AI has 4% success rate in automating work. LOLThis is what VCs and mag 7 are gonna burn $600 billion on this year. AI vegans BTFO
>>1081441184% already is huge. That's a ton of people, and it's only going to improve.
When intelligence meets artificial stupidity
just keep running it until you roll the 4%
>>108144118thats what happened when you have shitty prompters. they need to to go prompt school
>>108144337Does it make sense to spend billions to automate the 4% of easiest tasks? And even that is shaky because any customer must accept a 96% failure rate, so they always need a plan B and enough time.
>>108144478You're assuming 4% is the highest it will go
>>108144486Only 100 trillion more and 50% of tasks can be automated!
>>108144486This 4% was accomplished through billions upon billions of bubble funding, mass hoarding of components, and years of marginal improvements that are obviously plateauing. I'm not going to deny that it can improve, but it will take much more time and actual innovation.
>>108144494ai improvement is logarithmic so 100 trillion will only yield like 6%
>>108144486Yes, 6000000 million trillion billion more and it might get to 4.1%
>>108144118>source: it was revealed in my dreamsAI bad channels are so cringe
>>108144337>and it's only going to improve.and it's only going to get exponentially more expensive to improve it
>>108144337>>108144486We've been hearing that since 2022, and text based AI is basically the same since then. Granted, the visual side has definitely exponentially improved.
>>108144337>4% already is huge.consider how stoopid your average workforce is - you are talking about spending billion$ to replace bottom of the bottom tasks. Its burning cash to scrape barrels.
>>108144466mfs be like>do my job for me, mr computerno you have to be more specific
posts the increasingly nervous luddcuck for the 100th time in a row
>AI has 4% success rate in automating work. LOLSounds like a skill issue LMAOI've had no problems automating mine.By the way, the desperation is really palpable from how many of these AI hysteria threads you're making.
>>108144337>and it's only going to improve.saar it can't even generate profit
>New technology>Has to be 100% successful right off the bat or it's a failureWe would still be living in caves
>>108144921We will all be living in caves if AI succeedsHow many CEO interviews where they dream of overpricing and depopulation do you have to see to realize this
>>108144921None of this technology is aimed at improving your life. It will replace you, make you valueless and ultimately lead to you being discarded by the victors of it all i.e. the owner class of AI.
as long as LLMs pull data out of their asses they cannot be trusted and should not be used for anything critical. it is somewhat comfy to have AI summarize technical documentation, but anything beyond that is a fucking gamble.
>>108144999The same goes for 90% of software developerst. performance engineer
>>108144999And once they can be trusted ? Also even if they have 5% failure rate, being cheaper and faster will beat humans any and every time, it just needs to be good enough, not perfect.
>>108144494yeah. back in 1953 or whatever, one megabyte cost 6200$. your 1 tb harddrive of cost 6.2 billion dollars. weve gone from "it cant even finish a sentence" to "it writes a novel" to "its bad at pictures" to "oh itll pump out classical paintings in 13 seconds" to "oh it also do full motion video sequences from a 1 sentence prompt" to-yeah, its likely we will look back at today and go "damn, those primitive savages. they spent 100 trillion dollars to get just 4% success? today we create their entire worldly knowledge they had at the time for like, 3 dollars in like, 40 seconds"
>>108144921Would you be happy to take a flight in an airplane where it's success rate is 4%?
>>108145058when the wright brothers took off, the success rate for powered flight was 0%
>>108145066But would YOU willingly pay and take a flight?
>>108144118only 4% next year that'll be 3% then 0% then we'll all stop using AI and go back to hiring all those capable zoomer graduates
>>1081443374% automation != 4% workforce replacement
>>108145054Who will be we ? You think the billionaires will keep you around ?
>>108144921No one said that, but it's not even 20% LOL
>>108145078>social upheaval followed by covid then AI hysteria We pranked the zoomers pretty good
>>108144889Na even if you're specific the shit fails. I use Gemini + Claude+ ChatGPT at the same time to get shit done. Coording them is a full time job. i write the code myself now and just use them for code review or as a glorified search engine
>>108145058>Would you be happy to take a flight in an airplane where it's success rate is 4%?YesIf I had been around during the dawn of powered flight I would def have flown airplanes and I more than likely would have crashedWhat a way to go
>>108145066>when the wright brothers took off, the success rate for powered flight was 0%incorrect. Every mother fucker was building a plane at the time and some took off before they did.
>>108145083im not a billionaire, and i have a terabyte hard-drive. the phone i have in my pocket would be basically godlike to someone in 1952, and im not a god now.itll be the same shit it was every decade. what seems expensive to us today, will be trivially cheap tomorrow.
>>108145095You can still do the same thing today. Don't be a bitch. Being a test pilot or joining the military is still a thing
>>108145113Only that the phone didn't replace your economic value, this technology will replace your economic value.
>>108145102okay. find the guy who took off first, and apply the same thing. a drone would of cost all of americas gdp at the time the wright brothers or whoever the fuck you wanna find first tried the first powered flight. today you can get one at whatever replaced toysrus for like 35 bucks for one that fits in your hand or some shit
>>108144118What does success rate in automating work even mean? God damn, I don't want to fall for the AI hype because a lot of the people pushing it are the same retards who would be holding NFTs, but somehow the opposition is even more braindead.
It's funny how people here really went from ultra mega fedoratipping atheists to just going full mask off and worshipping LLMs as a phony god with zero irony
>>108145123sure. hell, seems likely. doesnt...really have anything to do with what im saying, tho. those people whose livelyhood was to walk up mountains and bring down ice? devastated by the invention of the refrigerator/freezer. technology moved forwards anyways.
>>108145078Holy based
>>108145155And once all white collar jobs go the way of the ice man ? And once robots and everyone else is competing for whatever is left in terms of blue collar work. Then again I ask you, why would anyone keep you around using up resources if you provide zero value ?
>>108145125no one spent all of American's GDP to make planes and drones though. Most of the shit we have today was supposed to be made to kill Nazis, unfortunately they died too soon and we used to scraps to build Zion.
>>108145162i dunno man. either they figure it out quick enough, or we all die, i guess.again, doesnt change what im saying or anything. people arnt going to stop this, or slow it down, or anything. we just fuckin yolo that shit. maybe it kills us this time
>>108145147>tips
>>108144515Just leave the money printer on till it gets to 50% easy. If it fails now itll take down the entire economy
>>108144396Just run 1000 ai at once for the same task and whichever result has the most repeat wins.Easy peasy
>>108144118They are just getting started, cashier
>>108144337Americans are given do-nothing jobs to keep them busy & from civil unrest.Automating a do nothing job is not an achievement.
>>108145274Easier than telling euros hey cant have FREE MONEY for the hard work of other nations.
>>108145147No, I get what you mean, once you notice how many posts just feel like reskinned rapture logic you really can't unsee it
people on the technology board absolutely amazed that new technology isn't immediately perfect
>>108145351>people on the technology boardall 20 of them kekroflmao
>>108144337i don't think you understand how percentages work
>>108144494>>108144505>>108144515>>108144530What if 90% of the work is getting it to go from 0% to 4%? That's how a lot of technology works. Getting it to accomplish *anything* is the hard part. Once it's reached that, you're just making it more efficient, which is easy.But let's pretend like this technology won't advance like every technology before it.
>>108145122>Don't be a bitch.Air travel is the safest mode of travel in the United States you fucking moronAlso, I'm an aviatorXD
>>108144337Only 70 trilliton dollars are needed
>>108144118
Based on the most recent global economic and employment data for 2024–2025, here are the estimates for the total number of jobs, the quantity of labor that 4% represents, and the associated financial cost.1. Total Jobs in the WorldThere are approximately 3.7 billion employed people in the world. Source Data: The World Bank and International Labour Organization (ILO) estimate the global employed labor force is roughly 3.7 billion as of 2024. This includes both formal employment (salaried jobs) and the informal economy (self-employment, day labor), which makes up a significant portion of the global workforce.[1]2. Amount of Labor (Automating 4%)If you were to automate 4% of all labor, that would be equivalent to: 148 million jobs (Full-Time Equivalents). Context: This is roughly equal to the entire labor force of the United States (approx. 168 million) or nearly the entire population of Russia.3. Annual Labor Cost (in Dollars)If you assign a monetary value to that 4% of labor, it would be approximately $2.35 trillion per year.
Why are people still in denial about AI? It’s pretty obvious that Claude Cowork will actually replace some jobs. It’s only going to improve. Accountants are flat out denying it and still encouraging people to major in it on Reddit. How can people not see this?
>>108145416Do you actually use LLM? How long have you been actively using LLM?
>>108145416Yeah unfortunately for you this isn't how it works. LLMs just guess the next token in a series of tokens. There's a limit to how good you can guess something, and every wrong guess can instantly brick the work.LLMs are a dead end if having actual AI which can do more than guess tokens is your end goal. Research into AI is barely getting any funding, it's all going to buying Nvidia crapware to train LLMs.
>>108145489Yeah yeah, Anthropic plugin will change everything. Now go play with your nannybot
>>108145416More like you need a some multiplier of more compute and data for each percent of improvement.
>>108145509>LLMs just guess the next token in a series of tokensHalf of AI skepticism comes from some perceived sophistication of the human mind which in reality doesn't exist. Nearly as bad as artists who suddenly become quasi religious and talk about soul in art.
>>108145549what makes you think it will work that way when AI continues to show exponential growth with increased efficiency? You're being willfully ignorant of what's actually happening
>>108145536How will Claude Cowork not replace jobs? In the area of accounting, it seems pretty obvious that agents will be able to prepare tax returns and even prepare the supporting Excel files. Even with current agents, I’m sure they could reference prior files as examples and successfully prepare new ones.
>>108145589This technology is still in its infancy and we're already discussing not if it will replace jobs, but how many jobs it will replace.AI deniers can go suck dick, they are ignoring everything around them just to be contrarian little shitheads
>>108145551It's not about sophistication, the "AI" fundamentally doesn't try to understand what it's doing and why it's doing it. It's just guessing tokens. Yes, with a big enough model and extensive training data the guesses can be very accurate or even 100% accurate if you're dealing with a purpose built model for a very specific problem which lacks extensive nuance.But this isn't intelligence. It doesn't learn by itself, it doesn't care to understand what it's doing, it has no thoughts about what it's doing. When it generates an image, text or a piece of code it doesn't have any reason for doing it beside you writing the prompt. The method it does so is just aligning tokens based on how likely they are to fit together based on the training data supplied to it and the number of parameters available. In reality there are infinite possibilities for the order of tokens, but infinite parameters and infinite training data is not possible so the LLM will always fall short.You're the only one talking about sophistication of the human mind. I'm talking about the limitations of an LLM.
>>108145589>>108145078
>>108145589Accounting is more about a sense of security and so the tax authorities don't look as closely at your returns. Where I live I also legally have to have an accountant, but I know in many places that's not the case.Many jobs that will survive will probably be for emotional or legal reasons. People now laugh at programmers, but most other jobs are still easier than programming, but those other reasons may save them.
>>108144337Tomorrow... it could even be 5%!!!!!!!JUST TWO TRILLY MORE
>>108145614That’s one example and doesn’t reflect overall hiring trends. I’m sure most of those roles won’t last, either.
>>108145424Are you a test pilot? Why not?
>>108145608You can apply that reductive thinking to almost anything. People don't actually have understanding, it's just chemicals in the brain.There is no way to know if token prediction will plateau as an approach to AI or not, but so far these tools are useful.That problem solving ability is what I would call intelligence, it doesn't have to use the same mechanisms as the human brain. There are other approaches to AI and none work nearly as well as transformers.
>>108145632The "overall hiring trends" are companies layoffs after over overhiring in covid era and saying it was because AI, when, in reality, it's just fake attention-grabbing bait. But you already made your mind that some shit plugin will chaaange everything no point in arguing
>>108144118they REALLY don't want to pay goys wages
>>108145608What does it mean to understand something?
>>108145650I think reductively, the human thought process is next token prediction, where the "tokens" are electrical impulses that come together to form complex thoughts, which is basically how LLMs work.Now granted, there is some valid criticism to the fact that there's only so far an LLM can go granted that their medium is limited to language and a text-based mediumBut already we're seen so much multi-modal advancement. Models don't just think in terms of text-based tokens, they can visualize pictures pixel-by-pixel, they can interpret sound and motion.And now the next big advancement in AI isn't going to be just language-based models, but world models. Those are already well underway and deep in advancement, and the Google Genie model isn't just a novelty, it's a alpha-version peek at what a world model looks like.So yeah, there's some validity, but it's overall reductionist, and either way, it ignores the fact that we're already balls deep in technology well beyond the LLM
>>108145650For fucks sake, if you're completely tech illiterate go to some other place that caters more to your ilk. I am not gonna engage in a discussion with you if you're categorically unable to comprehend what is being said.Intelligence isn't problem solving ability either, intelligence can manifest in the ability to solve problems. One follows the other, not they are one and the same. A pick and place machine solves a problem with extreme efficiency, yet it has absolutely no intelligence. I don't care what you choose to call intelligence, you're just wrong.
>>108145589Accountants need certification, they'll probably use the AI and then sign off on it though. It's like why AI can't replace lawyers even though they use them
>>108145608>You're the only one talking about sophistication of the human mind. I'm talking about the limitations of an LLM.We're talking about both. They are in competition. It's sort of the reason for ai to exist after all. You talk about it like there is some insurmountable gap between human minds and AI models. Human minds have this deep creativity and reasoning for doing tasks. In reality we are token guessers doing what we are paid. I think it could be humbling to see how limited human intelligence is.
>>108145718I've probably been programming for at least a decade longer than you have. You probably have a spooky, Platonic view of intelligence so you will just never call anything intelligent unless it's a person.
>>108145726Right, but that’s at manager level. I’m basically saying that staff roles will be eliminated. You don’t need a certification to be a staff accountant.
>>108145187The Popemobile, whose concept began back in the 8th century, only became bulletproof after the attempted assassination of John Paul II in 1981.The pope survived two bullets + severe blood loss, forgave the assassin and got back to work. The assassin repented and, after serving his sentence, he visited the pope's grave in 2014.So I guess God is indeed real.
>>108145066The Wright brothers didn't take off in powered flight at first.
>>108145754Amen brother.Christ is KINGPraise the LORD!
>>108144118Meanwhile, at big blue
>>108145827IBM the renowned innovator of the 21st century
>>108145827News of AI probably hasn't reached Big Blue yet>trips over magnetic data reel, sends pile of punch cards flying
>>108144337Bruh 4% is not huge. It's at the level of using it or not doesn't make any difference. Source: I'm a software developer and I've been working for more than 15 years. The 4% sounds about right.
>>108145147You should get a mental check if you see religion everywhere.
>>108145827They're retarded. Everybody in the field knew this was going to happen. Some predicted earlier than others but with time everyone got to the same conclusion, AI doesn't improve much let alone replace developers. The only real difference is that instead of asking a search engine you ask the AI. Everything else is bullshit that doesn't save much time or makes the code better.
>>108145459Wow, you can't calculate that yourself?
>>108145054>yeah. back in 1953 or whatever, one megabyte cost 6200$. your 1 tb harddrive of cost 6.2 billion dollars. weve gone from "it cant even finish a sentence" to "it writes a novel" to "its bad at pictures" to "oh itll pump out classical paintings in 13 seconds" to "oh it also do full motion video sequences from a 1 sentence prompt" to-This.>hurrr-durrr it's only 4%First of all: source?Second: it would likely improve very quickly.
>>108145956He thinks 4% ai accuracy will result in 4% less jobs needed. Buddy clearly isnt the brightest
>>108145509>Yeah unfortunately for you this isn't how it works. LLMs just guess the next token in a series of tokens. There's a limit to how good you can guess something, and every wrong guess can instantly brick the work.An extremely dumb argument, perhaps relevant in 2023. You know that all modern LLMs can repeatedly edit a file with tool calls? If they make a mistake, they can fix it with the next tool call. Separate reasoning output also allows this.
>>108144921transformer based LLMs are almost 10 years old
>>108145509
Why the fuck is none focusing AI on languagesWe can finally achive fucking the Universal Translator but no we use for fucking shit
>>108146016B-but it still edits the file token by token!
>>108146026>Why the fuck is none focusing AI on languagesAre you living under a rock?
>>108146045Yes
>>108146016Yes. Differentiation between "scratchpad output" and "final output" is all you need to mitigate this issue.
>>108145509>LLMs just guess the next token based off a weighted dice roll influenced by past data it observedI mean, isnt this how most "people" work too?
>>108146022It's been like 20 fucking years and these ugly ass wojaks haven't disappeared
Ai is soulless. Most of the applications people are spamming its usage for is either godawful soulles drawings, something to own "DA LUDDITES", or replacement of a human, which never works. Nobody with a brain likes this crap, genuinely. Why do so many people blindly support it, you are contributing to cold calculated garbage with a splat of gold tinted yellow on it
I'm afraid. We've got a big globalist consulting firm doing a consult with us soon on ways to vision our future implementations of corporate paradigms or some bullshit and "AI integration" is high on their list. I can't wait to explain the complexities of actually getting real humans to do the thing on the needed timeline only to hear, "no no agents can totally do that" having not heard or even tried to understand what I'm communicating to them. I will read with great eagerness their vision for "efficiencies" created by putting half-baked technologies in charge of everything we've spent years making work. Good times.
>>108144478It's not even the easiest tasks that it automates. It actually automates some fairly tedious shit. Knowledge based jobs, basically. It can be trained to parse documents semantically. So document research, the more tedious parts of coding, etc. So basically entry level white collar shit. And if nobody can get in at the bottom there's eventually nobody qualified to replace all the senior programmers etc.
>>108145589Fucking retard lol
>>108145095Ask chatgpt to engineer a plane for you with 4% success rate and make you dreams come true.
I run a small company and went from 8 employees to 5Ideally I’d get it down to two
>>1081441183 years ago it was 0%, the middle part of the sigmoid is where most of the change happens, despite spending the least time in the middle part, and you should expect to enter the middle part after you begin rising. I expect the success rate for agents to more or less symmetrically skyrocket over the next 1-2 years. It should be a wild ride
>>108147219>bro just 14 trillion more
>>108147219I just did some exploratory math and it seems that it doesn't matter how steep the sigmoid middle is, the time it takes to go from 4% to 95% is apparently always around 4.3 times the time it takes to go from 1% to 4%. So we'll get to 95% success rate somewhere in the range of 4 to 9 years (depends on when the 1% point was). A lot longer than I expected, but we'll be at 50% in around 2 to 4 years
>>108144632>Granted, the visual side has definitely exponentially improved.Mostly because of more powerful hardware. Which costs a lot more.
>roll 20 sided dice>need to roll national 20 for success>do this every time you use AIHow do AI people cope with this?
>>108147385Using this model, looks like we'll have the following instantaneous rates of change for success rates: now: 8%/y, 1-2y: 28%/y, 2-4y: 40%/y, 3-6y: 27%/y, 4-8y: 4%/yAnd the following success rates: now: ~5%, 1-2y: 22%, 2-4y: 59%, 3-6: 88%, 4-8y: 97%I imagine the overton window will move the most (and thus produce the most calamity) when the rate of change is the highest, or slightly later, so around 2-6 years. Ouch, the US election if there is one in 2028 is gonna be fuckin crazy and lit af probably
>>108146022>>108146035>>108147262>(you)
