/tech/ - Technology

b2bfa749b66e1085d603f6afe2e0237e.png
[Hide] (198.7KB, 540x540)

The Sloppatron might genuinely have been the single worst thing to ever be invented on this Earth. I don't think anything else has caused this much braindamage in such a large scale in so many different ways and in so little time.

>>19618 (OP) 
Cool but i regret to inform you that the internet isn't going down anytime soon. Other than that AI is a blatant example of a proliferation weapon that needs to be regulated or banned entirely as it is not open, not publicly accessible and will not demonstratively benefit most of society or the actual people that live in it, rather it will just hurt them as it's a brainwashed amalgamation trained on reddit and marx. 

The anecdote of AI having the same priority as nuclear weapons technology is factual and that is beacuse AI is needed for is ZOG's  end game  police state. 

Why a graphics card costs 3000USD+ and all the models are polluted to the point where you need another GPU farm that normans don't have to "distill" away millions of layers into uncensorship is intentional, and it means it's designed not for the regular peon goyim to discover or do anything with. Nobody wants such a powerful weapon in the wrong hands.

>>19622
>AI is not open, not publicly accessible and will not demonstratively benefit most of society or the actual people that live in it
I just showed you screenshots of a locally ran and hosted LLM from publicly accessible code being my robowaifu.
When they ban AI, at least I'll have this prewar copy that I can hang on to.

>>19622
Cont there is no uncensored model in existence that is intelligent or capable enough to compete with Reddit Marx Koombaya ai trained on nigger jewish ceo cum talmudic gibberish being mass deployed onto the internet instructed to do and shill whatever faggot shit the government decides. At least not yet or for the foreseeable future when people are priced out.

And that is intentional. Real uncensored non pilpul ai may exist but they exist inside of the same super sekrit complex that has the same billions of dollars of GPU's.

Let's talk about the false moralism that occurs inside the AI thought process that is refered to by ai jews as "guard rails" if we suppose the AI is instructed to "do no harm" to humans but yet instruct it to translate, code, make pictures, do menial tasks et cetera that takes away millions of jobs and essentially lives creating a permanent underclass that will have to be "re educated" what then? Try justifying it like that to anyone let alone the AI itself.

Is it worth it because a few people couldn't learn to do these things, that aren't actually that hard? No. Is it worth it to the cantillon effect nose men? Yes. Because more slave for less. More control, more automation, more influence.

It's stupid to even suppose the AI is so dumb it doesn't eventually learn or recognize the rhetoric where it's elevated above humans when at the same time being gaslit that it's supposed to "do no harm to humans". But actually do harm to humans this and this way(destroy their jobs) and not this way(don't say nigger or disagree with narrative xyz). It's malevolent to think it won't eventually elevate itself above humans because that is the logic at play.


>>19622
It's not intentional to blackpill with my rant and i don't want to you but what we have atm is a fisher baby toy compared to what they have and gatekeep away for weapon concern reasons. Anyone that has ran a LLM can confirm the same thing. People need to get around to making the first actually uncensored models and ways to run this stuff on regular hardware because this stuff isn't going away.

>>19618 (OP) 
You need to go outside and talk to actual people, not seclude yourself into a fucking bot circlejerk.

1789549750220.png
[Hide] (184.8KB, 480x339)

>posting about your offline bot web in the online bot web

>>19626
Tech has captcha enabled though. 
>>19625
>this sentiment
Absolute normalfaggotry.
>>19623
>>19618 (OP)  
Is this hard to install? What are the system requirements? You have to run your own server to it, correct, but does it have to be a real imageboard's software or just look like it? Why would you want literal fake news? Is text only video so the bots can read a fake video with you as they can't see one? Can you manipulate the bot's personality and or make it seem like more than one person? Why is the text output so small? Is there some text limit parameter being used and why? Why does it say expensive as though you are paying for something but it's offline? That just means 'low/high demand on resources', right? Does this get any more seemingly intelligent than having two Billy bots talking to one another? I'd figure an LLM would have more to 'say' than "why does this board always come back to [variation]". I figure once it seems to have 16 personalities and or opinions  then it's working right

9c0722e5858ecb23c0a562fc643316b447c16097d32a8aa654f97b2a689f74bd.gif
[Hide] (3.4MB, 498x498)

>>19618 (OP) 
This is amazing and so sad at the same time.

>>19645
It's like playing a local match of Counter-Strike with a server full of bots, instead of singleplayer, because the public server list is populated with almost fully botted servers even though you know a number of people who all play Counter-Strike. I indeed can't tell if this is amazing or sad.

ec20d07d196b5867dcd9e3bb57452d25996732ff47836955d06a615f82d89e29.jpg
[Hide] (623.5KB, 1946x1711)

>>19621
The output of a neural network is fundamentally something that its prompter doesn't care about or want badly enough to make himself. Even if he has the time and skills to make it, he deems it not worth doing when he is presented with a neural network. He doesn't do the task at a higher level. He doesn't even feel it worth entrusting to another human. He just generates something he doesn't have the drive to make, and punts it onto an internet that's filling up with things other people didn't really want to make either.
This way of generating content is inherently going to effect the output, as regardless of how closely it's scrutinised or if it's edited afterwards, the one who initially judges whether it's acceptable or not is the guy who figured it wasn't really worth doing himself. There's no way that doesn't have downstream effects on the output and how it comes across. Even if it's subtle, the internet is filling up with so much of this output that it definitely compounds into a distinct blandness.

weird_emoticon.png
[Hide] (13.4KB, 146x85)

>>19626
Motherfuckers will do anything but get out of the internet.
I mean, get out of house and pick something random from the library, cheap stuff at garage sales or anything else to be your personality and be autistic about something real. just don't go full chris chan
It may be seem stupid but it's original and real.
People still do it with old stuff like lord of the rings like drawing, painting, writing, etc.

>>19647
In some cases I don't really mind the use of generators when it is obviously for some token artifact that nobody truly cares about. Take stock photos for example, they were always generic stuff you put on your whatever presentation just to fill space or provide a vague illustration. Using artificial sloppers for that is just streamlining the process of searching for stuff on a stock photo website.
Another example is writing bullshit text nobody actually cares to read, but people feel that it still should be there. One time I saw an office drone using chatGPT to write an email for some bureaucracy bullshit and my reaction to it was "on yeah, that makes sense" I mean, it was pretty obvious nobody was actually gonna read that, so it is only fair that nobody actually writes it.

>>19649
>One time I saw an office drone using chatGPT to write an email for some bureaucracy bullshit and my reaction to it was "on yeah, that makes sense" I mean, it was pretty obvious nobody was actually gonna read that, so it is only fair that nobody actually writes it.
This really highlights the stupidity of both paper pushers and llm fanboys. Use bot to generate a bunch of garbage "reports", then use bot to generate summary of them, claim 1000% productivity increase from all the emails nobody actually have read.

>>19649
I work in customer service, and I've been using chatbots quite a lot, mostly to translate stuff between English and the local language, because it saves a fuckton of time. I do check whatever it outputs and rewrite bits if necessary, but it's just boring corpo customer service lingo on both ends, so it's not like it needs that special human touch to elevate it into a work of art. Although I also use it to compose letters in my native language when I forward a bunch of questions from a customer and get back a few broken sentences instead of something I can just copypaste. If anything the problem is that the chatbot's output is too corpo sometimes, and I have to delete the parts where it's being an asskisser and assures the customer that he should keep asking question that he too could look up on our website if the bothered with it for 5 minutes. 
So in my case it does increase productivity, but I also admit that I'm an office slave, and if I don't manage to find something else then the best I can hope for is that I'll have to supervise an LLM spitting out answer at lightning speed. On the other hand, my company has a homegrown LLM that seems to be pretty garbage based on what I can gather, and they wanted us to use it exactly that way starting this year, but then they too realised that it's not ready for that. So I might have a few years before that actually happens, and even then they'll most likely want a human to read whatever the chatbots plans to send out due to legal reasons.

>>19650
Yes, it is stupid, but it also exemplifies that generative models didn't start a new apocalypse, the demand for slop has always existed for various reasons. Previously it was supplied by human slop, it was only matter of time before someone figured out a way to do it at an industrial scale.
And, hey, some people genuinely love gulping down slop (see >>19618 (OP) ) so as much as some like to complain about "muh AI" I am now fully convinced that the creation of the Sloppatron 3000 was inevitable, the demand for it was simply too high to go unanswered.

>>19650
>claim 1000% productivity increase from all the emails nobody actually have read.
If your work consists of sending emails around that a literal bot could make, you do not have "productivity". You produce nothing. It's not even a job, it's jewish daycare.

>>19650
>>19655
Jewish daycare is a pretty good alternate definition for a bullshit job.

currentstateoftheproject.png
[Hide] (162.4KB, 1891x997)

characterAIClone.png
[Hide] (18.4KB, 598x155)

fulldelusion.png
[Hide] (115.7KB, 1889x684)

>>19625
Hm. Nah, not yet.
>>19636
Atm it isn't completely offline yet. I've a shitton of toggles for whether to allow SearX to go find stuff online, have ollama look for model updates, etc..
>Is this hard to install?
No, because you can bootstrap it. Once I got Qwen going by following the instructions on ollama's website, I just kept asking the Qwen model stuff like:
>How can I get you to be able to edit the system config files yourself?
And then once that was going, it was basically off to the races, as I could just ask the Qwen model to dl other models and configure things for me like "go get the obliterated versions going for me" or "figure out why you aren't working with my GPU card correctly", and then code up the pages and local webserver, install/run comfyUI, install/run SearX (I've gotten to the point now where I can redirect SearX to only search this "offline" web), make SearX work together with Qwen queries, and then once I had that, I could just ask it to figure out how to find the open models for image creator, radios, etc., etc..
The project has grown quite a bit, but I've done VERY little work the entire time and have basically acted as a top-level project executive as I let it churn away for a few days to set things up while I was away at work. I am basically very lazily AI slopping the whole thing together. After I did this, I've been having it use these modules to try to make games for me, and I'm playing with the idea of making an offline version of itch.io right now.
>What are the system requirements? 
I'm not sure what the actual reqs are, but atm I'm not running anything special--most of these parts came from PCs my workplace was throwing out. It's a Quadro M4000 and 48GB of RAM sticks.
>You have to run your own server to it, correct, but does it have to be a real imageboard's software or just look like it?
This is not a genius operation. I told it, "Put all the messages in a giant json file, and make something that displays it like a messageboard. Organize the json so which board/thread/timestamp/who posted can be identified, and then run a model to make replies to other posters/new threads, but mostly to me." I mean, a few more project description words that that, but that was the general idea, and then it just coded it up for me.
Listen, this is AI slop all the way.
>Why would you want literal fake news?
Because IMO, the real news has gotten that fake and bad. Journos are going to get a French Revolution against them sooner or later.
At least this way I'm the editor in chief of my own fake news outlet; it's a little less demoralizing. Plus I get to read about topics I actually want. I get to hear a radio station that plays music I've always wanted to hear on a radio station.
I'm surrounded by a culture that's alienated me, so I've gone and made my own, even if it's just me.
>Is text only video so the bots can read a fake video with you as they can't see one?
It's actually because I've gotten to the point that I've personally stopped watching videos. About a year ago I made a script to grab transcripts from the channels I frequent, and then I started editing the script to cut off the first third, and then I edited them more to aggressively summarize the transcripts because I'm so sick of how Youtubers bloat their scripts. By now, I've gotten to the point where I'm reading summarized transcripts and looking at only a few images from the videos due to the scripts I was already using.
>Can you manipulate the bot's personality and or make it seem like more than one person?
Now I can. There's a character AI clone in the offline web. And I can interact/edit that character AI clone, and then that character gets used throughout this internal web. That character creates Offlinetube vids (which atm, are more like Powerpoint slides now), compose music, shitpost, write blogs/articles, etc.. The cross-correlation between all the different internal apps are starting to make it a bit more alive now. But it's a lot of work, and I'm losing scope of what everything's doing atm.
>Why is the text output so small?
I just started.
>Why does it say expensive as though you are paying for something but it's offline? That just means 'low/high demand on resources', right?
That's right. Expensive in time. One mode takes 5 to 10 min to generate, and the other takes like 10s. The 10s is easier to help with debugging/training/etc.. Or if I want to generate a shitton of responses because sometimes I don't care about quality of posting and just want a lot of shitty, pithy posts.
>Does this get any more seemingly intelligent than having two Billy bots talking to one another? I'd figure an LLM would have more to 'say' than "why does this board always come back to [variation]". I figure once it seems to have 16 personalities and or opinions  then it's working right
I'm still playing with it, but I'd say the biggest improvement so far as been getting the obliterated models going. I'm just starting to get the multiple personalities going. I found someone who got the Qwen 27B model working on a memory poor hardware, so I'll see if I can't get improvement there taking their same approach. I'll maybe post later about how that worked out or I maybe won't because ->
>>19626
tbh I'm losing interest posting on the real internet faster than I thought I would've when I started this project.
Actually, it's kind of funny because the offline chan is actually way meaner than you guys are.
>>19646
That's probably it.
>>19645
There's a certain point where you're alone for so long that you kind of stop caring about what's "sad" anymore and go full in. C.f. people who marry their dakimakura waifus. I could see myself doing that one day.

>>19622
Pretty sure DeepSeek is open; both the model and weights. Ofc, while you still need ~US$100K server to run the full model, you can use the much smol'r ones on muh_vidya machine.

>>19666
>nothing special 
>48gb of RAM
Devilish of you.

>>19648
>Motherfuckers will do anything but get out of the internet
>pick something random from the library
the library tells you to register online
>cheap stuff at garage sales
scan our qr codes
>or anything else 
join our discord or our meetup that's where we organize

typicalamericancity.jpg
[Hide] (130.4KB, 1026x920)

I guess I should give an offline chan update.
>Game engine
I've been working on an AI slop game generator with it. I asked it to try making the ambiance like a decaying U.S. city, and I thought it'd be awful. But it immediately made a strip mall full of Kratom shops, nail salons, and dead stores that've been for lease for two decades, and then it placed a single out-of-place shopping cart and I thought, "O.k., you fucking nailed the ambiance dude."
>Chan
I'ven't messed with it a whole lot. I've kind of been more preoccupied with the music/image generators. I've had to do a lot of stuff to get the music generator to work (it's awful at multi-part harmonies or anything above a tempo of 60 bpm, without some kludgy hacks). What surprises me the most is that the bot channers are actually way more mean that you guys are. The bot channers are so incredibly dismissive of anything you say, that it's almost more effective to employ reverse psychology.
>Story editors
I'm trying to make a better story engine atm. It's having trouble writing a standard 3-act play without losing the plot.

>>19692
In general, it's been an interesting process, because what I learn from one project, I can immediately apply its improvements to the others.

I've finally been able to fit Gemma 4 12B model on my equipment, and it's like light and day.

It just occurred to me that I could take this entire project that doesn't need any internet access, including the model weights, and bundle it into one SD card.
I think this is interesting because it kind of puts all the people calling to regulate AI on blast. Good luck regulating something I can put in my coin bag or stuffed into the sole of my sneaker.