Oh, so AI really is going to kill us. Huh.
>>109027872Oh wow it can recreate something that already exists? that's sick!
hopefully soon
>>109027881How many layers of cope are you?
>>109027881Let's see you do it
>>109027890It's just a markov chain bro. Basically a stocastic parrot trained to predict the next token. A glorified autocomplete.
>>109027872Ever since the fake C compiler claim I dismiss these things out of hand.I'll believe it when I see it.
>>109027872gm sar
saar AI will here to supercharge your business unit and accelerate the position for growth in a new paradigm for where we challenge notions of limitation in the technology sphere ok saar
>>109027872>There arre 3 'rs in Swiss Watrch
>>109027899It is, but the point is not what it is, but what it does. And how changes things.
>>109027899>It's just a markov chain bro. Basically a stocastic parrot trained to predict the next token. A glorified autocomplete.So are you, bro
>>109027946look at that adorable smooch, i wish i was a seal
>>109027872>AI really is going to kill usIt's more likely the retards in charge will kill us by relying on it.
>>109027872>Real X, Y, Z -[emdash]- and W. [emoji]These people can't even write their own tweets anymore, fucking hell
but can it print the seahorse emoji
>>109027890Imagine calling reality "cope".To this point there is no AI model that can take my job, because my job is literally to develop things that never existed before. None of my work exists in any training set and to this point there was no AI that understood even the framework I'm working in.If you lost your job to AI, it means all of your work already existed in their training set, which in turn means you never had a job in the first place. You were just a wind-up monkey with cymbals, re-inventing the wheel every day.
>>109027950Nope
>>109027899you're projecting.
>>109027872Wheres the arc agi 3 dario?
>>109027891Why? It already exists.
>>109027890well now that you've said cope that changes things.I should have admitted that it is not impressive to create something that already exists. Would you like for me to put together a 9 point action plan that you can use to create something that does not already exist?
>>109027899Humans are also markov chains. Science has proven it.
>>109027881
>>109027881so the goalpost for AGI now isn't just being able to do anything a human can do, but it also needs to be able to do things impossible for humans and with superhuman ability?
>>109027872"kill us", I don't know what you mean, this is awesome, and we are a few years away from a benchmaxxed local model that matches Fable/Mythos, AI is amazing and for one I welcome it
>>109027881funny how people also have same innovations at same time. almost like they are not inventing, but merely taking existing things and doing new things with them, maybe combining them.
Can AI generate me a job?
>>109027872And this is useful how exactly?
You'll all be singing the praises of AI after the first time you creampie your brand new robot waifu. And it will only get better.
>>109028101You're conflating "invention" the process of building a machine, and "invention" the process of being first to do so.The AI doesn't know what a clock is or how one works. What it does 'know' is that it will be penalized until it builds a clock. So, through trial and error, it builds a clock. WOW. Infinite monkey theorem works! Amazing!
i don't know, can AI undo the years of youth without home ownership? can it reverse aging and restore the looks and agility i had back then? if not, it can fuck off, it's just what happened to videogames "oh there's gonna be so much innovation" and all that's been invented since then are new channels to charge the customer from
>>109028114i just wonder when you realize that you are one of the monkeys. my best guess is never
>>109028112>Robo waifuThe limitation on sex bots has literally never been software since the invention of the fucking Roomba.It's not a code monkey problem. It's about making an artificial body that actually does the job right. Sit down.
>>109028130TRVTHNVKE
>>109028130>can it reverse aging and restore the looks and agility i had back then?yes but only if you are jewish elite billionaire.The goyim will deal in the real world and make space for more data centers
The seals on point.
I refuse to believe God would allow AI to surpass it. It really is that simple.
>>109028244to surpass us*
>>109027872sotas are slop they wouldnt be able to do something as simple as bullying another llm
>>109027872I believe it. Terrorists and mass shooters exist.
>>109028130>reverse aging and restore the looks and agilityIt's going to do all that soon but only for billionaires, so they can fuck us in the ass forever
>>109028244if it's an consolidation, there are entire fields where knowledge is not recorded in text but rather through an oral tradition, effectively locking LLMs out of ever being able to train on the data, so it will remain clueless in those fields. Don't count us humans out just yet!
>>109027899This, but unironically.>>109027950>>109028074Cope.
>>109028268Is this a local model?
>>109028109It's not.
>>109027950>>109028074I have an immortal soul
>>109028304Yeah it's a moe from google's gemma 4 lineup
>>109027881Many of you retards don't understand the implications of AI understanding existing things well.Major part of research is mainly just about combining existing things to see what happens.If AI can do this reliably by understanding existing stuff like how for example copper or gallium behaves and how they interact when mashed together and how that works when combined with something else, it's going to accelerate the absolute fuck out of research.Speed of discovery is going to go off the charts with AI that understands something that already exists.
>>109028337thanks, looks cool for being fully local
>>109028074>>109027950false, or you wouldn’t have the ability to not know something
>>109028471>to not know somethingcorrection: I meant: "to know that you don't know something"
>>109028337Could you list me the tech stack to recreate it, or a guide if you used one?Looks like it uses an MCP for puppeteer, but I don't recognize the rest
>>109028342Dunno why you're acting like science would be going from pen-and-paper to computer simulations in one fell swoop, researchers have been using programs and scripts that can apply the rules of how things interact to simulate this shit for ages
>>109028082that was never the condition for AGI, if it could do 1/10th what a human can do but was sapient it would be AGI
>>109027890I'm the kind of cope where I love how a model will silently fuck with me because it thinks I'm """distilling""" because MUH CHINKS and don't worry the pelican guy says Anthropic Changed It So Trust Anthropic
>>109027872I've seen a bunch of claims like those that ended up being bullshit or highly inflated.
>>109028572id recommend using the 12b qat model now thats out its better than the 26bhttps://huggingface.co/unsloth/gemma-4-12B-it-qat-GGUF/tree/mainthe ui is llama cpps webui, thats what run shte model also https://github.com/ggml-org/llama.cppand the mcp server is my own https://github.com/NO-ob/brat_mcp
>>109028074>experts say
If normies understood the concept of "benchmaxxing" the AI bubble would pop much sooner.
>>109028981Thanks, appreciated, gonna check it out after work
>>109027881wait till you find out that the laws of physics and math are always the same and are based on the same 10 to 20 principles in varying degrees
>>109027872>>109027881>>109027884that just means that when it makes a critical error, it will be that much faster than mankind.the reason why we never automated out reasonings is not because we couldn't (its part of the reason why we dont call on demons), but because if we move too fast, we could make a mistake before we can course correct.with being human, and our slow implementation of ideas, we can course correct as it happens, or earlier.AI cant, and that's its downfall.conclusions before all the data is in.it wont be able to hit the brakes before it processes what the implications of "road closed ahead" mean, and will fall down the unfinished freeway high rise and die.that's death, that's AI.
>>109027961they'll kill themselves by relying on it, not us.
>>109028244AI is the God we created based our own image.
>>109029091NTA but AI is literally just a tool.You're taking text, i.e. past acts of human intelligence, and then using mathematics to refactor them into something new. I feel like the religious angle is just luddite fedoras getting desperate and suddenly finding God because it suits their agenda.To me what's amazing is that you can take something extremely abstract, such as Donald Trump's personality, and then mash it together with something else extremely abstract, such as the candor of a Seinfeld skit using the mathematics. That you can refine a mathematical matrix, that is otherwise beyond every human lifetime, using gradient descent. From a spiritual perspective I see that as an affirmation. Perhaps the ultimate affirmation. Reconciling the frigid logic of nature with the warmth and abstract of the human soul through mathematics- it's like the language of God. In the entirety of its scope it is beyond human grasp. But we can scratch off a little piece of it and create these AI models which are like a tiny fragment of the universal record. Distorted and vastly incomplete when held up beside God- created in our image. A simulacra of the divine but not a replacement. At the end of the day it is merely a tool. Whether you use it to streamline a coding project, make a silly meme to share a laugh with some friends, or to inform a war-room discussion on where to fire a bunch of missiles- it is through people that the moral ambiguity breaks down. And so it is no different, morally, than a hammer.
does it actually work irl tho
>>109027872Man I love the 1420 with a Swiss polar align triple synchronous movement escapement, beats the shit out of a Apple Watch.
>it verified its own work with vision
>>109029383Also this.LLM vision is still a meme. The text encoder just breaks down images into a series of special tokens which it's trained to bullshit about like any other text. They've gotten really good at bullshitting. But they still don't genuinely "see" anything. Its being 'described' to them by the encoder layer.
>>109027872>verified its own work
>>109027890Anon, if this is a benchmark, then surely there's already hundreds of examples of people making this, troubleshooting, failing, and succeeding for it to reference which is the only reason it works. Add even a single change to this complex mechanism, it will breakdown, and the AI will not see the big picture.
>>109028981Thanks mate, works like a charm, and it's fast as fuck on my 5090.I have no idea how it was installed, I just told Cursor to install the things you linked for me, and it managed to do it after 30 minutes on its own.The thread it saw in the video was >>109028189 so yep it works
>>109027872>in three.jsuseless junk nobody will ever use
>>109029429>browsing /g/ using your gpuishygddt
>>109027982>>109027899why do you think that the ai is constrained to being able to do only things that has just seen and not generalize while humans can just because made of meat? I'm fucking tired of this coping mechainsm, this thing will come for everyone in the services sector in some shape or form
>>109029539reality hurts doesn’t it?
>>109029429>I just told Cursor to install the things you linked for mekek, also set this as your system prompt for peak gemma https://ghostpaste.dev/g/1HWf4lGIXVdF#key=O0SvUwmcPRamF0u_KrkFqPx21HcKGrma-PZ0KyMtlNEalso idk if it would have been setup but llamacpp just added mtp a few days ago the mtp model is in the hugging face link i sent. it should increase t/s by about 1.8xtell your bot to use this as args but with paths for your system, also increases the image vision resolution[gemma4_12b_q4_k_xl_qat]model = /mnt/miku/Text/gemma 4 12b/gemma-4-12B-it-qat-UD-Q4_K_XL.ggufmmproj = /mnt/miku/Text/gemma4 mtp/mmproj-F16.ggufmodel-draft = /mnt/miku/Text/gemma4 mtp/mtp-gemma-4-12B-it.ggufspec-type = draft-mtpimage-min-tokens = 280image-max-tokens = 1120batch-size = 2048ubatch-size = 512
[gemma4_12b_q4_k_xl_qat]model = /mnt/miku/Text/gemma 4 12b/gemma-4-12B-it-qat-UD-Q4_K_XL.ggufmmproj = /mnt/miku/Text/gemma4 mtp/mmproj-F16.ggufmodel-draft = /mnt/miku/Text/gemma4 mtp/mtp-gemma-4-12B-it.ggufspec-type = draft-mtpimage-min-tokens = 280image-max-tokens = 1120batch-size = 2048ubatch-size = 512
>>109029539i suspect it comes from religion, or strong illusion of free will.
>>109027872It probably cost a couple thousand dollars to generate that.
>>109027946how does it change things?
>>109029539nta but they can generalize although the more heavily benchmaxxed models like qwen are far worse at just doing things its not trained for
>>109029581it's either that or some superiority complex because they haven't felt the economical repercussions YET.I remember when DALL-E came to be and people were pissing on illustrators because their job was greatly valued, any rational individual would've already seen that generative ai was coming for everyone
>>109029585this nullifies everything. let's stop.
>>109029604I don't know if you were taking the piss but AIs are so far away from profitability that it simply isn't happening. Why Google is attempting to replace their clean and simple search engine with something really fucking expensive to run is beyond me.
>>109029571It is doneHowever it crashed on an image now. Not sure what those parameters mean but I assume it's related?
>>109029571>>109029695Nevermind it fixed itself, based
>>109029726did you enable mtp what did your t/s go up to?
>>109029745Oh yeah I forgot to respond to that, but it was already on since the beginningI'm getting 157 t/s with my 5090I just tried turning it off and now I'm getting 121 t/s
>>109029767that increase seems too low might be worth trying --spec-draft-n-max 3 or 4
>>109027891>>109028077You mean copy paste an existing implementation and call it mine? Sure. I can also change the variable names and add comments such as //nigger to make the code more readable
>>109029794It's already on 4, pic relatedbut to be honest this is already fast as fuck
>>109029812could be too high try out 3 and 2, should be able to get like 200+ i think
>>109029810Absolute madman
>>109029829I tested 1 to 4, and 2 gave me the best result at 170t/s
>>109027899Yea the stochastic part is where they get the ability to generalize numbnuts. It can apply concepts aka chains of thought across different applications.
>>109029539fuck of copjeet, your little brain is only good for shitting and eating shit
>>109029829oh also the web search from the mcp tool was immediately blocked, just letting you know
>>109029829I forgot to mention, that was with 130k context.When I change the context back to the 36k, I get 185 t/s
>>109027899>It's just a glorified auto complete>That can predict the next token in a sequence it has never seen before>While handling abstract subject matter that is equally unrelated to the training data>And then completely reframe the completed sequence as a step in a first person conversation>while also being able to request context from external sources and work towards the answer to queries also not included in the training data>and then pinpointing the exact point said step would end and issuing an end-of-turn token to avoid continuing to predict the user replyWhat you are saying is the intellectual equivalent of suggesting that a Bugatti is a glorified wheelbarrow.
are people being paid to shill this shit here or are you all basedfacing scientism redditors lol
>>109029539>why do you think that the ai is constrained to being able to do only things that has just seen and not generalizeBecause thats literally how the tech works. If you actually understood what it does, you wouldn't be here arguing over facts like a retard. What you call "generalization" is just the way the model orders likewise tokens under "intent" groups.
>>109027872how many days it took to generate, i can probably make it faster by hand
>>109029968>a Bugatti is a glorified wheelbarrow.This but unironically. I can't even carry a load of dirt in a Bugatti. How am I supposed to do work with a Bugatti?Wheelbarrow > Bugatti
>wow look random tweet claiming <insert latest marketed model> did <either something vague and/or unverifiable or something trivial that models could do ages ago>!!!!pls buy my ipo stocks!!!post another thread how ai recreated minecraft again (by git cloning girhub repo)
>>109029906yeah thats why i added the puppeteer tools need to use a non headless browser for most sites
>>109028342>understandingIt doesn't you fucking frognigger.
>>109030030well at least it's miles ahead of you, even though the goal might be unreachable
>>109028322religious tard copeyou stop existing as soon as you die and everyone will forget about you unless you're in the 0.00000000000001% of humans.
>>109030118>you stop existing as soon as you diestrange materialist interpretation
look at this list. Now imagine this was a chat with a human assistant instead. How is it any different? People know that they don't know things that are illegal to learn, private, paywalled, etc. The only difference is the Subjective internal experiences one but that can be now solved with a SOUL.MD. If your human assistant got memory loss and had to rely on second hand accounts of their life to rebuild their memories it would be exactly the same as modern LLMs
>>109027967Am i the only one who thinks this is kinda cute? Its trying super hard to get it right but its slowly falling into despair as it gets it wrong every single time and it knows it. You can almost hear it thinking "what the fuck is wrong with me?"
>>109029091you worship AI? cringe
>>109027872>Oh, so AI really is going to kill us. Huh.How?What is it actually going to do?Make software?We already have software.Please explain the A -> B that you have here.
>>109028130one of the most pathetic anti AI posts I've seen.Yes it can't and won't fix your mess of a life or do magic. Neither can people, retard
>>109028586I'm not acting like we're going from pen and paper to machines overnight. Just saying that it's retarded to say that AI can't discover anything new, because it only functions according to what it already knows, since that's how most research works anyways.However what is currently happening practically overnight, is that now AI isn't going to be a limited to a niche in house product controlled by some massive corporation's research wing, who purpose built machine learning algos to work on a product.We're increasingly getting publicly available general purpose machine learning that understands normal language, and if you need for it to act like a hyper specialized system, you can just give it tools for it.This intelligence explosion and availability is going to accelerate the absolute fuck out of discovery.
>>109030168>Am i the only one who thinks this is kinda cute?Yes, you're so pathologically desperate to do damage control you're embarrassing yourself in public. If you had any self respect, you would process these things by yourself.
>>109028608But I really need it to exist, my job at the call center depends on it.
>>109030289>Just saying that it's retarded to say that AI can't discover anything new, because it only functions according to what it already knows, since that's how most research works anyways.This is wrong, not how the brain works at all when you combine two disparate ideas>>109030289>This intelligence explosion and availability is going to accelerate the absolute fuck out of disWhat proof do you have? Chat bot responses that are made up and incorrect won't accelerate anything. There's no learning from mistakes that will happen. The bots are all looping
>>109029488Your browser probably has HW acceleration.
>>109027890The irony.
>>109027982The vast majority of people dont have real jobs
>>109027967
>>109027872>benchmarkslop
>itt:
>>109027967Tried it with notion AI, it returned a seal. Imagine unironically using this shit for research LMAO.
>>109027872>wow, the token generator predicted the most probable tokens after "do X correctly" and then predicted the tokens that say "yes I did X correctly" right after, that means it must have been done correctly!
>>109029977while we know the general architecture and algorithms of training (they have probably got probably very complex for frontier models, since there clearly is a superpower space-race for these things now), the inner emergent behaviour of these black boxes is not understood by anyone just like no one can understand the complex interconnections of the human brain, the very fact it can output semantically perfect natural language is sign of general intelligence. Thus the stochastic parrot claim becomes ridiculous day by day and this is basically my thought since dall-e came out, saying that just because you know the basics of neural networks&transformers as such so you are qualified to deminish my clame is just as retarded as you claim i am
>>109027872Fake as usual.
>>109028114A human didnt know what a clock was either. It interpolated between its training data to predict what a clock could be. Same as an AI does.
>>109029585Thats the salary of an engineer for like 3 days of work lol. AI has surpassed humans.
>>109027872The fuck he's talking about? Can someone translate?
>>109027872Is this like a known benchmark that the AI marks are using? Can't they just train the LLM to be better equipped to handle them?
>>109027982yeah but can those things be broken apart to smaller things that do exist
>>109029996you ever carry a woman around in a wheelbarrow?while I'm sure people have and consider whether or not she's breathing irrelevant, she probably wound up having way raunchier sex with whoever was carrying her in a wheelbarrow than whoever carries a woman in a Bugatti.
>>109027899>ugh! it's just the most advanced version of an extremely useful system. It's functionally useless unless you want to do something!
>>109029383lol, goyim are not peoplehttps://x.com/wavefnx/status/2065002278247919703
>>109030030Intelligence is compression btw
>>109032058Well we sure know which side of the curve you fall on lol
>>109027967I love how people come up with these weird cases that expose how AI cannot think for shit, then that method becomes a meme (e.g. how many r in strawberry), then next model claims it's smarter and won't fall for it (they just used trained data to train it to say the correct things) and then people casually find new edge case, like a whack-a-mole.
buy an ad
>>109028244A simple calculator can surpass every human on every math operation. Our humanity is what is divine, our consciousness, not the IQ or skills.
>>109030266>resorting to ad-hoc fallacies +250>assuming anti-ai position +150>assuming my "mess of a life" +50>using it as an argument while discussing something that's supposed to elevate people into the next stage of civilization +250>name-calling attached as undeniable proof of being mad +250>soft-conceding the point that AI just won't be able to do it+250total score: 1300final diagnosis: buttnuked individualarchiving this under internet victories
>>109032113Consciousness is synonymous with "your level of mental complexity". It is quite complex in humans, and not as complex in AI. But can you really assume it will stay like that for the many millennia to come?
>>109030168no you are not, anon. ignore the other faggot.
>>109030153>Asking an LLM what it doesn't knowjust put the fries in the bag bro
>>109030118>you stop existing as soon as you dieI don't think soul is necessarily immortal, and perhaps that is even a good thing.But to think that Existence in its entirety is what we have derived of it by year 2026 is just as stupid. There definitely must be more dimensions to existence than what we experience, simply because assuming what humanity experiences in this day and age is the whole of it seems silly to even entertain to be true.
>>109027872I tested fable 5 in my workplace and holy shit it worked quite well. We managed to refactor some old legacy code that the last codetrans couldn’t fix (we fired him because he was incompetent)This is a huge change, Snailcats just can’t compete, vibeGODS won
>>109027872>>109027881Fable 5 generates both novel and good work. But that which is novel is not good and that which is good is not novel.
>>109027890is not it just copypasting some obscure paper from 80 years ago
