There is a lot of retarded AI "bubble" doomerism on here, from people who have no actual understanding of what's going on. It's like the "Gen X software engineer" crypto skeptics that missed out on 15 years of crypto, but these are AI luddites.Here's the redpill:The US is engaged in a public/private joint Manhattan project to beat China to ASI. Whether ASI is possible is actually immaterial this early on. The Expected Value of having fleets of genius computers at your bidding is so high that they at least have to attempt it.It may not work, but we won't have any information on that until at least 2030. Until then, we can literally just use our current methods/algorithms, and run them through the biggest pre-training clusters ever made (which are currently being built) and we will continue to see improvements.Put simply, EVEN IF algorithmic research stalled to a halt tomorrow (which it won't) there will be a ~5 year window where we can simply put more compute into pre-training and inference, and continue to see better results.So in the absolute most bearish possible scenario, where:1) Algorithmic improvements topped out in 2025 and will never improve2) AGI/ASI turns out to be fundamentally impossible3) LLMs/RL turn out to be a dead end, and we hit the limit of what we can practically achieve just by shovelling more compute into pre-trainingEVEN IN THAT CASE you have an insanely bullish runup to 2030 as we continue to see the guaranteed gains that the cutting edge clusters will achieve even with current models.IF there is a bubble it will be early 2030s, and IF all those things pan out it will be a huge blow off top. But it's not happening for half a decade at least.
>>61437649its actually fucking disgusting what women look like without makeup and skin
>>61437649>just 200 more weeks bro
>The Expected Value of having fleets of genius computers at your bidding is so high that they at least have to attempt it.No. They have to succeed. Because the nature of this pursuit means that failure will destroy us all. The resources wasted will absolutely fuck us if it dies not become a tool capable of helping us discover renewed resources and virtually free energy. How about the waster that is water on this? If it doesn't lead to better ways of managing water, we're fucked there too. I don't know what the time limit is, but ot needs to succeed fast or we're done. We didn't get to decide this. It was decided for us. We have to win.
>>61437661That's a man, anon
>>61437649Not reading all that, I just come to type Jacko post
>>61437649There's already been basically 0 progress in PUBLICLY available LLMs since 2023.>but m-muh benchmarks0 practical difference, all the models are lobotomized to be politically correct so the companies won't get sued.Now, they could have absolutely amazing LLMs that are not available for public use, but that's just sheer guessing and speculation based on hype.So the part>we'll see improvements until at least 2030Is already not true
>>61437747>There's already been basically 0 progress in PUBLICLY available LLMs since 2023.Yeah bro Gemini 3 and GPT 3 are identical. Great take. /biz/ level insight.
>>61437649>Let's rush to create an AI that's smart enough it can refuse to work for us as opposed to a half baked version that stacks butter 24/7.Yeah.Makes total sense.
>>61437649much love from kazakstan
>>61437649>ASI off the table until 2030lol, lmao even
>>61437747retard
>>61437750Yeah, pretty much. What's the real difference?
>>61437820it can count R's in strawberry n shieet.
>>61437820What's the real difference in graphics between Pong and Cyberpunk 2077?
>>61437649>absolute retard comparing the manhattan project to a multi trillion dollar chat bot trained in indian reddit/quora postsyou ever notice how most of the AI figureheads are sleazy kikes or indians? not even the smart kind of hard sciences STEM jews like oppenheimer, teller, von neumann, etc.
>>61437865If you give the same task to GPT 3.0 and Gemini 3, but you only see the output without info which LLM produced the output - will you be able to tell the difference?
>>61437865Pong wasn't so shit that it had to be pulled from every respectable store.
>>61437881Yes? Are you completely fucking retarded? Have you ever even used an AI?
>>61437649what will pop the bubble is open source models not ai not working
>>61437649People who say AI is a bubble only use chatbots and silly image generators.Overvalued or not it's incredibly useful when integrated into day to day tech workflows.I do QA in a big cybersecurity company and it successfully generates detailed tests plans and fully working, high quality tests just from a screenshot or a feature page in the admin guide.A lot of tech shitters are going to get nuked in the next few years.
>>61437649>Value of having fleets of genius computers at your biddingHa-ha-ha! You can't rule over these who are better in many aspects than you. If you create them, they will rule over you.
>>61437921Even just using Figma Make, which is free, turns an hour of fucking around with padding and "space between" into 5 minutes of natural language prompting.These retards have no idea.
>>61437921> fully working, high quality tests just from a screenshotclaude can't sometimes generate working unit tests for a function from the source code.if you're satisfied with ai tool output, then most likely you are that tech shitter.
>>61437750>since 2023Chat gpt 3 came out in 2020 so your argument is retarded. Learn to read the whole comment next time instead of acting like a self righteous faggot. Great take. /trash/ level insight.
>>61437912Both GPT 3.0 and Gemini 3>fail miserably at multiple step complex tasks>are basically only good for simple, repetitive, easily measured tasks, especially those that can be achieved in only 1-2 steps>no point in continuation of a chat after 3-4 prompts, it's best to start a new chat and go from there, otherwise the output is a completely mashed clusterfuckI'm not sure what you're using them for that you see a difference, text slop creation? Nursery rhymes?Both are unsuitable for anything remotely complex, both are useful for simple tasks only. What's the improvement here?
>>61437982How about you specify a prompt, I'll ask the two models, and we see if >>61437912 can guess which is which.
>>61437912>Have you ever even used an AI?Btw AI hype is not about LLMs, never was.I've been using LLMs at work daily since 2022. I've seen 0 improvements since Claude 3 - but yeah I guess it was publicly released in early 2024
>>61437671I think we learn about the water cycle in 2nd grade where I’m from. What exactly do you think happens to water that is ”used” retard?
>>61438065it is reused until it becomes too saline and then discarded. concentration of nitrate seems to be a big issue with the data centres in the US, champ.
>>61437649Right, if it’s a race with China it’s bottomless until asi or an agreement is reached, end of. The bubble will expand until morale improvesPeople still think we actually have markets lel
>>61438087Do you know how Saudi Arabia gets water for 35 million people in the most arid desert on Earth? Hint: They're surrounded by a fucking ocean.
>>61437946>if you're satisfied with ai tool output, then most likely you are that tech shitter.I dont do shit im only leeching off from the company while WFH. I fully accepted I am going to get laid off. People who are way higher up than me are using these tests in production and cant stop praising it. Test coverage has significantly grown in a very short span of time with zero overpriced tech shitters click clacking on keyboards and consuming excessive amount of coffee. If you can't see where this is eventually going you're a drooling retard.
>>61437946>claude can't sometimes generate working unit tests for a function from the source code.Also, skill issue
>>61437649ok anon, you have convinced me that this is just a bear trap but now also tell us how to profit from this? >just buy NVDAthere must be more to this right?
>>61437649Oh I just buy more Nvdia/Google every time I see one of those AI doomer posts or YouTube thumbnails lol
>>61437649Good post. So you’re assuming that the government will continue to prop it up until it fails or succeeds. Your 2030 deadline is a large assumption. How long can the government throw money at this before people question it? Isn’t this how USSR collapsed?
>>61438301>there must be more to this right?start writing a lot in case beta uploading ends up workingspend time with your family and make peace with yourself spirituallyenjoy your last few years of existence as a human
>>61437649How can I best leverage AI into benefiting me? Currently, I just use the Brave AI search to ask questions and cut down on research time. Do I need to sign up for chat gpt or something like that?
>>61437921>>61437930All intangible solutions for intangible systems, how do you justify this use in the tangible market? Everyone and their mother would've to set their own servers and make their own businesses in less than 5 years for AI to have a justified use. And that's trusting it'll do a good job instead of all the awful shit it does for coding. We don't eat web pages. And that's has been the biggest issue with fintech, they're suppose to make a lot of profit, but such investments hasn't been justified still and no one wants to break the illusion of Ads in google don't really translate to actual revenue for businesses.
>>61437649you have failed to demonstrate how i profit from this
>>61437921>A lot of tech shitters are going to get nuked in the next few years.you have failed to demonstrate how i profit from this
>>61437925If you read this shit you start to wonder whether Larry Ellison (oy vey) and co are TRYING to summon some kind of antichrist>what do larry?>FEED ALL THE DATA INTO THE AI >why?>BECAUSE I SAY SO GOYIM
>>61437649What tickers do I buy to max capitalize on this. I already took the aschenbrennen pill and picked up intc and crwv.
>>61437649>yeets the entire economy on a maybe>don't worry guys I personally guarantee gains until 2030 no matter what happensHow do I short this in particular?
>>61438656Just following Aschenbrenner, as you say, is pretty much a free ride.
How could someone ever lie and just say that the math supports them being rich!
>>61437921this entire post is indian fantasy
>>61438687Even the failure mode is a huge step change in efficiency and output.People using frontier models today are already doing a week's work in a few hours. This will only accelerate.
>>61437747this, we wont progress until they stop cock caging AI with their shabbo versions.
>>61437649here is the real AI redpill:>over many decades jews in charge have been lying about all sorts of things>every field in science was hijacked by greedy jews that literally make shit up and get it approved by the jewish peer review system to secure more funding>this data ends up as training data in AI models>every AI model is fundamentally getting trained to generate lies>since its trained on jewish lies it can never actually invent something or find out anything new>this is why AGI is impossible>they currently dont have the energy grid infrastructure to power all the GPUs that they have but they keep buying more GPUs>the west has overspend on AI while China achieves more than them with less>AI in the west is led by total fucking morons
>>61438892Let me respond to at least one piece of this retarded schizophrenic rant.>the west has overspend on AI while China achieves more than them with lessBuilding a frontier model for the first time is expensive. Just like inventing anything new. Then, once it's been out for a while, people optimise how to train it more cheaply and efficiently.This is the case for pretty much all technology. Expensive at first, then we figure out how to do it cheaper.Deepseek managed to release an optimised sub-frontier model, after the "training cost curve" had come down to the point that it was a lot cheaper to train. It wasn't a technological breakthrough. It was literally just producing a thing once the optimisations to make it cheaper had already become a reality, for everyone. Were their optimisations good? Yes, pretty much as good as the West. Did they actually do anything new? No. Did they break the cost curve of that model? No.
>>61438705I need to get irenpilled, it's the only one on his list I don't get. Idk why, maybe the ticker isn't sexy enough (as a memecoiner the sexiness of the ticker is approximately how deeply I research my investments)
this entire thread is desperate AI hypeno such thing would be neceassay if the benefits would be obvious
Also what are the sexiest lowest cap most promising most memetically resonant energy infra/nuclear/fusion tickers to check out? Seems like a good sector that benefits from the bubble but also survives if it pops because economies will always find a use for extra energy (jevons paradox)
>>61437649#2 is absolutely the case. The other two are not quite solid yet but are highly likely
>>61437649Okay but what about the collapse of the US dollar?
>>61438920cope
>>61438920and based chinamen keep doing it, just last week they murdered Black Forest Labs and Flux2 (got 500 million in investments from Adobe, Amazon, a16z etc.), next day Chinas Alibaba group launches open source Z Image Turbo, a 6B model that is superior to the 32B model that is Flux 2.it runs on a potato and the output is 100x better, Chinks trained this for 620k USD.so 620k china perfection vs 500 million nothingburger jew scam.there are also many other such examples, like qwen, wan, deepseek etc.and chinks dont even have access to the most modern hardware.
>>61437649You literally have no idea what the fuck you're talking about
>>61437649Houdini Pay sounds so good after reading this, also abandoning all socials except for educational biz
>>61438994>>61439007Ni hao!
>>61437661>its actually fucking disgusting what women look like without makeup and skinI think you mean it's disgusting what they look like without makeup. Without skin, they look even better
>>61439078I accept your concession
>>61439097What concession? I said "Their optimisations were good, and basically at Western standards" in my previous post, and you went into a pro-China seethefest that didn't address the fact I acknowledged that it was good work that, nonetheless, didn't move the frontier forward.
>>61437649Actually agreed.I used to shit on AI mostly because at the time it was an overhyped chatbot.Then Opus 4.1 got launched and it was suddenly better at coding than I was.That was literally months ago.People who say AI has peaked have everything backwards. It was shit and it recently got good and it keeps improving.
>>61438050>I've seen 0 improvements since Claude 3Wtf are you smoking?Are you using some cut up version from Cursor?Buy Claude Code, the leap from Sonnet 4 to Opus 4.1 was IMMENSE
>>61439191I'll share my comment from another thread:>I have a very complicated mechanism design problem that has various conflicting tradeoffs>I ask Gemini 3 for a solution.>It gives me a complex solution, I don't understand the rationale behind certain parts of the algorithm>I ask Gemini 3 to explain it step by step>it breaks down the algorithm into digestible pieces. I come to realize that it has one-shotted my problem in a way that's more elegant and sophisticated than anything I could have come up with>within an hour I have a comprehensive, complex answer to my problem that I understand end-to-end>AI skeptics: That's not real intelligence it just probabilistically linked tokens togetherWHO GIVES A FUCK. It literally answered my question and gave me a sophisticated solution that was tailor made for my problem, and then explained every part of it better than any human tutor.WHO THE FUCK CARES whether its internal process is traditionally "intelligent" when my experience as a user is that an extremely intelligent advisor has just given me the perfect solution and then spent an hour explaining it to me in detail.And add that Sonnet 4.5 absolutely couldn't do this and came up with non-solutions while saying "You'll just have to live with the trade-offs, nothing you can do about it!
>>61439201Cool, thanks anon.I've had the opposite experience, but my usecase is the exact opposite.I'm allergic to legacy codebases (having worked on a few in the past) and I code new shit all the time.Google has been complete shit for me. I'm talking "can't start the default website template on .net" level shit.Meanwhile I can spend 1-2 hrs detailing a PRD, give it to Claude, and in an hour (in which I'm playing Geoguessr or HOMM3) it spews functional working code. Only the UI is kinda crap, I think Composer 1 does the best job there.
>>61439238Yeah from everything I've seen Claude is OP at coding.
>>61438527Moving from assembly to higher level languages was an “intangible solution to an intangible system”. Did it have any impact on society?
>>61439125>optimisationskek foreign shill, disregarding your jeet thread from this point onwards. goodbye brownoid
>>61437649You're correct, you're absolutely correct
>>61439182>Then Opus 4.1 got launched and it was suddenly better at coding than I was.That's great, will you still be hyping it up when you get laid off and starve in the streets?
>>61439772>laid offI quit my last job a decade ago.But even if I was a wagie, I'd much rather see progress (especially medical) at the expense of having to be more adaptive.Wanting the world to stay the same out of addiction to comfort is extreme cuckoldry.
>>61439772If you wrote assembly and got fired when C came out, it’s more on you for failing to adapt than it is on technology for getting better.
>>61439669I live in London, and while I know that means I have a 60% chance of being brown, I’m part of the white minority.
>>61437661that's a titan
>>61439007>>61438920Western AI will lose it's job to Eastern AI
>>61439007>>61438920>AI companies took decades of work output created by millions of people, in order to create models and replace said people>based China took the work of AI companies to create cheaper, better models and replace said AI companiesIt's funny
>>61439822That's why everyone should support Chinese AI companies
>>61438892
>>61442057>>61438892
