This is not what many would call a surprising finding. What I'm more curious about is whether this hastens the popping of the AI bubble or if it stubbornly remains intact.
AI company valuations are dependent on AI becoming a 30 trillion dollar industry, so basically creating so much AI productivity that it rivals the entire US gdp.
If you want to take a super myopic view, a cancer cure only affects the health sector But obviously health affects everything else and other industries.
The article takes a very limited view of AI, restricting it to tech, to justify the headline.
The people selling AI were certainly hoping to sell it to more than just tech companies. The dream of executives in every industry was it replacing labor across the board.
This is a dumb comparison because AI is explicitly marketed as directly impacting every sector of white collar work: HR, admin, accounting, everything.
tech/IT is considered a "cost center" in most other industries. if they can cut that, this is no different to them than the price of diesel falling. most people outside of tech would love this.
this isn't curing cancer which saves lives (worth 7-10M per person by US accounting). this is reducing the cost of infrastructure.
All we did is move the bottleneck from A to B but actually it's always been B and business has always been too blind or unwilling to accept B is the problem.
Tech sector has always been ultra-fast at adapting new technologies, other sectors will take between 5 to 15 years.
My partner was recently working for an insurance company that had only a few years ago moved away from their paper files to digital, and even then still had rooms of files that they'll never move to digital.
Her job will eventually go to AI, insurance assessor is mostly reading huge files and making a decision based on the laws & policy, but I'd give it minimum five to seven or so years before it happens.
And they will still need assessors, but they'll just stop hiring and let the workforce slowly dwindle down to maybe an 80% reduction in headcount.
This will happen to a lot of desk jobs/white collar jobs, especially the less senior ones, but it will take a decade.
Agreed. I've worked in wealth management and financial planning for 15 years. At this point I'm a senior planner, but my firm kinda just stopped hiring junior planners because our financial focused AI can spit out a rudimentary plan as a baseline that I modify and tailor quickly. My job is likely not at risk of AI overtaking it in my retirement timeline but a whole shitload of junior planners wont even start their careers, there's no where to start
I work in the insurance industry. Some jobs will, but any job that needs human connection and human decision making in conjunction will be around for quite some time.
Plus people want someone to blame and blaming a computer isnt the same as another human
My company cant get AI to scan and process invoices. Something we had automated reliably for many many years. Now we are paying literally millions in temp overhead to both process the backlog and help train the AI. This is pathetic.
It's made me money, I've built a bunch of custom tools for my business that save a fair bit of time, have ditched a few subscriptions & generally use it to get a bit more done faster.
I think maybe a lot of corporations just try and shove it in without having a clear use case, which to me seems doomed to fail.
The issue for the AI companies is that in order to pay for the hundreds of billions they're spending to build up their models, they need corporations to widely adopt AI and be willing to pay not-so-small fortunes for its use.
No amount of small businesses paying $20 a month is going to pay for an $800,000,000,000 a year construction binge. That requires mass adoption across corporations, and there's only so long these companies will be willing to eat the cost while awaiting future payoffs.
I am not surprised corporations are having trouble. There are tons of layers so getting traction will take a ton of time. Also most folks at corporate level don't have any clue how their tech stack works so getting value out of ai without knowing that will be tough.
I work in a mid sized company and I use ai daily and automated much of my reporting. I would say at least 50-60% of my workload has been moved to AI.
If you don’t mind me asking, what’s your role? I ask because ”automating much of my reporting” can mean wildly different things depending on your role and department. Pulling a list of leads and marketing/sales funnel info? Bug reports? Accounts receivables? SOC2 Compliance posture? Etc etc
Because the CEOs that demand it be used don’t know what they are doing. They use CoPilot to summarize emails and then think the same free version of CoPilot can replace engineers and software developers.
I just think it's something that will just take a lot of time. Once you have it set up AI is wonderful but it take time to set up the right semantic layers and processes for companies to get value. It not something that is super useful out of the box. In time I have no doubt more companies will catch in but that may be another 5 years away
I did yea, the £20 per month ChatGPT subscription, a Perplexity Pro subscription (I got that for free as part of a promo for a year) & the cheapest Claude sub.
I could have done a fair bit on the free subs, but you run into limits / can't use their smarter models.
In the lead up to the Dotcom Bubble we witnessed companies getting billion dollar valuations just because they launched a website and promised online sales (I'm looking at you, Pets.com and DrKoop.com). Businesses would be abler to greatly expand their sales and marketing channels through the magic of the internet. The only problem was that the general public wasn't really adopting online e-commerce very quickly. Most people who had internet got it through slow dialup, and it would take another decade before broadband internet access became common. Those high flying businesses with a website quickly ran out of investor cash, causing all those college grads they gave $100k starting salaries to have to put their Porsche for sale along the highways. E-commerce as an idea was viable, but it would take time and refinement before it could really be profitable.
This AI story looks very similar to what we saw with the Dotcom Bubble. I think we will see some of these high flying names like Anthropic run out of investor money and see their debt come due before long causing them to shutter. The real viable and useful tech won't be truly known for several years to come.
But your point about accessibility doesn’t make sense. People have access to AI and they have access to agents in a way that dialup users did not have access to e-commerce.
Rather, I think there is a real argument that cloud AI is unsustainable because it’s being too subsidized by VC, that real costs would drive most of the usage today to be self hosted/on-premise, and maybe there will be a catastrophic agentic event that causes governments to seriously hinder frontier labs.
My point about accessibility was really more to show why e-commerce struggled to get going and the adoption rate by consumers was low. There were other factors too, such as people not trusting purchasing things online back then compared to now.
There are some challenges to adoption of AI now that we are seeing, particularly by people who are afraid of the claims that AI will take their jobs. We also see reports from companies that told their employees to use AI, saw the bills for the usage come in without a commensurate rise in productivity or profit, and then either scaled down or backpedaled. Those things will provide a drag on AI adoption/scaling that may exceed AI firms' cash burn to profit path.
One of the main reasons e-commerce did not takeoff during the dot com bubble was that people did not trust online payments and trusted easy payments methods were not there yet. We have the same trust problem with AI during this AI bubble. It was not only access issues but also trust issues.
They are still deep in the red. High growth startups that take on massive debt also need massive growth, but Anthropics growth rate is down by half compared to the last quarter, so their expected valuation was just cut by 25%
They are? While bleeding billions or renting to other companies like xai. Unless you call Google, amazon, Microsoft ai companies, well then tesla isn't car company and spacex is ai one.
I think they mean that they would be profitable if they stopped spending billions on R&D. Of course, they have to, or their competitors would overtake them.
I have a feeling based on personal experience that AI is going to hit the economy in waves and it sort of feels like we’re past the first wave. Lots of people have figured out how to use it to do specific tasks within their existing roles and now it seems like we’re at a lull in finding new/creative ways to use it. Maybe that will change, but for now I feel myself going to it less frequently than a few months ago because I’ve restructured my work around things it helped me build but now that those things are built I don’t need it for the day to day operations.
There’s a negative incentive for services industries to promote their use of AI. Expensive consulting firms like McKinsey have been using AI but aren’t saying a lot about it because it makes their services seem less valuable. Same deal for legal services, financial services, regulatory services, etc.
If you can really save a big slab of staff cost with AI then your service has to become cheaper in a competitive market.
Unless there are new goods or services being created in the economy this isn't necessarily a benefit to individual companies.
The trivialised idea that your revenue can stay the same but your cost goes down by a big % is not realistic. Economics just doesn't work that way.
Productivity per employee shot up in the years following Netscape's release, the same trend has not happened since GPT became available.
In software development I can see some tasks become very much faster, but it just bumps into the next rate limit in the org - and it's easy for devs to spend far more than a problem is worth on tokens.
Yep, AI outside of computational fields is still so early, so brittle and so weak. But there is no reason why it cannot work. AI in math was brittle until math got (basically) solved. To be fair, math is easy to train on.... time will tell the truth!
The fact AI is working in the tech sector suggests AI is coming for other fields too. Everyone who has looked into the matter knows that coding AI is many years ahead of other fields. The only potential roadblock could be training, since teaching AI to do more empirical fields is probably harder than teaching it to work on the computer.
UPS’s AI route optimization saves about 100 million miles driven, 10 million gallons of fuel (around $300–400M a year).
Walmart’s using AI and computer vision at Sam’s Club to speed up checkout by 23%, which cuts labor costs and improves throughput. They’re even licensing some of that tech now.
And in healthcare, AI reduced radiologists’ workloads by about 33–44% in mammogram screening, while maintaining or improving detection rates. And AI scribe tools cut after-hours work by 30% and time spent in notes per appointment from ~10.3 min to ~8.2 min (20% reduction) for physicians.
These are all traditional applications of ML which are a subset of the AI the article is talking about. You don’t need an infinite amount of data for those applications.
The AI the article is talking about is these overly built out LLMs that’s trying to masquerade as an AGI. The problem is that’s a very expensive way to reach the same goals as specialized ML applications do and have for the last 15 years.
Straw man. I did not say you need a frontier LLM to route a truck or read a mammogram.
Most of what S&P500 companies are deploying is exactly the specialized ML you described. That's the adoption story.
And the "same goals cheaper" line doesn't hold. Scribe tools parsing open-ended clinical language aren't a 15-year-old ML model. Documentation time dropped 20% because the model handles unstructured language, which specialized ML never could.
The AGI pitch is overheated. That's a critique of the marketing, not the deployment. The numbers are the numbers.
This was either written by an LLM or you spend too much time talking to LLMs. The syntax and diction is very Claudesloppy. “Line doesn’t hold up”
LLMs are what frontier AI research and sales mostly consists of. Other versions of ML that you listed existed before the huge venture capital explosion in OpenAI and other LLM companies. You’re intentionally mixing the achievements of those specialized ML applications with LLMs from companies like openAI to make it seem useful outside of tech
So not denying being a bot and using bolding like Claude. No point arguing with a bot. You can’t change an LLMs mind because eventually the conversation will fall out of the context window. I can’t change the training data.
So the only point you listed earlier that applies to LLMs is a better note taking app for doctors. Trillions of dollars invested and that’s what we get from LLMs. A software that can save 2 minutes for a doctors appointment. Yeah guys, not a bubble.
Oh please, I've been bolding and italicizing on reddit for 15 years. Learn to format some time, you must be gen z.
Nice try narrowing the goalpost. The claim was no payoff outside tech. I gave you 3 examples that contradict that. You decided 1 counts then called it a note taking app and priced the whole industry off it. Miss me with that bullshit.
2 mins per appointment across every doctor everyday isn't small.
Neither is 100M miles or 23% faster checkout. All of that leads to lower costs, which leads to higher profits. Scale it out, and it could be BILLIONS saved. Less burnout for doctors too.
You edited your comment I after I responded. You only had one listed in your reply to me. Also you added back the other ML things that are not LLMs that were out long before venture capital build up.
Still bot slop grammar and diction. You’re making no sense too, because I did have a counter argument. I don’t have to make counterpoints here because you’re just listing the same info and I’m saying that a note taking app for hospitals is not enough deployment outside of tech to justify the trillions of dollars spent.
I can’t engage with you if you’re gonna edit everything you say after I respond to you. It’s one thing to correct typos, it’s another to rewrite sections. Goodbye.
AI did not invent routing but it made it meaningfully better.
The difference is ML improving the forecasts feeding it. Traffic, delivery windows, weather, package volume. The engine is OR, the inputs got way better. Plus drivers actually follow the routes now.
The audience depends on the market. There are people who run businesses that want skilled ai users to make marketing content that reflects the product accurately and doesnt look like ai slop.
I've done a freelance gig at the workflow level, working on making automated pipelines. That was for a studio whose client was a fashion brand.
I've sold digital images on e-commerce sites. This is just digital products. I cant speak to what the audience looks like there because I am just putting up a product with a price on it and people come upon it and buy it.
Attention isnt a charity. Ultimately it doesn't matter what tools someone uses, the result is king.
The animosity towards these tools, which I only encounter on social media, seems really misguided.
Having studied propaganda extensively, prior to formally studying psychology, no, often there are not. There are malicious actors working to stoke social division in troll farms, in the most antagonistic ways possible.
Propaganda is not about getting you to think one thing or another, its about fostering social division.
The premier means to accomplish this is fomenting a false sense of social consensus, and text based social media is the most low cost way to accomplish this.
Through just a few thousand accounts on a site like reddit or Twitter a false consensus can be manufactured.
This false consensus acts to influence the worldviews and behavior of real people through the psycholgical principle called social proof. https://en.wikipedia.org/wiki/Social_proof
Trolls, by definition, dont actually hold the opinions they claim to.
They arent against Ai. People are discouraging westerners from using ai. The US is inundated with bad faith actors on text based social media.
Like the Maga Trolls and the hyper lefist Trolls, they are largely coming from outside of the country, paid for by countries like Russia and hosted in countries places like Nigeria, Bangladesh, India, etc.
I’m a professional musician and I’ve never heard of this scene that you speak of. In fact, most people I interact with have an opposite viewpoint. Are these communities online or in person? I’m very surprised you’ve made money with AI art, unless you mean you used AI tools to assist you as an artist- big difference.
All these comment sections have anecdotes from "small time business folks" who have seen "amazing efficiency achievements" and then go on to list vague, confusing enough sounding things to get people not to question them.
I have really noticed I don't tend to get specific answers when people tout AI. Im sure there are people out there who have found a lot of uses outside tech its just weird they never seem to post
I work in engineering consulting. We have a hard time getting people to write project descriptions - they completed a project...what did they do? Because when we close out a project and send a deliverable, we move on to the next project. We fall way behind on writing project descriptions and then "what we did" becomes institutional knowledge and not searchable.
We use these for marketing and when we're writing proposals for new work. The fewer we have, the more friction there is and the more time it takes to talk to the right people about "what they did" and then convince them to write a project description.
I created an internal tool that uses AI to scan through a whole project folder, find the right documents, and draft these project descriptions in our templated format.
Explicitly ai generated digital images. Online, E-commerce sites.
I attend in person meetups of artists. One meet-up is gallery artists and professors, people who make installations or work at MOMA or have showings of paintings or sculptures, and for the last year there has been at least one speaker who uses ai explicitly in their creative workflow.
I spoke to a woman at the last one who used ai to make a musical instrument thats controlled by eye movements.
I attend another meet-up thats explicitly focused on the creative use of ai. This meet-up is often populated by vfx artists, developers, editors, directors. It runs the gamut between people who are interested in upscaling a family photo to people who are actively developing tools at the major companies.
Edit to add: I live in a major city, big enough that theres a community for everything.
It's been said that you don't get rich during a gold rush by digging for gold, you get rich selling shovels.
And that's how these AI "tech" companies are making any money - by selling their "tools" the gold diggers who aren't finding enough gold to make back what they spent in tools.
I’m starting to think that the AI bubble/hype was just a vehicle for governments to publicly fund massive data centers to spy on their citizens. When they prove to be unprofitable, governments take them over and then it is too late.
Being 66 having 4.7 million in the market my bag can’t afford to have Ai not pan out. I know younger folks don’t have any hope at the moment but my sack is exposed to Ai. So it’s going to stay steady. 🤦♂️
jayfeather31 | 8 hours ago
This is not what many would call a surprising finding. What I'm more curious about is whether this hastens the popping of the AI bubble or if it stubbornly remains intact.
ThatNewspaperDude | 6 hours ago
The headline is the equivalent of saying “No sign of cancer cure payoff outside the medical sector.”
Like no shit, but tech (and medicine to keep the example) touches every other industry.
nacholicious | 3 hours ago
AI company valuations are dependent on AI becoming a 30 trillion dollar industry, so basically creating so much AI productivity that it rivals the entire US gdp.
You can't do that with just tech
HowdyDiarrhea | 5 hours ago
>“No sign of cancer cure payoff outside the medical sector.”
I don't understand this no comparison at all lol
ThatNewspaperDude | 5 hours ago
If you want to take a super myopic view, a cancer cure only affects the health sector But obviously health affects everything else and other industries.
The article takes a very limited view of AI, restricting it to tech, to justify the headline.
Alt4816 | 5 hours ago
I also do not follow your comparison at all.
The people selling AI were certainly hoping to sell it to more than just tech companies. The dream of executives in every industry was it replacing labor across the board.
CrackJacket | 3 hours ago
Less people dying from cancer = more people around to work, consume, etc.
punninglinguist | 3 hours ago
This is a dumb comparison because AI is explicitly marketed as directly impacting every sector of white collar work: HR, admin, accounting, everything.
Itchy-Professional16 | 6 hours ago
tech/IT is considered a "cost center" in most other industries. if they can cut that, this is no different to them than the price of diesel falling. most people outside of tech would love this.
this isn't curing cancer which saves lives (worth 7-10M per person by US accounting). this is reducing the cost of infrastructure.
semisolidwhale | 5 hours ago
> this is reducing the cost of infrastructure.
Except, apparently it's not
grizzlybair2 | 8 hours ago
All we did is move the bottleneck from A to B but actually it's always been B and business has always been too blind or unwilling to accept B is the problem.
tigeratemybaby | 7 hours ago
Tech sector has always been ultra-fast at adapting new technologies, other sectors will take between 5 to 15 years.
My partner was recently working for an insurance company that had only a few years ago moved away from their paper files to digital, and even then still had rooms of files that they'll never move to digital.
Her job will eventually go to AI, insurance assessor is mostly reading huge files and making a decision based on the laws & policy, but I'd give it minimum five to seven or so years before it happens.
And they will still need assessors, but they'll just stop hiring and let the workforce slowly dwindle down to maybe an 80% reduction in headcount.
This will happen to a lot of desk jobs/white collar jobs, especially the less senior ones, but it will take a decade.
weristjonsnow | 6 hours ago
Agreed. I've worked in wealth management and financial planning for 15 years. At this point I'm a senior planner, but my firm kinda just stopped hiring junior planners because our financial focused AI can spit out a rudimentary plan as a baseline that I modify and tailor quickly. My job is likely not at risk of AI overtaking it in my retirement timeline but a whole shitload of junior planners wont even start their careers, there's no where to start
cupofchupachups | 2 hours ago
So they blew up their talent pipeline? Where are the senior planners supposed to come from?
weristjonsnow | an hour ago
It would appear that is a tomorrow problem, having asked the same question and gotten no real response
Fine_General_254015 | 7 hours ago
I work in the insurance industry. Some jobs will, but any job that needs human connection and human decision making in conjunction will be around for quite some time.
Plus people want someone to blame and blaming a computer isnt the same as another human
user485928450 | 6 hours ago
Position: human scapegoat
Pay: minimum wage
Experience needed: 2-6 years of experience as a scapegoat or masters degree in groveling
slippery | 3 hours ago
How about a masters in masochism? Many years of experience.
InfamousEconomy8677 | 7 hours ago
cope
AwkwardTickler | 7 hours ago
My company cant get AI to scan and process invoices. Something we had automated reliably for many many years. Now we are paying literally millions in temp overhead to both process the backlog and help train the AI. This is pathetic.
OverCategory6046 | 8 hours ago
It's made me money, I've built a bunch of custom tools for my business that save a fair bit of time, have ditched a few subscriptions & generally use it to get a bit more done faster.
I think maybe a lot of corporations just try and shove it in without having a clear use case, which to me seems doomed to fail.
AnAlternator | 4 hours ago
The issue for the AI companies is that in order to pay for the hundreds of billions they're spending to build up their models, they need corporations to widely adopt AI and be willing to pay not-so-small fortunes for its use.
No amount of small businesses paying $20 a month is going to pay for an $800,000,000,000 a year construction binge. That requires mass adoption across corporations, and there's only so long these companies will be willing to eat the cost while awaiting future payoffs.
razealghoul | 7 hours ago
I am not surprised corporations are having trouble. There are tons of layers so getting traction will take a ton of time. Also most folks at corporate level don't have any clue how their tech stack works so getting value out of ai without knowing that will be tough. I work in a mid sized company and I use ai daily and automated much of my reporting. I would say at least 50-60% of my workload has been moved to AI.
Jonoczall | 4 hours ago
If you don’t mind me asking, what’s your role? I ask because ”automating much of my reporting” can mean wildly different things depending on your role and department. Pulling a list of leads and marketing/sales funnel info? Bug reports? Accounts receivables? SOC2 Compliance posture? Etc etc
Ragnarok314159 | 7 hours ago
Because the CEOs that demand it be used don’t know what they are doing. They use CoPilot to summarize emails and then think the same free version of CoPilot can replace engineers and software developers.
razealghoul | 7 hours ago
I just think it's something that will just take a lot of time. Once you have it set up AI is wonderful but it take time to set up the right semantic layers and processes for companies to get value. It not something that is super useful out of the box. In time I have no doubt more companies will catch in but that may be another 5 years away
flbnah | 8 hours ago
Did you need a subscription to anything to accomplish what you did, or did you use/would the open source models have worked?
Ragnarok314159 | 7 hours ago
$20/mo Anthropic subscription will do a lot of coding for you.
The free version of CoPilot is really awful. You can give it clearly defined tasks and gameplan, and it will go off the rails constantly.
OverCategory6046 | 8 hours ago
I did yea, the £20 per month ChatGPT subscription, a Perplexity Pro subscription (I got that for free as part of a promo for a year) & the cheapest Claude sub.
I could have done a fair bit on the free subs, but you run into limits / can't use their smarter models.
Rocketsponge | 5 hours ago
In the lead up to the Dotcom Bubble we witnessed companies getting billion dollar valuations just because they launched a website and promised online sales (I'm looking at you, Pets.com and DrKoop.com). Businesses would be abler to greatly expand their sales and marketing channels through the magic of the internet. The only problem was that the general public wasn't really adopting online e-commerce very quickly. Most people who had internet got it through slow dialup, and it would take another decade before broadband internet access became common. Those high flying businesses with a website quickly ran out of investor cash, causing all those college grads they gave $100k starting salaries to have to put their Porsche for sale along the highways. E-commerce as an idea was viable, but it would take time and refinement before it could really be profitable.
This AI story looks very similar to what we saw with the Dotcom Bubble. I think we will see some of these high flying names like Anthropic run out of investor money and see their debt come due before long causing them to shutter. The real viable and useful tech won't be truly known for several years to come.
PricedOut4Ever | 5 hours ago
I think there is an AI bubble as well.
But your point about accessibility doesn’t make sense. People have access to AI and they have access to agents in a way that dialup users did not have access to e-commerce.
Rather, I think there is a real argument that cloud AI is unsustainable because it’s being too subsidized by VC, that real costs would drive most of the usage today to be self hosted/on-premise, and maybe there will be a catastrophic agentic event that causes governments to seriously hinder frontier labs.
Rocketsponge | 4 hours ago
My point about accessibility was really more to show why e-commerce struggled to get going and the adoption rate by consumers was low. There were other factors too, such as people not trusting purchasing things online back then compared to now.
There are some challenges to adoption of AI now that we are seeing, particularly by people who are afraid of the claims that AI will take their jobs. We also see reports from companies that told their employees to use AI, saw the bills for the usage come in without a commensurate rise in productivity or profit, and then either scaled down or backpedaled. Those things will provide a drag on AI adoption/scaling that may exceed AI firms' cash burn to profit path.
Once_Wise | 2 minutes ago
One of the main reasons e-commerce did not takeoff during the dot com bubble was that people did not trust online payments and trusted easy payments methods were not there yet. We have the same trust problem with AI during this AI bubble. It was not only access issues but also trust issues.
madhewprague | 5 hours ago
Anthropic is on clear path to profitability. Its revenue is going up much faster than the expenses.
nacholicious | 2 hours ago
They are still deep in the red. High growth startups that take on massive debt also need massive growth, but Anthropics growth rate is down by half compared to the last quarter, so their expected valuation was just cut by 25%
Flimsy_Meal_4199 | 3 hours ago
By some accounting they are already profitable as of Q2 in 2026
kariam_24 | 3 hours ago
With billions in debt?
satansxlittlexhelper | 3 hours ago
Yes. Being profitable is what allows you to pay off debt.
kariam_24 | 2 hours ago
Ah so it is imagined money like Tesla stock valuation?
Flimsy_Meal_4199 | 3 hours ago
The difference is that the AI companies are profitable today on their core business.
kariam_24 | 3 hours ago
They are? While bleeding billions or renting to other companies like xai. Unless you call Google, amazon, Microsoft ai companies, well then tesla isn't car company and spacex is ai one.
00raiser01 | 2 hours ago
They aren't, what fiction are you talking about.
livingbyvow2 | 55 minutes ago
"On an adjusted basis" dream world?
Only boring people with a brain live in GAAP reality.
gs101 | 2 hours ago
I think they mean that they would be profitable if they stopped spending billions on R&D. Of course, they have to, or their competitors would overtake them.
Humble_Insurance8294 | an hour ago
They were supposed to, until Chinese Open Source competitors did the most epic rug pull of all time
It was a gigantic, historical mistake for the US to build an empire made out of software, while the Chinese focused on the HARDware.
yeahsureYnot | 7 hours ago
I have a feeling based on personal experience that AI is going to hit the economy in waves and it sort of feels like we’re past the first wave. Lots of people have figured out how to use it to do specific tasks within their existing roles and now it seems like we’re at a lull in finding new/creative ways to use it. Maybe that will change, but for now I feel myself going to it less frequently than a few months ago because I’ve restructured my work around things it helped me build but now that those things are built I don’t need it for the day to day operations.
catmoon | 3 hours ago
There’s a negative incentive for services industries to promote their use of AI. Expensive consulting firms like McKinsey have been using AI but aren’t saying a lot about it because it makes their services seem less valuable. Same deal for legal services, financial services, regulatory services, etc.
bnej | 32 minutes ago
If you can really save a big slab of staff cost with AI then your service has to become cheaper in a competitive market.
Unless there are new goods or services being created in the economy this isn't necessarily a benefit to individual companies.
The trivialised idea that your revenue can stay the same but your cost goes down by a big % is not realistic. Economics just doesn't work that way.
Productivity per employee shot up in the years following Netscape's release, the same trend has not happened since GPT became available.
In software development I can see some tasks become very much faster, but it just bumps into the next rate limit in the org - and it's easy for devs to spend far more than a problem is worth on tokens.
Turbulent-Sign-6067 | 13 minutes ago
Yep, AI outside of computational fields is still so early, so brittle and so weak. But there is no reason why it cannot work. AI in math was brittle until math got (basically) solved. To be fair, math is easy to train on.... time will tell the truth!
spinosaurs70 | 5 hours ago
Frontier model AI costs a ton per person and doesn’t do the whole job nor seem to even be able to replace software engineers.
So the whole outcome of this is a lot of cash burned till TPUs, neuromorphic chips and new non-LLM approaches makes it cheap enough.
Turbulent-Sign-6067 | 11 minutes ago
The fact AI is working in the tech sector suggests AI is coming for other fields too. Everyone who has looked into the matter knows that coding AI is many years ahead of other fields. The only potential roadblock could be training, since teaching AI to do more empirical fields is probably harder than teaching it to work on the computer.
InclinationCompass | 4 hours ago
Not true. Just a couple examples:
UPS’s AI route optimization saves about 100 million miles driven, 10 million gallons of fuel (around $300–400M a year).
Walmart’s using AI and computer vision at Sam’s Club to speed up checkout by 23%, which cuts labor costs and improves throughput. They’re even licensing some of that tech now.
And in healthcare, AI reduced radiologists’ workloads by about 33–44% in mammogram screening, while maintaining or improving detection rates. And AI scribe tools cut after-hours work by 30% and time spent in notes per appointment from ~10.3 min to ~8.2 min (20% reduction) for physicians.
1098duc_w_the_termi | 3 hours ago
These are all traditional applications of ML which are a subset of the AI the article is talking about. You don’t need an infinite amount of data for those applications.
The AI the article is talking about is these overly built out LLMs that’s trying to masquerade as an AGI. The problem is that’s a very expensive way to reach the same goals as specialized ML applications do and have for the last 15 years.
InclinationCompass | 3 hours ago
Straw man. I did not say you need a frontier LLM to route a truck or read a mammogram.
Most of what S&P500 companies are deploying is exactly the specialized ML you described. That's the adoption story.
And the "same goals cheaper" line doesn't hold. Scribe tools parsing open-ended clinical language aren't a 15-year-old ML model. Documentation time dropped 20% because the model handles unstructured language, which specialized ML never could.
The AGI pitch is overheated. That's a critique of the marketing, not the deployment. The numbers are the numbers.
lil_dinger_guy | 2 hours ago
This was either written by an LLM or you spend too much time talking to LLMs. The syntax and diction is very Claudesloppy. “Line doesn’t hold up”
LLMs are what frontier AI research and sales mostly consists of. Other versions of ML that you listed existed before the huge venture capital explosion in OpenAI and other LLM companies. You’re intentionally mixing the achievements of those specialized ML applications with LLMs from companies like openAI to make it seem useful outside of tech
InclinationCompass | an hour ago
Translation: "I don't have a real counterargument so you must be a bot"
How am I mixing achievements? The claim is that there's no AI payoff outside tech. That's a broad blanket statement and does not hold up.
Here's payoff outside tech:
UPS: 100M miles and 10M gallons a year
Sam's Club: 23% faster checkout (they're licensing the tech now)
Scribes: JAMA study shows documentation time down 20%
Call it specialized ML instead of AI if you want. Doesn't change the numbers. LLMs are loud, ML is what's shipping.
The VC hype doesn't change the deployment numbers.
lil_dinger_guy | an hour ago
So not denying being a bot and using bolding like Claude. No point arguing with a bot. You can’t change an LLMs mind because eventually the conversation will fall out of the context window. I can’t change the training data.
So the only point you listed earlier that applies to LLMs is a better note taking app for doctors. Trillions of dollars invested and that’s what we get from LLMs. A software that can save 2 minutes for a doctors appointment. Yeah guys, not a bubble.
InclinationCompass | an hour ago
Oh please, I've been bolding and italicizing on reddit for 15 years. Learn to format some time, you must be gen z.
Nice try narrowing the goalpost. The claim was no payoff outside tech. I gave you 3 examples that contradict that. You decided 1 counts then called it a note taking app and priced the whole industry off it. Miss me with that bullshit.
2 mins per appointment across every doctor everyday isn't small.
>Using an artificial intelligence-powered system to “listen” to patient visits and take notes increased clinician face time with patients and reduced after-hours work.
Neither is 100M miles or 23% faster checkout. All of that leads to lower costs, which leads to higher profits. Scale it out, and it could be BILLIONS saved. Less burnout for doctors too.
>Reimagining documentation, reducing burnout
"I can't change the training data" is a funny way of saying you don't have a counterargument.
lil_dinger_guy | an hour ago
You edited your comment I after I responded. You only had one listed in your reply to me. Also you added back the other ML things that are not LLMs that were out long before venture capital build up.
Still bot slop grammar and diction. You’re making no sense too, because I did have a counter argument. I don’t have to make counterpoints here because you’re just listing the same info and I’m saying that a note taking app for hospitals is not enough deployment outside of tech to justify the trillions of dollars spent.
I can’t engage with you if you’re gonna edit everything you say after I respond to you. It’s one thing to correct typos, it’s another to rewrite sections. Goodbye.
DoorFrame | 3 hours ago
Route optimization wasn’t perfectly doable with pre-AI tech?
InclinationCompass | 3 hours ago
AI did not invent routing but it made it meaningfully better.
The difference is ML improving the forecasts feeding it. Traffic, delivery windows, weather, package volume. The engine is OR, the inputs got way better. Plus drivers actually follow the routes now.
https://staging.bsr.org/en/case-studies/center-for-technology-and-sustainability-orion-technology-ups
mad_method_man | 3 hours ago
map quest would like a word
Honest_Ad5029 | 8 hours ago
Anecdotally, I have made money with AI in the art sector, under branding that explicitly speaks to the use of AI.
Theres an audience for ai assisted or generated work. The knowledge of how to use these tools well is very unevenly distributed.
DeepWaffleCA | 8 hours ago
I'm curious - what does that audience look like? And what kind of media did you generate?
Honest_Ad5029 | 7 hours ago
The audience depends on the market. There are people who run businesses that want skilled ai users to make marketing content that reflects the product accurately and doesnt look like ai slop.
I've done a freelance gig at the workflow level, working on making automated pipelines. That was for a studio whose client was a fashion brand.
I've sold digital images on e-commerce sites. This is just digital products. I cant speak to what the audience looks like there because I am just putting up a product with a price on it and people come upon it and buy it.
Attention isnt a charity. Ultimately it doesn't matter what tools someone uses, the result is king.
The animosity towards these tools, which I only encounter on social media, seems really misguided.
True-Hold-1140 | 7 hours ago
I always find the insistence that the animosity towards x or y is online only very odd. Clearly those are people behind the posts and comments.
Honest_Ad5029 | 7 hours ago
Having studied propaganda extensively, prior to formally studying psychology, no, often there are not. There are malicious actors working to stoke social division in troll farms, in the most antagonistic ways possible.
Propaganda is not about getting you to think one thing or another, its about fostering social division.
The premier means to accomplish this is fomenting a false sense of social consensus, and text based social media is the most low cost way to accomplish this.
Through just a few thousand accounts on a site like reddit or Twitter a false consensus can be manufactured.
This false consensus acts to influence the worldviews and behavior of real people through the psycholgical principle called social proof. https://en.wikipedia.org/wiki/Social_proof
True-Hold-1140 | 6 hours ago
What malicious actors would be against AI? The people most equipped to bot farm are the people that want AI to succeed
Honest_Ad5029 | 6 hours ago
Trolls, by definition, dont actually hold the opinions they claim to.
They arent against Ai. People are discouraging westerners from using ai. The US is inundated with bad faith actors on text based social media.
Like the Maga Trolls and the hyper lefist Trolls, they are largely coming from outside of the country, paid for by countries like Russia and hosted in countries places like Nigeria, Bangladesh, India, etc.
FocusDelicious183 | 7 hours ago
I’m a professional musician and I’ve never heard of this scene that you speak of. In fact, most people I interact with have an opposite viewpoint. Are these communities online or in person? I’m very surprised you’ve made money with AI art, unless you mean you used AI tools to assist you as an artist- big difference.
SpitefulSeagull | 7 hours ago
All these comment sections have anecdotes from "small time business folks" who have seen "amazing efficiency achievements" and then go on to list vague, confusing enough sounding things to get people not to question them.
True-Hold-1140 | 7 hours ago
I have really noticed I don't tend to get specific answers when people tout AI. Im sure there are people out there who have found a lot of uses outside tech its just weird they never seem to post
throwaway00119 | 6 hours ago
I'll post.
I work in engineering consulting. We have a hard time getting people to write project descriptions - they completed a project...what did they do? Because when we close out a project and send a deliverable, we move on to the next project. We fall way behind on writing project descriptions and then "what we did" becomes institutional knowledge and not searchable.
We use these for marketing and when we're writing proposals for new work. The fewer we have, the more friction there is and the more time it takes to talk to the right people about "what they did" and then convince them to write a project description.
I created an internal tool that uses AI to scan through a whole project folder, find the right documents, and draft these project descriptions in our templated format.
True-Hold-1140 | 6 hours ago
Assuming theyre accurate thats a pretty good use
Honest_Ad5029 | 7 hours ago
I sell explicitly ai generated images on e-commerce sites. People buy them. AI is in the name of the brand.
If you never try to do something, of course you'll never see a result.
Honest_Ad5029 | 7 hours ago
Explicitly ai generated digital images. Online, E-commerce sites.
I attend in person meetups of artists. One meet-up is gallery artists and professors, people who make installations or work at MOMA or have showings of paintings or sculptures, and for the last year there has been at least one speaker who uses ai explicitly in their creative workflow.
I spoke to a woman at the last one who used ai to make a musical instrument thats controlled by eye movements.
I attend another meet-up thats explicitly focused on the creative use of ai. This meet-up is often populated by vfx artists, developers, editors, directors. It runs the gamut between people who are interested in upscaling a family photo to people who are actively developing tools at the major companies.
Edit to add: I live in a major city, big enough that theres a community for everything.
kent_eh | 3 hours ago
>No Signs of AI Payoff Outside the Tech Sector
It's been said that you don't get rich during a gold rush by digging for gold, you get rich selling shovels.
And that's how these AI "tech" companies are making any money - by selling their "tools" the gold diggers who aren't finding enough gold to make back what they spent in tools.
Euphoriacsoul | 5 hours ago
I’m starting to think that the AI bubble/hype was just a vehicle for governments to publicly fund massive data centers to spy on their citizens. When they prove to be unprofitable, governments take them over and then it is too late.
Mbstranche08 | an hour ago
Being 66 having 4.7 million in the market my bag can’t afford to have Ai not pan out. I know younger folks don’t have any hope at the moment but my sack is exposed to Ai. So it’s going to stay steady. 🤦♂️