A.I. Companies Are Losing A LOT Of Money
HTML-код
- Опубликовано: 7 июн 2024
- Free and premium investor resources, check out our website: www.differentiatedanalytics.com/
In this video we take a deep dive into the generative AI industry and question its economic viability.
For inquires related to Differentiated Analytics email us at: founder@woodandwoodenterprises.com
Check out our second channel Broken Business Models where we discuss unusual or otherwise suspect businesses that may be unviable: / @brokenbusinessmodels
For business inquires: Mary@creatormanager.co
For other inquiries: Wallstreetmillennial@gmail.com
Check out our new podcast on Spotify: open.spotify.com/show/4UZL13d...
All materials in these videos are used for educational purposes and fall within the guidelines of fair use. No copyright infringement intended. If you are or represent the copyright owner of materials used in this video and have a problem with the use of said material, please send me an email, wallstreetmillennial.com, and we can sort it out.
#Wallstreetmillennial
------------------------------
Buddha by Kontekst / kontekstmusic
Creative Commons - Attribution-ShareAlike 3.0 Unported - CC BY-SA 3.0
Free Download / Stream: bit.ly/2Pe7mBN
Music promoted by Audio Library • Buddha - Kontekst (No ...
------------------------------
0:00 - 1:23 Intro
1:24 - 5:39 Estimating costs
5:40 Can AI companies make money 
Кино
selling the shovels wins again
Nvidia stock to the moon 📈
@@random-po6er If the users of the shovels don't find gold, then the market for shovels will eventually dry up.
@@mattiasolsson2499 Crypto Bros will come up with another stupid idea
@@mattiasolsson2499 but at least you sold some shovels and made a profit. If you never find gold the only thing you did was lose money on shovels.
@@mattiasolsson2499but this isn’t any shovel though…it’s also a spoon.
This whole investors-backed economy thing is starting to feel more like a den for gambling instead
Yeah it's certainly some of the dumbest economic activity we've ever seen. Honestly it feels like a bunch of impatient toddlers with a short attention span are running everything now it's incredibly frustrating they're going to get bored of this in a minute and then it'll be some other bulshit scam.
Investing has always been about gambling. Everytime you take a risk it's gambling to some extent
I been saying that about the stock market and 401k’s for years
They are gambling, just "informed" or more realistically FOMO gambling
it is, gambling with a glasses and a turtleneck on.
The point of most AI companies is not to make AI. It's to get more investor than try to sell themself to a bigger company.
That was the point all along lmao
Bigger than Microsoft and Google and META?
AI is not cryptocurrency...
AI does have a legitimate potential.
Even as it is (rough on the edges) it produces value.
It might be over hyped but it certainly has an intrinsic value.
Problem is to find the right business model.
Lots of tech startups exist to sell themselves rather than turn a profit
@@here_be_dragons9184 i imagine reducing power usage has to be a priority
Yeah every stock that has ai in it instantly gets a premium evaluation. Real bubble energy here.
Get in while you can.
@@carpediem4512 no don’t
Get in, get out, then short sell like hell
It's not even just stocks. EVERYTHING is more expensive when it has "AI" in its name.
I played Apex Legends for many years. Over the years you see some wild cheater shit. I'm talking people having rapid fire, flying around, using abilities of other legends (like a Lifeline shooting a grenade with Fuse's launcher???) etc.
So i looked around what these cost. I wanted to know what people pay. Some only pay like $1 or $2. But other cheats also cost like $20 PER MONTH. Why? Well because their aimbot is supposedly "100% non invasive" because it supposedly uses AI to determine enemy player positions which is complete BS because the videos showcased hardcore aimbot as well as so called "silent aimbots" (aimbot where your bullets always hit the enemy even if you aim god knows where). How tf is AI supposed to achieve that, huh? But doesn't matter. People read AI so they buy and pay way more.
Only a matter of time before this goes the way of every financial bubble that came before it
“Hard times creates efficient production, efficient production creates large wealth, large wealth creates reckless spending, and reckless spending creates hard times.”
lmao nice
Stealing this
Mo money, mo problems?
Original author?
@@jonasbaine3538 He made that up based on the popular saying "Hard times create strong men, strong men create good times, good times create weak men, weak men create hard times"
So AI is going to reduce the number of human employees but you need more employees willing to use AI to bring the price point down to an acceptable level. Yeah...what could go wrong? (Cue the Bladerunner soundtrack)
Cue the Terminator soundtrack 😂
@@doalwa Nah. That soundtrack is way too upbeat. Nothing says we are doomed like music by Vangelis. 😏
It will increase the number.
I dont understand your reasoning. You need more employees willing to use AI to bring the price down? Nope. The AIs will do the work, the employees ill become the unemployees
@@chad0x
What exactly is this work that AI will do?😂😂😂😂 I'm a warehouse worker, AI isn't piloting autonomous vehicles so i guess white collar people are sweating bullets😂😂 These
Typical startups these days - burn money with little or no plan to make the firm profitable..
Cause most start up CEO are tech bro not business people.
@@khanhnguyen-tt3ff I'd say AI startups are a bit different than the old unicorns like uber and doordash. Those their biggest cost were labor* which never really goes down. Compute costs do go down overtime tho espiecially when every hardware company is working on AI accelerators
Like Temu?
@@daniyalkhan9856 but it's also very competitive, so it doesn't mean declining costs is automatically increased profit.
while 10-20x in share value
The only one making money off ai is NVIDIA
That’s true currently, but R&D always has upfront costs before it results in a profitable or useful product.
But, just like with the mRNA vaccine went through clinical trials in less than a year once the right profit/social motive was in place, AI’s receiving even higher levels of interest, attention, and spending.
Most of the startups are gonna fail and be a waste of money, but the few that succeed are gonna be so successful that they more than make up for all of the other one’s losses on net.
Unless current AI methodologies end up being technological dead end, which i think is unlikely.
Some time in the future, there will be a reckoning for the startup millionaires who are lining their pockets. If they manage to fleece the wrong people, it may get ugly.
For now. They’ll figure it out. Just like with every other market.
Sell the shovels, don't mine the gold...
And they are making hand over fist.
It’s so much fun to ask copilot stupid things because you know this is costing microsoft actual money.
🤣
I still wouldn't because you're literally burning through local water supplies in order to do it.
@@Freakattaker Data centers don't waste the water they use on cooling. It recirculates just like your car radiator. What they use alot of is electricity.
@@tripplefives1402I was a contractor for Microsoft and I helped setup up a data center in Phoenix (or somewhere near Phoenix) and it used evaporative cooling to keep the data center cool. Idk if they changed the design because it seemed like a dumb idea when I was working on it 5 years ago. The amount of maintenance required was crazy. I can’t imagine their water bill lol. Granted, I’m sure they were given a significant discount. I know Bank of America got crazy good deals on their energy bills in Charlotte. I’m sure that’s most big companies.
@@tripplefives1402 google AI use of water
AI Image generation is starting to look like Self Driving Cars to me. It looks great at first impression, but the more you look the more realise it's not ready and probably will never be. Sure there's lots of people using it to pump out cheap logos (and porn, so, so much porn) but any professional or even semi professional client goes to a human. My worries about being able to find commisions have basically gone away, a few clients have told me they tried AI and realised that the number of revisions they needed outweighed paying for a human they can just hand a brief and get a ready to use product 7 days later.
Yeah, generative AI as it currently exists is mainly a tool, not an agent. A CGI/Photo editing professional can use various AI tools to save time in his workflow, but it can't work without the professional. Just like digital drawing and editing cut down the time a professional needed to create or edit an image, so will AI.
Where have you been the past year? The exponential improvements that AI Image generation has made over the last year are absurd. It’ll be replacing quite a few markets in the coming years. Especially artists.
I can tell you I've been using AI for software programming and it works pretty good.
Sometimes it hallucinates stuff that doesn't exist but most of the time it's on point.
Sure you don't cut and paste code as is but it brings you ideas and solutions and most of the time they are fine.
Even for an artist AI can provide shortcuts like, ask for a forest background, and then have the artist focus on character, or have the AI generate the initial picture and then have an artist do some touch up to fit your exact needs.
Use AI for storyboards or proposals for client... There's a lot of potential...
@@camdowgExpect the typical business concerns regarding trademarks, copyrights, legal accountability, and intellectual property protections to undermine the impact. Either as a matter of due diligence or because Disney wants the government to crack down on the AI systems that are generating a thousand fresh porno images of Mickey Mouse every hour.
@@reappermen yes, things like Dall-e and midjourney are tools and not agents, the human using them is an agent. Its just that things like ChatGPT or Gemini get progressively more agentic
"Why don't you just pay artists or secretaries... let's say $30?" - people
"Ackhyually, our $2 Billion Server Farm could do better to democratize said skill for $5 a month...
per service per subscription per version per brand per Megabytes per user as per EULA that we can change at any given second." - Microsoft.
$30 an hour is $50 TC, which is $2k a week and $8k a month…
Instead hire someone for $15/hr, no benefits. Give them a $100/month subscription to premium, tailored-AI. Same output as the $30/hr guy for $2500 a month.
@@akkikishore3770you are scaring me with this one wow
@@akkikishore3770Congratulations, you're a ghoul.
I don’t imagine AI getting very far before it starts consuming itself. It’s like play dough. Keep adding colours and it becomes brown.
Part of me kind of wants to make a github account and start posting absolutely terrible code for solving x problems. Like "here's a really efficient sort" that's really just bubble sort, or includes unecessary computations to make it even worse than bubble sort. If AI is going to plagiarize everything, let AI suffer the consequences of plagiarizing my lousy code.
What a fucking line 😂
😂😂😂
@@FireMageLaynI wonder how they actually train models to distinguish between good code and bad code. My short experience with AI code-generation shows that it really don’t know the difference between working and non-working code, without getting into good vs bad.
@@maniak9005 it probably has soemthing ti do with how many people are commenting on it, or perhaps how many other people are using it. Which means errors can be introduced into the model.
I work for industrial software company… it launched several AI/ML. tools for manufacturing facilities… all are running in net loss… the value is not just there for such big investments for clients..
My biggest concern for these companies is differentiation. In my experience, there's not much difference between LLMs when being used as productivity tools for things like coding and writing. Similarly to ride sharing we may end up in a cash burn fuelled race to the bottom as near identical services drive each other toward a cliff's edge.
People thought it would replace the need for personal vehicle ownership 😂
where competition on quality ends race to the bottom on price begins ...
AI reminds me of those bums years ago that kept saying “BIG DATA” oh my god the next big thing, then it was “INTERNET OF THINGS” oh my god next big thing- all scammers. Anyway screw these people.
don't forget nfts, blockchain, cloud, ar, vr, metaverse, self-diriving, fusion, ... the list is truly endless, so much tech hype so little results.
@@chimagamer4157 right its as if jump on the hype get investor cash and then disappear…
@@chimagamer4157 lol little results? did you not see the price of Bitcoin it's at a very high price you mean high results.
@@chimagamer4157i mean it's not just vaporware empty hype. Self driving cars, big data, AI, cloud, Blockchain, fusion all have their own uses and are useful and will be even more useful in the future.
@@chimagamer4157 Cloud technology is actually very useful. The rest are either mediocre or just straight up scams (nfts lol).
All I hear are more dumb ways to monetize that will cost us as the general consumers more in the long run. Which is exactly the problem with damn near every new invention. It only costs the general public when it should be benefiting us.
and the joke is, the public funded the initial research into the technology.
@@chimagamer4157they still do. Most of the model architecture is still done by open source or on grants. Obviously the data on training as well. The only thing that isn't purely public is the hardware but one may argue CHIPS act etc could be public funding for hardware.
Generative AI, cryptocurrency, fully automated vehicles. All solutions looking for a problem that doesn't exist, and everyone is perpetually chasing one after the other hoping to catch the next instance of lightning in a bottle. I think we've hit the point where technology has advanced too fast and we've looped back into the "throw it at the wall and hope it sticks" phase.
I think it's obvious what solutions AI crypto and self driving vehicles solve.
If you read Bitcoins White Paper, the problem it is trying to solve is having to go through govt or banks to exchange money. It does in fact solve this.
@@lightningwight4154 well it could theoretically. in reality its niche in real life applications and the coin exchanges just are the old banks now and its super volatile...
" looking for a problem that doesn't exist". Yeah, no need everyone becomes an artist
producing pictures or videos
.
@@lightningwight4154It doesn't solve squat, no matter how much it says it wants to.
It simply cannot scale up enough, it's hugely wasteful, slow, volatile and wholly dependant on an infinite supply of new users who will buy in no matter the price.
AI is like a pandora's box that everybody wishes they hadn't opened, but now they all have to pursue in order to not be left behind by the competition and their enemies.
Not really. They are all developing AI products because that's what investors are throwing money into. As soon as the bubble pops most of them will completely stop working on AI.
Its called race to the bottom or Moloch, where everyone would like to stop, but they cant cooperate due to fear and thus have to continue
More like a bag of burning dog poo on the door step.
except when it helps fix problems that would take a human with paper and pen years or decades... and does it in seconds to minutes
@@Wary_Of_Extremes but its not solving any meaningful problems, is it?
Ai just wants to make art, tell jokes, make music samples and scrape the web for easy derivative research.
You can't devalue entire industries then expect the same industries to boom.
Interest Rates will determine how long investors will have stomach to fund money losing investments.
The money is the people making the AI chips, the cables and power equipment for the grid and the power plants. It's another gold rush where the people that sell the shovels are making the most money.
Sam Altman is a venture capital huckster, like Danny Zappin of Maker Studio. Dumping millions of dollars into an ambiguous tech startup, run by a 30+ college drop-out has a "Sam Bankman-Fried" written all over it.
I mean AI is not like crypto or NFTs, bad comparison
To Sam Altman, "AI" are just two vowels he can milk. @@TheManinBlack9054
I disagree. The products are absolutely amazing. Just not market fit.
@@TheManinBlack9054 The hype cycle is the same, unfortunately, this cycle heavily influences the direction in which companies take this technology.
That's not true. ChatGPT Plus works and so does Gemini. I mean, you can try it yourself. It's not perfect, but it's impressive...in a frightening way, but impressive nonetheless!
0:51 And that is the danger of AI coding: an overly strict validator that many devs will accept without question. It’s mindless copying from StackOverflow on steroids.
one point you missed is in improvements of the models. For example, Mistral LLM 7b model, outperforms old LLama 13b model. In other words, you get better performance with smaller model. ChatGPT does similar thing when they unload questions to different models. Models will get more efficient, so we have to wait an see.
It's going to be interesting to see where A.I. companies are in 5 years from now
Depends on what kinda AI the company has, I can see a weather AI can earn money in 5 years
Most of them will be sitting in the gutter with the NFT Bros.
"going to be"
🤔
will be ?
should be ?
@@fix0the0spade what makes you so certain?
@@TheManinBlack9054 They're charging $10 for a product that costs $20 to deliver, then still needs human input to make useable. It isn't the same problem as NFTs (they're a flat out scam) but sooner or later a business has to make a profit. If they triple prices it's cheaper to skip them entirely, making it cheaper requires using less server time, which makes the product worse and it might be more efficient to skip them. It isn't like the Dotcom bubble either, that happened because everyone bet on internet access getting cheaper faster than it did, but it did get cheaper. AI seems to be getting more expensive with each generation. Eventually the majority of the AI players will decide they've lost enough money and give up.
You mean running supercomputers 24/7 for kids to write grade school essays with isn't cost effective? Who could have guessed
Until we get hardware that is better optimized for AI workloads (eg neuromorphic hardware, not just a million graphics cards) it's difficult to see how/if/when this whole sector will become profitable.
I think much like the internet it'll become integral to daily life (it kinda already has), but the profitability is only gonna be in a handful of a few players who are successful, like OpenAI. They'll eventually find a way to cut costs and increase revenue when people become more dependent on it.
However there's already a whole bunch of CEOs making AI claims or pivoting just to try to pump the stocks of said CEOs
I was thinking the same thing. How much longer can we expand this processing envelope? Fill the whole world with servers?
@@althunder4269 We've already filled the whole earth with servers decades ago and we even have some in space.
Real neurons don't actually work the way AI people think they work. They are more like frequency multipliers.
@@tripplefives1402The whole earth? I think there is still some way to go.
New technology exists, hype as investors pour in hoping that the horse they bet on becomes as big as the internet, more investors pour in as marketing hypes it up further, products fail to meet marketing expectations, products fail to meet initial expectations, products struggle to find an actual use case, hype dies, onto the next thing
Did you say the same about the internet or the smartphone?
@@TheManinBlack9054 How many companies fell in the DotCom Bubble? How many smartphone manufacturers exist today vs when the hype for it was at an all-time high?
It's not that the underlying technology is bad, we just have a lot of trash companies that need to be weeded out, just like in the DotCom Bubble.
@@TheManinBlack9054Did you forget about the Dotcom Bubble?
@@TheManinBlack9054 ai isn’t the the internet or the smart phone more like crypto and NFTs
My kneejerk reaction is to not be too surprised or broken up about this. I'm sure there will be real-world usage for AI sooner rather than later. But right now, companies are promising some wild stuff. And the AI is cheerfully hallucinating its drivers off. Good times indeed.
Bruh chat-gpt has been a real world application?? That's like saying the Internet hasn't affected the world yet
I hope it goes nowhere
Yeah, I think right now we are at the stage where loss leading providers are trying to get their stuff out there in the hopes that people will find uses for it. But for the moment it is more a solution looking for problems than a real self sustaining ecosystem.
@@sobbski2672 The only thing ChatGPT has revolutionized so far is spam
investors after years of QE will put their money in anything due to FOMO.
Looking at the pricing of all the AI companies, I've long been wondering what the long-term strategy is going to be. People in Western economies can maybe spend 50 Euro a month extra on AI services. Nobody, absolutely nobody, will pay 20 bucks each to ten different companies offering various AI services. Either is all integrated into current software, with maybe one or two extra subscriptions, or it's not going to fly. Regular people don't *need* it enough at the moment.
The long term strategy is parasitism.
Case in point is OpenAI's licensing "deal" with the Associated Press, in which OpenAI scraped all the AP's shit off the internet, trained ChatGPT on it, threatened the AP's market, and then eight months after release announced that they were "partnering" with the AP to, basically, wedge OpenAI products into the AP workflow and also shake them down for any data they might have hidden away from OpenAI the first time.
They didnt tell *anyone* how to block their scraper spider GPTBot until months after launch. It's so transparent it makes me furious just thinking about it.
Because most of them suck... just like the meme coins
Meme coins don't suck they just don't do anything at all 😂
Love your content bro, you're very good at this!
He's AI too
My roommate has convinced himself AI is going to be the next microchip revolution. He thought the same thing about self-driving cars. Before that it was hypersonic flight.
I have far less optimism for the future.
Your roommate is correct, extremely correct
I read all the glowing AI headlines but my local Walmart can't accurately tell me how much soap they have in stock.
This was an amazing video explaining the current hype around AI, the best video so far. Although there were lot of technical terms discussed, it made clear sense to me. Thanks and keep up the good work!
This video didn't explain shit about ai. He listed 2 examples and then tired to justify his opinion.
I work in design. Entire work teams are gone because of ai. Entire work flows in businesses have already been replaced. Everything you use when you open your phone to click on Amazon, click on shit on Google, every other ad you see has pasted through ai. The idea that ai isn't ready to be used is the dumbest shit because it's already been used and optimizing things you are not aware of.
The luddite left's grudging acceptance of the need for nuclear energy being driven by AI was not on my bingo card.
I'm amazed that the deep thinkers on the right are starting to understand why renewables are going to power everything in the near future.
😅
They probably already know that the military has been doing it forever…
This is just what I needed to see. The AI bubble is ONLY getting started! Thanks WSM
The AI plan: a one-person company (CEO) typing prompts and making billions of dollars... The issue: in a world where nobody has a job ( because they were all stolen by AI) how is the economy going to work?
The reality: one person (CEO) typing a command, AI responds, and many people check whatever AI produces for accuracy
@@maniak9005actual reality one person (CEO) of company funded by friends and family, goes bankrupt after a year after not finding any paying customers.
But the AI is also going to be the CEO
One angle that is missing in this video from my POV is that Nvidia's margins are >50% and that a significant amount of downward pressure on them by inbound competition will significantly change the dynamic.Another thing to consider is that companies are getting smarter at how they train and how they invoke models. AI today is essentially a brute force solution, we have discovered a hard way to do something valuable, but daily those edges are being chipped away significantly. My prediction is that compute will become rapidly commoditised whilst models will focus on being niche and performant instead of general and taxing.
Glad to hear it.
Every time I stumble upon this channel I get the feeling this voice/narrator is made by AI.
He is, its no soul..l can hear and feel it.
Aye, I believe it's a pretty slick AI that's been well edited. It doesn't take breaths.
like most of these companies, there will be a consolidation on applications. similar to how only 1-5% of tech start ups actually become sustainable
Just wait until the lawsuits really start gaining traction, AI text to image like Mid-Journey might have to pay UBI to the artists they illegally sourced data from globally, the consensus being a universal basic income paid to artists upwards of 20 years, the period copyright is expected to protect from theft before falling to the public domain. AI companies violated the rules claiming the data harvesting was for "research purposes" only, and then they monetized it for profit anyway... any artist with work published online before the data was collected will be eligible...
more likely they go bust before that, courts are notoriously slow, until then you might have the next bubble burst.
The lawsuits are all thrown out and chances are Disney will also just gonna use AI
Every artist 'sourced' from a bunch of other artists...unless they were the very first artist ever and had never seen art.
I hope that when the AI bubble inevitably pops, at least some of the companies that genuinely were trying to further the field of AI (not just profit) remain standing.
There are plenty of AI/ML applications in boring industries, such as around non-destructive testing of physical items (=scanning them and analyzing for defects). The low overhead, high value companies will probably make it.
Liquid neurons, superconducting noise-surfing analog chips, optical, lot of approaches neglected in favor of silicon. I really don't want an SMR powered data Center in town.
whatever job you do for money, I hope people come along and do it for free so it's not just done by you for profit...
When the Internet bubble popped, the Internet still existed. If AI does the same thing it’s still better to be invested than not in the long run.
@@henrytep8884 According to Claude, there is too much frothy hype in the AI space and investors should be wary. Yes, the irony was delicious, an AI telling me AI is overrated.
I worked as data scientist for 4 years Switched from data scientist to software engineer job last year. I don't want to be there while this hits the fan.
Hardly any c level exec understands the math behind the solutions. There is very few actual research going on - it's hard. Throwing money won't make algorithms smarter.
Similar: Getting into Business Analysis made me quit chasing a job in ML! Nothing quite puts a dampener on things like reality.
Hey do u think ai /ml engineering good ??
@@mirzaabdullah9006 Nope!
@@mirzaabdullah9006 it's good but EOD remember it's algorithms or framework that will be part of a product.
You can't keep researching without a good product brings money.
Generalize your skillset...don't specialize
Companies trying to put cart before ox, it will not end well. The execs are pumping stocks and cashing out.
And Microsoft is the last company that will innovate...hahaha. anyone who works there can corroborate, it's a diversity hire/ govt company...like IBM.
Nvidia is commodity hardware that is comically overvalued. The most popular ML/AI code libraries only supported their cards because they were the easiest to program (the code looks like C/C++). Everyone works in high level languages now (python etc) and the libraries are not hard coded to nvidia anymore (see google’s JAX). This is yet another hype bubble.
I mean rn most of big tech realized that cutting out nvidia chips infavor inhouse chip, like tpu would definetely save them tons of money in a long run.
@@Sapose11 the issue with that is that Nvidia is very very good at what they do. Apple invested Billiosn for their own chips and they still cant compete with Nvidia.
@@nox5555Google has been buying Broadcom's TPU's and very soon, so will Meta, as Broadcom's CEO, is now on Meta's board of directors.
"Everyone works in High level languages now" Implying C++ are super low level language means you never work with C++ before lol. Most low level stuff already have own libraries depending on the company. Just like Unreal C++, you don't even have to bother garbage collection. Unless you working with Hardware, which is hey, something they are heavily working with.
Python is not usually used in hardware programming you see.
@@youtubedeletedmyaccountlma2263after CS i worked in the games industry for 13 years writing C++. now work in python doing ML/AI. you should listen more and laugh less at this early stage of your career.
As long as it’s not everyday people losing money chasing wild promises then I just say meh. Big time investors can afford the hit if they’re wrong.
Friend of mine works in VC in UK had a AI company on there books (£2m fund) a US big VC called and asked to put in just because it matched their portfolio criteria and they had/have to spend the allocation of funds. I hate this whole world!
all that waste is crazy
@@PGG1996 I know right…meets criteria ok let’s put several million in just because.
It costs $300,000 to raise a child from 0 to 17 years old. $300k should be the target upper bound cost on the most performant model.
I donno what you've been feeding your kids, but my son only cost 10k. lol jk
@@BillAnt 😂😂
My cost was ~$45k. I dunno where $300,000 comes from. May be it's just US thing.
@@alexm9104 it’s a published average. $45k? Did you raise your child in a tent in the woods?
@@Alley00Cat I was raised with such a sum, not my children.
$45k are the approximate expenses my mother spent on me from my birth to adulthood.
The thing is, I was born in the '90s in Moldova and then lived for 25 years in Russia.
Even now, you can live for three years in a rented apartment and spend approximately $12k here in Moldova.
US, North America and west EU are not the whole world you see.
can't believe I've missed yet another bubble.
The services I use to do AI image generation used to let you generate images free. Now they're using credits. They claim it's very taxing on their computers when so many images are generated. Everything starts out free and then starts having a cost, and then you realize their business model isn't really much a business model as the costs climb. I wonder how this will affect all the rhetoric these business leaders have been pushing about "AI will eat all jobs." Not if the costs are that high for you, it won't.
0:49 That regex is wrong, and slow.
It doesn't include a country code, and it's unnecessary to store dots and dashes.
Just format your input and do ^[0-9]{15}$
And yes, I am being petty.
Probably want to allow a plus, white space and don't yanks use brackets?
@@IsleyNumber1 Personally, I'd just strip all that out before I even tried to validate it. If you're sending the number to a user, you can add the syntax bits back in, in a way the user is familiar with (depending on their country)
Because most of them suck... just like the meme coins.
Reminds me of the eSports League of Legends scene. The teams were never profitable. It was all investor money 😂
marketing money paid for those teams
Those teams were never to make money in the first place just marketing
The amortization exercise you did of the NVIDIA Al chips by their customers is interesting . I like that you explicitly called out that the required revenues you calculated were just to cover the 2023 investment.
I wonder what the long term annual capital investment in these chips will be. I suspect that the long term investment and associated annual amortization will be greater than the $3.5 billion you mentioned. NVIDIA shareholders seem to believe it will be significantly higher that the $40+ billion spent in 2023. If that is the case, AI revenues would have to be absolutely massive to make these investments profitable.
The big thing is those models require constant training. It's not just a one time upfront development cost - it will be a constant expense not reflected in the gross margin.
Perhaps they should invest in research on quantum computing.
AI has far more potential than quantum computers
@@TheManinBlack9054 quantum computers could literally crack most forms of encryption within seconds, fym "ai has more potential."
Centralized ‘cloud’ AI isn’t what end users really want, it’s great that they can provide the service but ultimately people will want to run their own choice of models and that’s the best competition, and probably better use of massive data centers. In theory it should be possible for a cryptocurrency network to let people lease the computation power of their machine into a network running AI models, in fact I’m pretty sure I have an old version of a network that even used specialized AI hardware as ‘miners’ doing basically that sort of thing.
Oh boy, a giant attack vector for spreading viruses, fantastic!
WsM decided to tackle the question of the century, brave effort.
Assuming we aren't all annihilated by an H bomb...
AI is seen by the players as global dominance, in which winner takes all.
So not surprising they are betting big.
Once the technology matures, normal computing AND web use will be as irrelevant as steam power.
The skeptical tone of this report (as we all love WsM for the schadenfreude aka trainwrecks...) is premature in this case imho.
Aside - it will also condense political power even more in the hands of a tiny few, so big societal shifts coming too.
So far he is right. Unless something changes, he wins.
And just like how the Zelda C.D-I games said: Good.
Wow. Cd-i owner here (not used for 27+ years now). Cheers from Portugal
@@luiscunha9393 Cheers from Brazil
These tech companies have billions but you don't have food or house.
Wake up.
I have both. I think you're projecting. Maybe spend less time on the internet and more time upgrading your skills so you can afford food and a house.
@@allanshpeley4284 Sure buddy, you are just an embarrassed future billionaire ;)
@@netanelaker4437 You don't need to be a billionaire to put food on the table. Not even a millionaire. Stop making excuses for your situation. It's time to man up.
@@allanshpeley4284 I hope a billionaire would see your comment and will let you ride him.
Yeah, it will probably be another decade before AI starts being profitable because not only are there issues with copy write, but the tech isn't quite there yet. There defiantly needs to be more regulations for AI. I think it would be a good idea to make it so that anything made with an AI needs to have some sort of indicator that it was made by an AI in order to prevent scams and the spread of misinformation.
Another decade? Do you do any research on how it is used. This video mentions like chatbots and web services and that's it as if a ton of ai hasn't been used yet.
What do you think happens when you open Amazon to buy something? How do you think the icons that show you something come to being? When you look at reviews. There are some millions of reviews. What do you think reads those reviews? When advertisements occur are they random? No.. ai does those are things. There are so many uses other than just a chatbot..
Yeah, that's a fair point. I was thinking more about the AI mentioned in the video when I commented. By the way, what do you think should be done to regulate AI?@@tomk5238
Doing the lords work my friend
10-year amortization sounds generous. I'd think 3-5 years is more realistic based on how quickly the models are changing
Chatgpt kinda sucks, you have to fact check everything it says to the point where you’re really just doing your own research with a partner confidently spewing incorrect guesses at you. It just feels like this will be like the self driving car hype. It’s a neat party trick for inconsequential stuff but we are decades from this stuff hitting the road
I dont really think its decades, years maybe
Might be longer than decades. Legal wrangles aside, there are no adas 4 cars in the market, some claim adas 3, but they are adas 2 with some added features. The major adas 4 vehicles I see are trains, which were available 20 years ago.
This is the same with these 'AI' companies. LLMs will not be the core behind generative ai whenever it is created.
Just recently, Google Gemini REFUSED to create an accurate image of a WW2 German soldier or a Founding Father. Pathetic. The biggest problem with AI are the humans building it.
No duh it's going to have flaws it's not 100% perfect just like with many things but it's going to improve as time goes on. Mobile games get updates to fix glitches and other problems.
@@huckleberryfinn8795thats a hidden prompt issues which can only be fixed when Google admits its a religious cult and literally delete a few lines of code.
AI is not just a product, but a platform and a tool that can enable and enhance many other products and services. AI is creating new markets and opportunities for innovation and growth, not just for big tech companies, but for small businesses, entrepreneurs, and consumers. The video also overlooks the positive social and environmental impacts of AI, such as improving health care, education, entertainment, and sustainability. It's an investment that no successful company will be able to ignore; thus, they aren't losing money, they are buying the bills they need to survive.
I like the comparison to the early internet. The real AI revolution starts once any kid can start an AI Company in their garage, just how any kid can start a web company now. That's when real innovation and exponential growth begins. Three companies doing only the most obvious projects won't accomplish too much, what we really need is cheap AI hardware that is sold to consumers.
GPU, not more costly than early PCs, plus most of your kids already have them at home in their gaming rig.
but if we take the internet analogy, you are looking down a further 10 years until it gets actual mass adoption, and that only happened because of smartphones. all the heavy lifting has been done in the last 10-20 years by government funded institutions, just like with the internet.
But you can already do just that (renting hardware from the cloud). And not many revolutionary innovations are in sight. Unless you mean training a brand new LLM (why?), that will currently cost you $80K for a one like the original ChatGPT (down from $1M a year ago).
You did not mention the “cost” of the raw-material or data which OpenAI & others of their ilk like Microsoft think is “free to use” to feed their power-hungry behemoths. That hardly makes any sense & NYT has a lawsuit going on this matter. Apart from this fundamental matter is another crucial one. This is about the “verifiability” of the end-result & not only is this impossible to do so, the resulting super-confident assertions even when apparently hallucinating makes it 100% unreliable for any serious enough mission-critical application. So while it’s human like grammatically correct output looks impressive on the face of it, it is less than useless in applications that demand truly verifiable levels of confidence. So while impressive to wow a party crowd, 100% unsuitable to make a life or death decision such as a Nuclear first strike. If indeed one has a death wish, better to play Russian roulette. Little capital expense to buy a revolver & no power hogging is an added advantage.
Correct, AI will be limited to entertainment for the masses as well as clue/idea generation to seed further research by professionals. Not replacing said professionals any time soon, contrary to what the peddlers are pretending.
Amortising enterprise hardware investments over 10 years is... optimistic
Agreed, I'd imagine the hyperscalers will be selling off their H100s to smaller players as part of their upgrade flow to Blackwell architecture - so not the old ways of running stuff into the ground.
One thing you forgot: gpt4turbo are 10 times cheaper than the old model. See also LCM image generators also SDXL turbo are fast and cheap and running in consumer computer.
I can't help but think that LLMs like GPT will inevitably mature into increasingly more open source and self-hostable tools that will be just another part of a programmer's toolkit. Think of just how many technologies exist currently that are freely available and open sourced, that's what happens with most software: the hype dies down as the technology becomes so ubiquitous that everyone just uses an increasingly robust open source version most of the time. There's so little that people need to pay for in the present day compared to the 90s and 2000s.
You can already self-host models equivalent to ChatGPT 3.5 turbo on your local PC. The main problem is that they are just not that useful for anything much, really.
The reason why the AI stuff will not really ever be truly open source is that the training is so costly. Training new LLM from scratch todat would be $80K in hardware rent alone, but you need to add to it expenses for obtaining and filtering data (depending on what you want to do, this can become really, really expensive - the LLMs have the "luck" of feeding from publicly accessible Internet), possibly initial tests with different model architectures, alignment and benchmarks aimed at your application, possibly retraining again if anything turns out wrong... this is not something a big bunch of individuals can realistically do with small contributions of their spare time. And it involves a lot of meticulous work which is not very fun to do (another exclusion factor for open source projects).
So what is your guess for stock investors during this high price phase of gen AI?
Once the bubble pops, buy the survivor.
Nothing will ever be bigger than the internet, unless of course its some kind of superinternet!
internet down 10k years with Nx10^12 of people using it.
On Super Earth, we will have the Super Internet, to spread democracy!
I am glad.
I called it last year, the AI hype is just that, and there are few real world use cases that are financially viable.
Someone asked on my LinkedIn last week about how i'd separate the hype from a realistic AI investment. I replied but it felt futile because I know in my heart that nobody gives a shit about realism, they want their minds blown by far-out theories and they want to fritter their money on piss / the next 'big thing', so unfortunately everything I have to say will fall on deaf ears. That's fine if i'm being paid to give advice that'll be ignored, but f*** doing it for free on LinkedIn.
Those GPUs depreciate, and it becomes an arms race to keep up. If the end markets don't value the output, you're stuck on a treadmill just buying chips to keep up with the next startup blowing money on chips.
Well yea, cause VCs are giving script kiddies like Devin that just stitched together open source projects behinds a cute UI 150mil
A new AI winter incoming.
All I can see is over promise and under deliver.
The actual cost of AI is what we need to answer the question about AI replacing human labor properly. If the cost is still high, it would be better to just hire humans for now.
I am not a fan of the "business model" and think it is most definitely hyped far past what it can do as far as money it saves. That said, it may be a good idea to at least consider the likelihood of the discounted usage is actually providing vaule as far as training, testing, and other data/feedback. Just the proprietary data on how people use it, how often they use it, and if/how it actually ends up helping, would be an extremely valuable product too! That doesnt put it in the green by any stretch, but just to be fair.
I think your ".com bubble" analogy is likley accurate too. Gamblers gonna gamble.....oh Im sorry, I mean; "respectful, successful, productive, stock traders" are gonna trade.
It's not just the money in itself. It's also a dominance game. Any company that develops powerful AI, their metaphorical stocks will grow.
Profitability is so 20th century, its all about market domination and monopolization now, with investors making big bets the company can become dominant in the market and raise prices in the future. Many companies only really care to sell themselves to a larger company anyway.
Remember self driving cars threatening to end the trucking industry 10 years ago? Most of the tech bro hype is just that, hype. You don’t get investors excited by telling the truth…
Self driving cars failed because of their potential of killing people on the road / or regulation. Ai chatbot and other form of AI do not have this problem at all.
Hard to see path to profitability with a monopoly supplier... save me AMD, you're my only hope!
Prolly cuz none of them are achieving actual AI
A lot of this AI stuff, is just tulip hypermania... Like NFTs and other such crap before, it will all run out of steam soon! Sure... Some practical use cases for AI will remain... But the rest will evaporate into irrelevance!
Just a matter of time... Before we get a HUGE correction!
Very good video. AI businesses and investments might be in a bubble. The business model will be viable but only after many years of cost reductions.
Dude just used AI to generate the thumbnail while taking traah about Aai 😂
Don't forget the Ai voice
Would it be possible to cover GPU depreciation models? 10:11
A clear win for AI companies is not in what they're selling, but in what they are using the data sources, the customers, to develop and may later use themselves.
The issue is that there’s so much speculative investment money right now with so little profitability expectations that nobody is able to charge a fair price for the ai. Maybe 5-10 years from now will be when investors will expect returns
Actual information on the aggravate investments for artificial intelligence is always interesting, although investors having aar involvement was acutely irresponsible.
I wonder if anyone has done a trend analysis of % failed ventures of AI vs historical "paradigm shifts"
Like were railroad or oil speculation incinerating cash at the same rate with the same number of scams
I'm not quite sure what you mean, but of the AI ventures that haven't failed so far (90%), their average returns are a piss-poor 0.3% - 4.2%.
@RocketPropelledWombat there are categories of technology that promise to change the world, 5 years ago it was crypto, 20 I was the internet, 70 and it was nuclear power, 150 and it was railtoads, etc
All of these examples did create things that changed the world, but for all of them there was a massive speculative investment cycle filled with scams, look up the proposed Ford with a nuclear reactor onboard
I want to compare the hype and wasted money in those bubbles with the current one, because the AI one feels about as scammy and scummy as crypto
Surprised nvidia hasnt named one of their line of data center gpu's "the shovel"
I've yet to see large value add from LLM-s that would justify the hype. Sure, it has a tiny value add for a software developer. But so do most of the plugins in an IDE. They earn some money, but are not billion-dollar companies.
Instead of AI systems that do everything and, by nature will be huge models with very large compute resource needs, the most likely path to profitability will likely be systems very narrowly tailored to meet the needs of specic tasks.
Also, refinements to modeling will very likely greatly increase efficiency and thus dramatically reduce infrastructure requirements and costs.
As was pointed out, we are at the very beginning of the industry and improvements will come over time.
Just had an mazing example of just how far away generative Ai can be from the end market. Canva has both text, picture and video gen AI, apart from text which most of the time needs edits, picture and video AI are far from usable apart from cases where you need simple graphics (which it already has in its library) or pictures without any toes, fingers, people holding things etc. So far it has been a benefit to designers who actually know and have tools to edit and tweak AI designs, but not the end users who have neither the know how or the tools to do so.
AI adoption is all well and good until confidential information enters the chat. Professional services companies (think McKinsey, KPMG etc) at whom products like copilot are being aimed, are likely to not adopt AI simply because of the huge risk to confidential information being used to train those AI systems. The fact is they’ll all look to build internal tools instead which they won’t do until the costs come down to the floor.
Its the ultimate dumb market glitch, future growth driven investment meets possible exponential growth product
I think these companies are racing to create real productivity gains, for the world to change you need bubbles