I'm curious, is AI actually more expensive than humans performing the same task, or is that simply what companies like OpenAI and Anthropic are charging for it? In other words, if you bought enterprise-level GPUs and ran your own LLMs, would it then be much cheaper?
"AI" is also thousands of times more expensive than some of the scripts it replaced... so instead of using it a couple times to make better scripts, it has become the script >.>
I don't know who needs to hear this, but no one believed the mass layoffs was because of AI to begin with. AI is only starting to get good at programming and Idk what other office jobs it could replace other than excel only workers...the excel formulas are now easy to do with AI giving them. But most jobs don't use excel as their only part of the job....Most of this post is just about how shitty companies are not how shitty AI is.
Is he stupid or something? Since the invention of making stuff people always tryed to that stuff better or faster or both thats why were here today. 2001 we had a .com bubble 25 years later the Internet isnt even still a thing its everywhere andthey make huge Profits, give ai another 5 years and a lot of people are fucked because they become obsolete
AND the price is only going to go up as they are subsidizing costs to get users addicted and most of those users use it to commit fraud and other crimes
I just can't believe corporations are buying in. If they replace everybody with AI, then the companies selling that AI have an organized power to sell those tokens at whatever rate they want. It's exactly what a union should be doing for people, except in this case, it's consolidating "labor" power in the hands of techbros.
These sociopathic CEOs and upper managers are myopic. They can't see the big picture. They don't understand the social contract. They don't seem to understand even from a self interested perspective that any modest "productivity gains" from AI are meaningless if everyone is laid off and consumers don't have any money to spend. They really think they can just cause another great depression and their business will still be doing well because the AI can do some tasks slightly faster??? Morons.
In my work AI requires heavy oversight. It is constantly changing and growing so the AI response you train today may be messed up by tomorrow. The only way to know is to proofread everything.
A study a sister company did showed that having the best Claude model do a code review costs about 100$ for a 45 minute task, and it found about 80% of the errors. SO a human still has to double-check it. Cheaper models found less problems.
Legitimate question: how does Moore's law faxtor into this? Is it more expensive and less good now? Sure. And in 5 years, when it's 1/8th as costly? And in 15 years when it's 1/1000th?
I saw something recently where they taught AI how to use ‘me speak’ reducing token usage by 70% it basically translated input into how Kevin spoke in that episode of the office ‘why make big word when few word do trick’ it’s that’s true, then this may change…it’s the tokens that make it more expensive
The combination of horrible cadence and disgusting subtitles on this one makes it unwatchable for me, which is a shame because I'm actually interested in the information.
Oh, did the techbro trash spin lies and propaganda for investment money, fail to actually have a 'next big thing' and now have to fake the narrative as long as possible to dodge collapse until their next 'next big thing'? ...Yeah. They do that in a fucking cycle every half-decade or so. And again: None of this shit's AI. Stop passively agreeing with their propaganda as a first step, how 'bout.
Won't happen, but there should be a legal cap, that the CEO cannot be paid more than like 10-20x the lowest paid worker in a company. There would also likely have to be some sort of clause including contractors related to their line of business, because otherwise they'd just go "Well, they're not _technically_ employees..." (it would be an even harder sell to include all contractors, since many companies probably outsource janitorial, etc.)
I'm just suggesting that the minimum wage isn't actually what you want to fix. It's just a minimal goal within the current system, and as long as there is inflation it will always have to be revisited.
Right, let's make people more expensive to employ and bring the relative cost of AI down to like 60% or maybe even only more expensive than 50% of humans! That way AI can be justified and used even MORE by businesses. Good thinking.
You are so monumentally fucking ignorant that if you have any actual intelligence and ability to learn this is going to be something you still cringe over 20 years from now.
Take several minutes to look at real world data. Common sense is bullshit. It's almost always wrong. Without hard data and experimentation "logic" can "prove" anything.
Look at facts, not what you "think" makes sense, or what other people whose politics you agree with tell you.
That was literally decades ago. It's time to institute a MAXIMUM wage, that includes all benefits.
Absolutely no one should make over 10 times more than the lowest paid employee. Also, having interns should be illegal, and there should be strict guidelines, time limits, and compensation minimums for contract workers.
On top of that, your 1st residential AND business property up to 2mil/5mil should be tax free, while additional should be heavily taxed exponentially, as should wealth over 10m.
It was always more expensive. AI services were (and mostly still are) predatory priced products with companies selling the product below cost, subsidized by the AI investment bubble.
Once the bubble pops and the bill comes due for consumers to pay the real price of the real estate and resources used to create a marginally useful tool, it’s gonna suddenly be about who has the most efficient LLM and AI agents, instead of who’s got the biggest most complex models.
AI can only ever even sniff being cheaper than a human when it's cheap or free. If it costs what it costs to run it, it is suddenly more expensive than most any other option. Also, AI is very rarely given a project it can fail at alone and take the blame for- there's usually humans who either have to fix bad output, or just throw up their hands and do it manually. But because an AI was used, it was an AI project.
I am constantly astounded by how myopic and stupid these CEOs and upper management types are. So their plan was to replace workers with AI, which is MORE expensive, AND more error-prone, AND once they've all done this, nobody would have any wages to spend on products or services that their business produces. This weird abstract notion of "productivity" is completely dependent on consumers having money to spend, they really can't see that?
Also, while finding and training staff is difficult, they're kind of over a barrel for compensation without a union. If a big chunk of your labor force is provided by a company that has little incentive to keep working with you, they can suddenly start charging anything they want.
Which is what it's useful for. I hate seeing those "help the blind to read" AI commercials. Seriously, nobody is against that. But that's not what you're actually selling. They're seeing a slop bot that can't get it right even plagiarizing everything in the Internet and evening off the Internet they can get their hands on.
I heard the same thing 25 year ago about a thing called the Internet.... and about 18years ago about houses and still both are bigger than they ever were 🤔
"But AI can ____." Ok, show me, show me the best selling movie, tv show, application, book, game, artwork, etc. that is 100% AI. You can't because it doesn't exist. You can show a lot of rephrased mediocrity, but so what? And it doesn't CREATE it rephrases! All you're getting is a sampling of the internet reworded in a conversational format. The information already exists people are just too stupid to find it.
but but but what about all this 4/10 porn and extremely formulaic, bland, soulless writing? Surely the world needs an ocean of that at all times, right? It has to, because we spent trillions of dollars on this and so it has to be worth trillions of dollars. It can't only be good for scams and cheating on your homework!
You're probably never going to find literally 100% AI, because like any other tool it's being used and directed by a human, but it still depends on what you mean by that. Neural Viz on Youtube I believe mocaps his face and voice acts, but then casts his own voice to others using AI. Can that be called 100% AI or not? Now, Gossip Goblin is probably closer to 100% AI, but it still involves re-running the same gens until you get a good one and manually video editing them together.
I have yet to understand how LLM is more useful than a google search, to me it seems actually worse than a google search... and the slop artwork everyone is using for their posters, fliers, restaurant menus etc always looks really bad. I think the real use is for misinformation, further muddying the waters so we can't tell what's real anymore so that the fascists can consolidate their control by sowing even more chaos and confusion.
Recent studies have found that Google's AI overviews are correct about 91% of the time. Do you find correct/useful information more often than that with a Google search? Personally I've found Google searches less and less useful these days due to how they bend everything to advertisers.
The machine learning algorithms train the models on rule patterns for assembling words into coherent language. It can assemble words and concepts in unique ways based off those patterns in grammar, syntax, context, and conceptual associations. Saying it is just rephrasing training inputs is an inaccurate oversimplification. It can reproduce some content from its training data based on those associations, but it's not limited to that.
The data it outputs, the actual facts, are never original. The surrounding text may be novel, but it's literally just rephrasing a delivery. And try getting it to come up with a 100% original joke. Then check if the output is ACTUALLY original, and then ask if it's ACTUALLY funny.
It can reproduce facts learned from training data, but it can also generate novel falsehoods not from its training data. LLMs have also recently started accurately working out novel solutions and counterexamples to math conjectures that were previously unknown, which runs counter to your claim. Generative AI models are also used to discover novel protein sequences and figure out how they fold geometrically in aqueous solutions. They're also used to construct novel drug candidates to trial.
Did you lose track of your own argument? You're arguing that it can't create anything new. The fact that it can create novel falsehoods shows you are wrong on that point. Of course it can't think. You then ignored every other point that I made about generative AI currently making actual novel discoveries using learned patterns because you cherry picked one point that you thought you could use to argue against me. I'm not even going to bother if you can't argue in good faith.
The whole concept that a chat bot will suddenly stop lying at some point at be reliable, when the core design is to confidently lie, is madness. There are lots of good uses for different algorithms, and even a few for LLMs, but they are not, and should never have been thought of as "general purpose"
It's currently cheaper than humans at nearly a quarter of use cases and compute gets cheaper and faster every year, but you're saying that not only will it not get cheaper to run, but that the tech itself will not advance beyond its current capabilities? I mean I get the hatred, but that borders on delusion. The entire history of technological advancement runs counter to your argument.
AI is not profitable. It is subsidized for customers while AI companies burn through investor cash and magical accounting. It will collapse long before it approaches profitability.
Tech is generally logarithmic rather than exponential though. Eventually the early explosion evens out and gives diminishing returns requiring novel developments in tech and technique to eke out improvements. The idea isn't that they'll stop improving, it's that the explosive development is going to even out once they've sacked the library of human knowledge enough that new training data provides trivial boosts and the only way to improve outputs is to build out an infinite increase in compute.
All this knowledge and power concentrated into single places... and all we do with it is tell it to make tree tiddy anime cat girls? Lets cure some diseases...
unclesporky
I'm curious, is AI actually more expensive than humans performing the same task, or is that simply what companies like OpenAI and Anthropic are charging for it? In other words, if you bought enterprise-level GPUs and ran your own LLMs, would it then be much cheaper?
nullbr
For high managerial work like CEO's is where AI would offer a better result for the price
freakdiablo
I think a big thing to point out is HOW it costs more than human workers.
Like in my experience, you still need a human to sit and babysit the AI. So why not have the person do the work in the first place...?
J3lek
The end goal of capitalism is to put all the wealth into like 20 bank accounts all owned by the same person.
StrangePaegan
Hahaha fuck them companies
CuntCrotchula
Sauce is Damon Cassidy for those curious
WeeemRCB
God, his voice is annoying.
pleaseuseotherdoor
https://media0.giphy.com/media/v1.Y2lkPWE1NzM3M2U1OHcxbDZrbWtnYW1lYjZzenRxeXYxbnllY2s3cjBzMnZ3eGM3Zng3diZlcD12MV9naWZzX3NlYXJjaCZjdD1n/l0FecnkPKOOveZNrq/200w.webp
IGetAllMyFactsFromHere
"AI" is also thousands of times more expensive than some of the scripts it replaced... so instead of using it a couple times to make better scripts, it has become the script >.>
teddybearsucide
I don't know who needs to hear this, but no one believed the mass layoffs was because of AI to begin with. AI is only starting to get good at programming and Idk what other office jobs it could replace other than excel only workers...the excel formulas are now easy to do with AI giving them. But most jobs don't use excel as their only part of the job....Most of this post is just about how shitty companies are not how shitty AI is.
Grinch01
Is he stupid or something? Since the invention of making stuff people always tryed to that stuff better or faster or both thats why were here today. 2001 we had a .com bubble 25 years later the Internet isnt even still a thing its everywhere andthey make huge Profits, give ai another 5 years and a lot of people are fucked because they become obsolete
redundancy
MIT study is called "Beyond AI Exposure: Which Tasks are Cost-Effective to Automate with Computer Vision?" https://futuretech.mit.edu/news/beyond-ai-exposure-which-tasks-are-cost-effective-to-automate-with-computer-vision
laDisparitionDesHeures
January 2024?! That's ancient!
youreathing
Thanks for the source, which @apeofdeath didn't want to post for some reason
Toqom
AND the price is only going to go up as they are subsidizing costs to get users addicted and most of those users use it to commit fraud and other crimes
Sunflier
sweetandsmokey
But it's faster.
RevolutionOnHerLips
I just can't believe corporations are buying in. If they replace everybody with AI, then the companies selling that AI have an organized power to sell those tokens at whatever rate they want. It's exactly what a union should be doing for people, except in this case, it's consolidating "labor" power in the hands of techbros.
channelranger
Well the trick is that they do get human labor if they got what they wanted. It's just that what they want is slaves.
Rhythmaster
These sociopathic CEOs and upper managers are myopic. They can't see the big picture. They don't understand the social contract. They don't seem to understand even from a self interested perspective that any modest "productivity gains" from AI are meaningless if everyone is laid off and consumers don't have any money to spend. They really think they can just cause another great depression and their business will still be doing well because the AI can do some tasks slightly faster??? Morons.
willpostanything
Laughing in AI
toogoodtomiss
In my work AI requires heavy oversight. It is constantly changing and growing so the AI response you train today may be messed up by tomorrow. The only way to know is to proofread everything.
CASchoeps
A study a sister company did showed that having the best Claude model do a code review costs about 100$ for a 45 minute task, and it found about 80% of the errors. SO a human still has to double-check it. Cheaper models found less problems.
DyspronsiumAddict
Legitimate question: how does Moore's law faxtor into this? Is it more expensive and less good now? Sure. And in 5 years, when it's 1/8th as costly? And in 15 years when it's 1/1000th?
Is there a future without AI?
I (sadly) struggle to see it.
KnowGoYesStopWin
I saw something recently where they taught AI how to use ‘me speak’ reducing token usage by 70% it basically translated input into how Kevin spoke in that episode of the office ‘why make big word when few word do trick’ it’s that’s true, then this may change…it’s the tokens that make it more expensive
AnonsAndMinions
The combination of horrible cadence and disgusting subtitles on this one makes it unwatchable for me, which is a shame because I'm actually interested in the information.
ElbowDeepInAGoblin
Oh, did the techbro trash spin lies and propaganda for investment money, fail to actually have a 'next big thing' and now have to fake the narrative as long as possible to dodge collapse until their next 'next big thing'? ...Yeah. They do that in a fucking cycle every half-decade or so. And again: None of this shit's AI. Stop passively agreeing with their propaganda as a first step, how 'bout.
SavageDrums
Big tech is broken, and it is 100% because they have business idiots running everything now.
Leave Microsoft, leave Google.
ThatOtherGirlYouKnow
https://media0.giphy.com/media/v1.Y2lkPWE1NzM3M2U1dzBvdzByOWpvdWR3MzgxNmF5czk2d2Y2aTBwbHQ2NDE0OXlsYXg2NyZlcD12MV9naWZzX3NlYXJjaCZjdD1n/PXCBR9byJHSUw/200w.webp
ElbowDeepInAGoblin
Oh, what does he know. I heard he's his own grandpa!
ThatOtherGirlYouKnow
https://media0.giphy.com/media/v1.Y2lkPWE1NzM3M2U1N3dkMHYxMGM2YTJwdnFhaXd5c3B3dnhyZThoM3FhbHJnZDBtMmZweSZlcD12MV9naWZzX3NlYXJjaCZjdD1n/4ibbiDYA023Fm/200w.webp
rhinokitty
Time to raise the minimum wage.
MrOne2
https://media1.giphy.com/media/v1.Y2lkPWE1NzM3M2U1ajEwbWJrZWxnOWZ5ejdxd3AzMHY2Ym54M2RpZ3E4NDhqbm42YmVwNSZlcD12MV9naWZzX3NlYXJjaCZjdD1n/h0MTqLyvgG0Ss/200w.webp and all the billionaires went
idiotsonfire
If the min wage had kept up with CEO bonuses, it'd be $44/hr now.
RainierAlt
Won't happen, but there should be a legal cap, that the CEO cannot be paid more than like 10-20x the lowest paid worker in a company. There would also likely have to be some sort of clause including contractors related to their line of business, because otherwise they'd just go "Well, they're not _technically_ employees..." (it would be an even harder sell to include all contractors, since many companies probably outsource janitorial, etc.)
VagabondEngineer
Time to replace the idiot CSuite with AI for actual productivity gains....
IfOnlyICouldUseMyPowersForGood
Funny but please don't actually do this. Replace them, yes. With AI, no way.
KuldFyt
Time to cap executive compensation and introduce clawback penalties for short-term overperformance that evaporates after bonuses are calculated.
tgiaa
Time to enact a maximum wage
cxmpxsxd
This... but good luck getting it elected.
SemperPi
I stopped reading at “cap executives”. Sold!
Travelcedric
The time to raise minimum wage was decades ago
kaoticgirl
time to burn down Wall Street
khora
Sweden doesn't have minimum wage.
honeybutterbiscuits
This wouldn’t work in America. In America, no minimum wage = free labor to big corporations
khora
I'm just suggesting that the minimum wage isn't actually what you want to fix. It's just a minimal goal within the current system, and as long as there is inflation it will always have to be revisited.
TheGgreen100
Right, let's make people more expensive to employ and bring the relative cost of AI down to like 60% or maybe even only more expensive than 50% of humans! That way AI can be justified and used even MORE by businesses. Good thinking.
jasondeslin
You are so monumentally fucking ignorant that if you have any actual intelligence and ability to learn this is going to be something you still cringe over 20 years from now.
Take several minutes to look at real world data. Common sense is bullshit. It's almost always wrong. Without hard data and experimentation "logic" can "prove" anything.
Look at facts, not what you "think" makes sense, or what other people whose politics you agree with tell you.
idiotsonfire
Republican propaganda detected.
NumptyDumptyLump
Fuck your percentages. AI is going to kill us. Do you care about corporations or people just trying to get by? It's humanity. Not slopbots.
jasondeslin
That was literally decades ago. It's time to institute a MAXIMUM wage, that includes all benefits.
Absolutely no one should make over 10 times more than the lowest paid employee. Also, having interns should be illegal, and there should be strict guidelines, time limits, and compensation minimums for contract workers.
On top of that, your 1st residential AND business property up to 2mil/5mil should be tax free, while additional should be heavily taxed exponentially, as should wealth over 10m.
Easykehl
It was always more expensive. AI services were (and mostly still are) predatory priced products with companies selling the product below cost, subsidized by the AI investment bubble.
Once the bubble pops and the bill comes due for consumers to pay the real price of the real estate and resources used to create a marginally useful tool, it’s gonna suddenly be about who has the most efficient LLM and AI agents, instead of who’s got the biggest most complex models.
channelranger
AI can only ever even sniff being cheaper than a human when it's cheap or free. If it costs what it costs to run it, it is suddenly more expensive than most any other option. Also, AI is very rarely given a project it can fail at alone and take the blame for- there's usually humans who either have to fix bad output, or just throw up their hands and do it manually. But because an AI was used, it was an AI project.
Rhythmaster
I am constantly astounded by how myopic and stupid these CEOs and upper management types are. So their plan was to replace workers with AI, which is MORE expensive, AND more error-prone, AND once they've all done this, nobody would have any wages to spend on products or services that their business produces. This weird abstract notion of "productivity" is completely dependent on consumers having money to spend, they really can't see that?
astrangehop
Also, while finding and training staff is difficult, they're kind of over a barrel for compensation without a union. If a big chunk of your labor force is provided by a company that has little incentive to keep working with you, they can suddenly start charging anything they want.
NotSomoneElse68
Which is what it's useful for. I hate seeing those "help the blind to read" AI commercials. Seriously, nobody is against that. But that's not what you're actually selling. They're seeing a slop bot that can't get it right even plagiarizing everything in the Internet and evening off the Internet they can get their hands on.
Grinch01
I heard the same thing 25 year ago about a thing called the Internet.... and about 18years ago about houses and still both are bigger than they ever were 🤔
comeonjointhejoyride
Like, i get that its the people that are gonna pay the price because, always. But how? How will they make us pay?
losersmanual
Wages down, utilities up.
vanishinggirl
I’m for AI replacing every CEO, CFO or whatever stupid acronyms they use for their actual useless “job” that they perform!
DYLANLEE79
"But AI can ____." Ok, show me, show me the best selling movie, tv show, application, book, game, artwork, etc. that is 100% AI. You can't because it doesn't exist. You can show a lot of rephrased mediocrity, but so what? And it doesn't CREATE it rephrases! All you're getting is a sampling of the internet reworded in a conversational format. The information already exists people are just too stupid to find it.
channelranger
but but but what about all this 4/10 porn and extremely formulaic, bland, soulless writing? Surely the world needs an ocean of that at all times, right? It has to, because we spent trillions of dollars on this and so it has to be worth trillions of dollars. It can't only be good for scams and cheating on your homework!
unclesporky
You're probably never going to find literally 100% AI, because like any other tool it's being used and directed by a human, but it still depends on what you mean by that. Neural Viz on Youtube I believe mocaps his face and voice acts, but then casts his own voice to others using AI. Can that be called 100% AI or not? Now, Gossip Goblin is probably closer to 100% AI, but it still involves re-running the same gens until you get a good one and manually video editing them together.
Cyanide555
Also google has made finding it way fucking harder. Ads, sponsored results. Making the "" and - not work correctly.
DYLANLEE79
Yeah the Boolean operators are now just "suggestions".
Rhythmaster
I have yet to understand how LLM is more useful than a google search, to me it seems actually worse than a google search... and the slop artwork everyone is using for their posters, fliers, restaurant menus etc always looks really bad. I think the real use is for misinformation, further muddying the waters so we can't tell what's real anymore so that the fascists can consolidate their control by sowing even more chaos and confusion.
unclesporky
Recent studies have found that Google's AI overviews are correct about 91% of the time. Do you find correct/useful information more often than that with a Google search? Personally I've found Google searches less and less useful these days due to how they bend everything to advertisers.
Rhythmaster
Yeah I'd say that's about the same for me. When I google search something I can find what I'm looking for about 91% of the time.
JustAName00
The machine learning algorithms train the models on rule patterns for assembling words into coherent language. It can assemble words and concepts in unique ways based off those patterns in grammar, syntax, context, and conceptual associations. Saying it is just rephrasing training inputs is an inaccurate oversimplification. It can reproduce some content from its training data based on those associations, but it's not limited to that.
DYLANLEE79
The data it outputs, the actual facts, are never original. The surrounding text may be novel, but it's literally just rephrasing a delivery. And try getting it to come up with a 100% original joke. Then check if the output is ACTUALLY original, and then ask if it's ACTUALLY funny.
JustAName00
It can reproduce facts learned from training data, but it can also generate novel falsehoods not from its training data. LLMs have also recently started accurately working out novel solutions and counterexamples to math conjectures that were previously unknown, which runs counter to your claim. Generative AI models are also used to discover novel protein sequences and figure out how they fold geometrically in aqueous solutions. They're also used to construct novel drug candidates to trial.
DYLANLEE79
Repeat that first sentence.
JustAName00
Did you lose track of your own argument? You're arguing that it can't create anything new. The fact that it can create novel falsehoods shows you are wrong on that point. Of course it can't think. You then ignored every other point that I made about generative AI currently making actual novel discoveries using learned patterns because you cherry picked one point that you thought you could use to argue against me. I'm not even going to bother if you can't argue in good faith.
MightyIink
You mean it was a grift all along for big companies like Nvidia to try and sell you something? You don't say!
GolbatChan
They laid people off because they expected a down turn. They used AI to take the blame and boost the stock price.
c85fkbvw9r
No, the goal is to get it cheap enough and good enough to replace people. It's just not going to happen right away.
Nim449
It is also not going to happen later on too. The whole show is a sham.
DMSledge
The whole concept that a chat bot will suddenly stop lying at some point at be reliable, when the core design is to confidently lie, is madness. There are lots of good uses for different algorithms, and even a few for LLMs, but they are not, and should never have been thought of as "general purpose"
JustAName00
It's currently cheaper than humans at nearly a quarter of use cases and compute gets cheaper and faster every year, but you're saying that not only will it not get cheaper to run, but that the tech itself will not advance beyond its current capabilities? I mean I get the hatred, but that borders on delusion. The entire history of technological advancement runs counter to your argument.
youcouldbedrinkingwholeifyouwantedto
It currently is not if you think the post is accurate. Why would companies rehire layoffs if AI was cheaper than humans?
DMSledge
Cheaper in the short term but still makes mistakes all the time, because it isn't thinking. It is a chat bot.
Malekith899
AI is not profitable. It is subsidized for customers while AI companies burn through investor cash and magical accounting. It will collapse long before it approaches profitability.
AndroidSoul
Tech is generally logarithmic rather than exponential though. Eventually the early explosion evens out and gives diminishing returns requiring novel developments in tech and technique to eke out improvements. The idea isn't that they'll stop improving, it's that the explosive development is going to even out once they've sacked the library of human knowledge enough that new training data provides trivial boosts and the only way to improve outputs is to build out an infinite increase in compute.
stonerhino
All this knowledge and power concentrated into single places... and all we do with it is tell it to make tree tiddy anime cat girls? Lets cure some diseases...