognized anything beyond the “in your face” elements of using AI (or using anything, really) in a process.
Very few people would think “what’s the risk profile for this business of giving this thing these responsabilities”.
So they seriously overestimate what are valid use cases for AI, something that the hype around it also pushes for: not a single AI vendor will ever mention just “error distribution” or anything close to it.
Obviously, when the thing blows up catastrophically by doing something which for a human is “obviously a bad idea”, THEN people recognized that AI is unsuitable for that, but by then its often too late.
(Easy example: lawyers using AI to make submissions to the Court and ending up disbarred because those submissions “quoted” invented case law).
So I don’t expect Agentic AI to fuck society up by taking a large fraction of the jobs, I expect Agentic AI to fuck society up by an accumulation over time of random catastrophic mistakes that kill people and collapse otherwise stable companies, mistakes that humans in such positions would never do or at least be way less likely to do.
It’s going to be akin to death by cummulative poisoning.
Agree and despite what it may seem like it really is gradual right now. What people are avoiding (god the C level discussions I have been witness to is mindblowing) is the long term damage of their choices today.
The amount of times I have heard executive talks with “you know we both have kids around the same age what do you see this doing?” And they always wrap it with something positive. These fuckers most likely have their kids in private schools, not to mention their kids have all the connections these fucking parents can afford them.
In short the execs making the decisions have their heads equally shoved so far up their own asses they are ignoring the problems on the horizon.
Its not because humans make those mistakes all the time. It doesn’t need to be %100, it just needs to be like 95% to be better than humans.
My point is that for Agentic AI mistakes with catastrophic consequences are just as likelly as minor mistakes, which is not the case for people because humans can spot the “obviously stupid” or “obviously dangerous”, plus they make more of an effort to avoid mistakes that can have very bad consequences, so they tend to make catastrophic mistakes will less probability than minor mistakes.
People giving psychological advice are incredibly unlikely to tell suicidal people to “kill yourself”, those giving food recipes are incredibly unlikely to say that pizza should have glue on top or those deploying software in Production are incredibly unlikely to delete the whole fucking Production environment including backups.
So even if the total rate of mistakes of an an Agentic AI was less than a human, its rate of catastropic mistakes would still be much higher than a human.
This is however not obvious unless one actually analises the risk profile of using Agentic AI in a specific place in a specific process, a skill very few people have plus it requires information about and/or understanding of Agentic AI which itself very few people have and the AI vendors activelly do not want people to have.
So you end up with an e-mail fluffing and defluffing machine being used to summarize and store medical info about patients and then down the line somebody gets given something that kills them because the data on file had a critical mistake.
This is why I said that its “the worst possible consequence of a mistake done here” that limit Agentic AI suitability: because generally you’re going to have way more catastrophic mistakes with an AI that you will even with even an human with no domain experience.
That’s just not even true. People with no experience are going to fuck shit up completely. We have a human president and look where that’s getting us.
When human makes a mistake, they learn, they continue to enrich humanity, they make a blueprint how not to make the same mistake again, if not for humanity, but at least for themselves. It also fuels some creativity so one mistake might lead to something good later.
When a mistake generator makes a mistake, it’s just another mistake in a pile of mistakes that only worsen our collective human experience.
Very few humans do that. Vast majority is far more sloppy than any AI slop I’ve ever seen.