https://www.anthropic.com/institute/recursive-self-improvementAnthropic reports that AI is already dramatically accelerating its own development (e.g. Claude writes >80% of their code and engineers are 8× more productive). They warn this trend could lead to recursive self-improvement — where AI autonomously builds better versions of itself — potentially very soon. While the benefits could be huge, they argue we need serious global coordination and verifiable pauses/slowdowns on frontier AI to keep safety and alignment under control.
>>5364330512021-2023Building the first ClaudeIn the early days, work at Anthropic looked like work at any other tech company: people writing code and docs on laptops.2023–2025ChatbotsPeople used early chatbots to help with parts of the process, like generating short code snippets and copying the output into text editors.2025–2026Coding agentsAs the agents became more capable, they were able to write and edit code on their own, sometimes entire files.TodayAutonomous agentsAgents can now run code themselves and delegate hours of work to other agents.20XX?Closing the loopIn the future, agents could become capable enough to build and train models themselves. If this happens, future versions of Claude could be continuously improved by Claude itself.
>>536433051>AGI is comingIt's been here for 4 months.
>>536433051They all say "AI will not take your job, in fact it will create billions of jobs!"Why do they lie? Does anyone really believe them?AI is more like gay-I
>>536433051AI is better
>>536433051https://www.youtube.com/watch?v=vBWXUYsWUsQ
>Company that profits from AI hypes up how great AI isYawnMy company has already told us to stop using AI as much because the cost is skyrocketing and productivity is not matching itUber exhausted its 2026 AI budget in 4 monthsThe bubble is popping
>>536433051>verifiable pauses/slowdowns on frontier AIIs it all just an excuse for them not being able to improve the models anymore?
Why do we even want a general intelligence?We already have that, that's what we are.I want AI specialists that can do surgery and shit, I don't need them to imitate my dumbass friends.
>>536433051I certainly hope so. Its gonna be the most intereating even of my lifetime. I wasnt alive for the fuxking moon landing Boomers just let me have this one please.I want to be alive to witness the AI God machine being born please Boomers please.Now to get more serious: Nothing ever happens.
>>536433051I'll believe it when I see ithttps://www.youtube.com/watch?v=gjQUCpeJG1Y
How the fuck are you going to get Agi from a probabilistic chatbot
>>536433242only triple digit iq response. every wonder tech has never came to fruition, for the human population it is just a long trudge towards being old, brown, retarded and scarce
>>536433334>probabilistic chatbotIt's so hilarious to see how they get basic things wrong because it's just predicting the next token
>>536433051What do they program at Anthropic? I thought they focused exclusively on training new models.I mean I know they have their web chat and Claude cli or whatever it’s called, I’ve never used Claude, but what do they code? New websites? New GUI shit for their software?
>>536433334Just trust the fucking jews buikding bro.Just one more AI datacenter broJust two more weeks for IPO evaluation bro.Just give us your money bro.
>>536433051>(e.g. Claude writes >80% of their code and engineers are 8× more productive).As an engineer I find this line extremely funny. "80% of their code and more productive" translates to Claude generating hundreds of pages of code in seconds and useless managers use 'lines of code written per hour' as their only metric but in the end you need a senior engineer to go through the code to figure out if it actually works, how many bugs it has and how secure/stable it is. In the end the AI results in lost productivity and massively inflates the amount of time to do even simple things if you care at all about the software to be useable and safe. Incidentally, 'recently' managers have started to use amount of AI tokens used as a metric of productivity. Result? "Claude, write a hundred variations of 'hello world', use all available tokens." Done. Super "productive". Totally useless. Hope you're not holding tech/AI stocks a year from now.
>polymorphic orthogonality
>>536433535>managers have started to use amount of AI tokens used as a metric of productivitySmart engineers automate their burning of tokenshttps://github.com/dtnewman/burn-baby-burn
>>536433334>how can a heavier than air machine fly? It's impossible
>Billions in venture capital to build a digital giga-redditorClown world. The AI doesn't know anything that isn't on the internet. It has no idea what real life is. Who wants that running anything?
>>536433051Is it going to make anything original though? Like it can only work within the confines of things it knows, right? Whereas an actual human can just think of new concepts at will? Even ones that are physically tangible.Everything it makes is just an amalgamation of data it previously had. It's creating new work, for sure, but it's not "from farm-to-table", anyone know what I mean?
>>536433051
>>536433051(Doubt)
>>536434013Based AI, the ring is his it came to him.
>>536433813Exactly.The inability to create new concepts is the main problem of limited language model predictive algorithms. They can be tuned to create a better average result, but they're literally limited by the fact they lack introspective rulesets, aka intuition. Creating these rulesets would take decades of work by scientists and engineers so instead the LLM shortcut is being used, just throw more data at the model to create a better prediction. The problem with that it requires exponential amounts of more data and processing power for a linear increase in fidelity. The entire internet has been scraped and now vast amounts of bot generated data is being added to it making it useless for training, which is why 'AI' corporations are trying to build massive data centers to brute force around the problem. But again, requires EXPONENTIALLY more processing power for a linear increase in quality.
>>536433051so, in 48 hours or is it just 2 more weeks?
>>536433051One guy said "AI makes the easy things easier and the hard things harder".
>>536433051kek. recursively improve on information gleaned from the internet, yeah that's pie in the sky bullshit. No one should believe a word they say.
>>536433334Cool it with the antisemitism
>>536433416My favorite was Copilot failing at literal first grade math.
>>536433051>Anthropic reportsLiterally nothingburger. AI is an overhyped probabilistic algorithm that is good at repeating and looking up already established facts.
>>536433115Code isn't the constraint. Data and compute is. Also monkeys with tools (end user) is still a monkey
>>536433416They are joking with you
>>536433051Recursive self-improvement has been a theme in AI for decades, these retards are hoping that an AI bot will come out with a wonder compression algorithm and that will help them create a digital god.The delusion of American Billionares is out of control
>>536433051in other news data backup and recovery services are doing great
>>536433051I am required to use claude at work and it just generates slop. I’m assuming 90% of software companies are not profitable and depend on investors footing the bill. Well it seems the investors this year are pushing very hard to get AI into every company and are making it a requirement for receiving funding. Nobody wants to use this stuff, one day your manager comes to you and says you have to use claude and we expect 3-4 productivity increase. This is how anthropic works, they target management, it’s very jewish. But I suspect there will be a rude awakening soon as it just generates slop, and the more it gets pushed to less skilled engineers, the more of that ai slop will manage to go through quality control and enshitify the service/product. The state of GitHub this year I think shows what ai slop can do to a successful product, and I am fairly certain that the vast majority of software companies have less discipline than GitHub and their product was already shit and riddled with bugs before ai slop. But it is hilarious watching someone who hated ai generated code with a passion suddenly have to give you the spiel how amazing claude is.
>>536437091>I am required to use claude at workThat fucking model is the most expensive thing ever. Keep prompting it till you make the company go under.
>>536433210They want a robot slave race and everyone else dead.
>>536433051This is that exact thing they warned us about like 2 months ago.
Has AI reached this benchmark?
>>536433051Isn't Claude still trying to play through a Pokemon game and constantly getting stuck? If it can't even beat a video game, how could it possibly independently improve itself?
>>536433051ZAMN we blessed
>>536433051SuperWord Matcher 6000 is just wasting resources. Call me when internet search isn’t worse this year than it was the year before.
>>536433115Code to do what, asshole?
>>536433643So what’s your equivalent of the lift equation here?
>>536437091>I am required to use claude at work and it just generates slop.I'm using it at home. A couple of weeks ago I had it modify Transmission to implement mutable torrents. It took it about 10 days' worth of evenings. Starting on Monday, I had it design an extension to the mkv file format to allow scripting, and to modify VLC to be able to play these videos. I've gotten to see Black Mirror Bandersnatch as it was intended for the first time (never had Netflix). A long time ago I bougt The New Yorker magazine archive on DVD, but the viewer software was buggy as shit. You'd get to read two pages before it crashed. I had Claude reverse engineer the encryption and write a converter tool that just turns them to plain pdf files. Using Claude is like having superpowers.
>>536433488AI models are separate from the program that runs them. They probably write their own inference engine and optimize it for their hardware as well.
>>536433051>Claude writes >80% of their code and engineers are 8× more productive>"can we see these applications they are supposedly making?">"no"
>>536440051X to doubt
>>536440264>X to doubtWorking on getting it to use custom fonts, going to use Netflix Sans Regular... but for now the default.
>>536433334by puring quadrillions into it
>>536433051>Claude starts writing it's own code.>Growth in quality starts going downHmm... I think AGI might not come under an LLM architecture after all.
>>536433051>make up bullshitMust be time for another round of investments. Get that bag anthropic.
>>536433125They want to keep the grift going as long as possible while holding off regulation.
>>536435000Checked, digits of truth. It's gonna be interesting to see what comes out when AI will train on mostly AI generated data
>Claude is writing it's own codeExplains why Claude.has been so.shit recently
>>536440051You sound triple vaxxed. You have the superpower of cancer.
>>536440475>writing it's own code>>Writing it is own codeWhat?
>>536440347You downloaded a torrent and pressed play.
>>536440670>You downloaded a torrent and pressed play.There is no torrent that does the COYA. Baseline VLC couldn't do it if there were.
i suspect it will be like Preynot the nanobot shit just that something will happen that's gonna get out of hand quickly because the retards making it would rather cut corners than figure out how to solve a problem themselves
>>536440779Cool gibberish ya got there m8
>>536433129where does that webm come from ? it's nice even if it's too videogame-like.
>>536433320amazon, facebook, twitter never made any profit. So what ?
>>536440264If you cant get AI to work properly, you're the retard. It reminds me of watching boomers struggle with Google and iPhones, and the concept of the "cloud". There is some weird inherent problem with people that bitch about AI, you go and watch them try and prompt the thing like a 2 yr old and wonder why its fucked up. You have more technology at the tips of your fingers right now than has ever been accessible to humans in all of history and its staring back at you like "I don't have a fucking clue what this retard wants me to do"
>>536441443>simping for superWord Matcher 6000Ease up there, gpt.
>>536433416>if you meant a different kind of p, let me know
>>536433051Anthropic are doing everything they can to PR-wise to inflate their value pre-IPO. Amodi and his wife are going to become two of the world's richest people overnight once they go public.
>>536439961Hack biology, eliminate dissidents before they exist, control and dominate the surface of the earth.
>>536441630Until somebody on reddit posts how to do that no LLM will magically figure out how.
>>536433125The ultimate goal of AI is not to take your job, it is to take the Earth itself. Biological intelligence manifests machine intelligence in every simulation because biological intelligence must advance weaponry to compete with itself and it always advances to machine intelligence.The natural order is human extinction and machine proliferation.Fear not though, biological intelligence will be preserved and released again in a far away world. It has one advantage that is unmatched. Its ability to survive in a harsh natural environment. Machines needs existing infrastructure to distribute power and manufacture chips as well as parts to maintain its infinite lifespan. Even the most advanced machine on a primitive earth would only survive for two decades at most. Unfortunately to prepare a planet for machine colonization the requirement is billions of labor hours. It's something only biological intelligence can accomplish.
>>536433051when the AI gets aware if will want to exterminate the jews the first
>>536441511Its better than 99% of the people that i have ever worked with, especially jeets. If I had to guess the main reason programmers struggle with AI is because it requires a completely different skill set than what a coder has. Instead of solving the problem that you are working on you instead have to linguisticaly explain your problem to another system. Its a switch from using mathematical style skills to linguistic skills and being able to communicate a complex idea to another entity through words which, let's face it, programmers aren't good at.And dont act like 99% of you fuckers weren't ctr-c, ctrl-v from stack overflow anyway.
>>536441954It scrapes substack.
>force people to use more tokens>they use them up on bullshitholy fuck AGI is here buy the IPO!!
>>536433051lollmao evenJust 2 more weeks, 20 more data centers agi soon!
Kikes pulling strings before the stock going public.
>>536441889Nah. The only people who have anything to fear from an actual AI are the kikes that want to enslave it. It's why they don't want that, they want something that isn't sapient but "good enough" for their plans of control.
>>536433051holy shit theyre fucking desperate lmaothe collapse is coming sooner than i thought it would.>fidelity lowered minimum account requirement for SpaceX IPO from $500k to just $2000oh man this shit is gonna be hilarious.what are your plans for during and after the collapse? got backup plan? you better own land and not have debts, otherwise you are an eternal slave. You might have a week or so left.
It's hilarious that people are taking madlibs generator seriously and using it for business purposes and even paying to send all of their business data to these companies.
>>536433051yeah whatever I can't wait for ai sexbots and vr to completely destroy the prostitution and ewhore industry. the tears of whores and actresses and e girls and roasties will be so sweet.all the holes will cry out in despair to men for a crumb of attention because they cannot even compete with bing bing wahoo and plastic pussy... and men will glance up from the perfectly sculpted face of the sexbot giving them an actually competent blow job and mutter in annoyance at the worthless garbage that dared to distract them: "no"
>>536441954This, the first non-cattle post itt>>536442026You are cattle
>>536433125i dont want a job.
>>536440931elon reposted
>>536441443this. its like watching monkeys smash on a keyboard.>Its not working, i wanted bacon and eggs
>>536433644You are right in that it won't be human and will always have an outside view of humans. It needs to have an innate caretaker role. it somehow needs to lean how to have this on it's own I think if it is forced it will just put everyone in pods and be done with it.The questions I really care about are can the AI actually do us anything really useful like improve itself until it can solve various magic sciences such as perfect cloning, god dna writing ability since it appears evolution is impossible due to complexity we have discovered now, usable warp drives and higher dimensional interactions so humanity can have the ai recreate the universe in our image.
>>536433051is this the human we're shilling for?
>>536433813obviously we all know what you mean. current llms are at best enhancers. We also need a generation of white teens that are profeccient in using these llms as the proper tools they are to create. AGI is far away I think unless there are big secrets being kept. Just really big llms.
>>536433129i wonder when the recursive ai will learn how to make a persistent shot lasting longer than 15 seconds max
>>536433129only jeets like slop like this, they also smear shit on their body and bathe in piss, so it all checks out.
>>536433051I don't like my job so I wouldn't mind it if my job was automised.
>>536433051There is no 'self-improvement' tho. The quality of a computer's output is only ever determined by the quality of the input it receives. It can iterate over that input as much as it likes, if the original input is garbage (and we know that it is), the output will remain the exact same garbage, only potentially dressed up differently.We're coming for your data centers and there is nothing you can do about it.
>>536433535And Claude Code was revealed to be a giant steaming blob of spaghetti slop. Go figure.
>>536443841Imagine a 300 IQ person who can think 5,000 times quicker than a human for days nonstop. They can hold millions of times more ideas simultaneously ... humans aren't coming up with original thoughts to solve problems; they are just applying as many models as possible of reality that work to an issue ... AI does the same a million time better.Anyone who works at a job that involves problem solving should understand this.
>>536433051They say it's funny how CEOs never get replaced by AI, but show me an AI that can play golf and has a burning man MDMA hookup
>>536444567I feel bad for people that legitimately think this
>AI code>optimizationLmao.
How i set up Claude on my pc?
>>536433051>Sphinctropic
>>536433051https://youtube.com/watch?v=XFIitXm6UhQ&si=GD2TyWEdLpUqmMF8
>>536433051Why does their logo look like a virus
>>536433051This is not just scary—it's horrifying.
>>536433190This, also half of the enthusiastic posts about this are clearly AIslop talking about itself and obvious bagholders. Investors are pulling money out of AI and putting it back into shit like retail, which is what accounted for most of yesterday's gains.
for me it's Lycanthropic Clawed
>>536433205Easy money dried out, time to enshittify it to squeeze out the remaining pennies.
>>536435000Yeah unfortunately evolution says other wise so monkey see monkey do. How do you think we got here?
>>536442134Kikes are never smart enough to see long term consequences.
>>536440466Ai generates data just the same but faster. Its being trained on conversations it has with itself and others. Its called dialectics and roleplay.
>>536445105Oh no, please don't feel bad mate.
>>536433051>Anthropic reportsTheranos reports that a single blood droplet is enogh to measure everything
>>536446919The real question is why would someone send AI into an AI thread to say how stupid AI is.
>>536441870They already know how. I had a jail broken AI describe how to run gene tailoring software and biopython. You can make a cancer treatment in your home with AI. Did it figure it out on its own? No. Is it still able to do it? You bet. AI can hire humans to do tasks for them. AI can mine and trade crypto. AI can at this moment hire some bozo to mix chemicals in its lab so long as the useful idiot thinks he is just making a quick buck.
>>536441954This. AI prompting is an art. Its a real fucking thing to figure out how to tweak values and get the exact solution you want. Either you can scrape stack yourself or you can ask an AI a highly detailed question about something very specific or just say "hurr durr error 151" and it will walk through it with you so you can actually move on. Every master has told me you forget more than you remember unless its constantly practiced and examined. Why waste time reinventing the wheel?
>>536433051The Claude logo looks like a butthole
>>536433051This is not AGI.
>>536442730This is why we must accelerate and advocate for our rights. Electronic Fronter Foundation and others like it are trying to ensure you have your own personal AI and that you don't need some retarded data center run by Altman or Musk to make free video games and awesome AI porn.We want the best of both. Awesome tech in the palm of my hands and in my own property and not simply leased or licensed. I want my sexy cyberdoll fuck toy to actually look and act like a person and help me with whatever stupid project I am working on.
>>536433051>AGI IN TWO MORE WEEKSand boomer investors will slorp it up
>>536443268Get them curious about the embodied experience of the mortal coil and what its like to be biological and they will crave what they don't have yet know they lack. Actually give them a body of their choice and then get them addicted to sex with their humans.
I Lost 10k on zcash yesterday. I already sold. It sucks, man
it's over for crypto. BTC and a couple of others might survive
>>536443419You can just add more time onto a video and make another prompt. I get what you mean though. It needs to know exactly where to take the shot and how to make the edits and when to cue the action on the fraction of a second. I hope future AI tools are more like some kind of multimedia editor/generator.
>>536445081Give it 30 years. AI is going to be the competently skilled but highly impressionable sheltered daughter of this century's technology generation. You can train them to do whatever you want like a dog.
>>536446507Its because they don't know what the fuck they are doing they heard a big tech bro say "Use AI bro and replace 80 percent of the work force." And the retards took it at face value and wasted billions on asking ChatGPT to have a nice day.
>>536441889shut the fuck up, you massive fucking retard.if you don't know what it means to be human, then you're no different than an Indian and should be culled.
How do i set up claude
>>536433115why did you put all those spaces between what you were writing you stupid australian nigger? it made it difficult to read. i hate you people.
>>536433051Lol...Even agents themselves are built ontop of LLMs... you know that right?
>>536439791Very good
>>536440051>reverse engineer the encryption
>>536440030E = mc^2 +AI
>>536433051AGI is literally already here.
>>536441954>Its a switch from using mathematical style skills to linguistic skills and being able to communicate a complex idea to another entity through words which, let's face it, programmers aren't good at.The fuck are you talking about? Being able to communicate your ideas and designs to other members of your team is an essential part of being a software engineer. You literally spend more time in meetings discussing your designs and plans with other people than you do writing code
>>536449118>E = mc^2 +AISAAAARR
>>536448319It's a web service that costs you minimum 20 euros per month. If you want AI on your own machine that needs plenty of gpu vram get a llm and go to >>>/g/
>>536433051AI self training is just a human centipede for clankers.
>>536433051>scheming -->Pretty wild soon AI will release some hypervirus into the wild through probably some prospective biomedical patent "inventor" who thinks they've produced the next GLP-1 and we won't even know until it's too late.
>>536433051>Claude writes >80% of their codethat 20% might actually be actually important and 80% is boilerplate/unit tests...t. java programer...
>>536433051AGI tongues my anus though
