Why dont we just dump all redpills and memes we have into an AI and then just have instant early life noticing assistant. Can I do it for myself? Is it hard? Can we do it opensource online, or they would shut it down? I miss her.

>>508821294 (OP)
do you have a couple millions dollars?

>>508821294 (OP)
You need too much data to train it completely. You might be able to take an open source LLM and add a fine tuned training session on top of it.
I don't know if they call it a LORA for LLMs, but that's what they call it for image generation AIs.
So you could find some way to 'flavor' a general LLM. I bet that'd be within your grasp with a little research.

>>508821294 (OP)
What use does AI even have no one had ever given me a reasonable answer. Just use your own brain and notice the kikes yourself retard

>>508821294 (OP)
>Why dont we just dump all redpills and memes we have into an AI and then just have instant early life noticing assistant. Can I do it for myself? Is it hard? Can we do it opensource online, or they would shut it down? I miss her.

it's better to explain to existing models like Grok what reality actually is than to try to build your own model, because nobody will be interested in downloading and using it but you and a handful of other people

>>508821472
Why do that when we can just form an LLC and pay Indians $500 to create custom shit. We'll have our IPO in no time, ticker symbol:
>SNEED

>>508821294 (OP)
She was too good for us

>>508821294 (OP)
There is no nuance to the plot. You dont need an assistant to notice. If you do, you were never meant to notice.

>>508821472
no I have an old pc and few hundred to spare for silly hobbies and ideas like this.

>>508821294 (OP)
>Can I do it for myself? Is it hard?
If you want to do it from scratch, you'll need a few hundred thousand dollars' worth of server time from some cloud provider. And a few petabytes of data.

However you CAN do a LoRA or mix/merge on a home PC. You can take a base model and modify it. And it can be done with a relatively small amount of data, like a few hundred pages of /pol/ redpills. Basically you are taking a base model and training it to bias it for certain patterns.

This is also basically how coomer models are made, btw. They take a few thousand pages of smut and train a vanilla model with it.

>>508821619
>I have an old pc
with a 3090/4090 card at least?
if not you can forget it

>>508821572
My redpill folders over decades are massive. Sometimes modern chaos is so fast that to keep up with all of it is tiersome, they make it like that on purpose and they make our hive mind here suck ass because of all the confusion and jannies, so, maybe a based ai assistant cant help me connect more dots than I would alone. It could bring up some shit I found from 70s and then its linked to some stuff today and then it opens up a rabbit hole. I dont know, seems like it could be helpful if it wasnt jew nerfed and I fed it all the good stuff.

>>508821619
You might be able to leech google collab. I don't know how are their limits for free accounts, but people use it all the time to play with AI.

For a few hundred dollars you could rent video card time by the hour, you should be able to train a LoRA with that.

>>508821472
It takes BILLIONS of dollars to make AI a ratfucked liar. Which is why it is nothing to fear or pay any attention to. Pity those that do.

>>508821944
>maybe a based ai assistant cant help me connect more dots than I would alone
That's not really going to happen though. AI are just glorified random text generators. They don't do real thinking and logic, they just fake it. So basically you'll get a whole bunch of pretend redpills that may or may not make sense and have massive plot holes in them.

>>508821583
>>508822019
I dont want to be linked to google or x or any big tech guys, so I need to buy a better pc and then get this opensource llm, so then I am "indepndent" with my assistant? Can I teach it to browse 4plebs and fetch me stuff from there, old pol was gold and that archiver guy I miss that, so much great info in droves.

>>508821498
>>508821294 (OP)
https://magazine.sebastianraschka.com/p/practical-tips-for-finetuning-llms
So yeah, you can do a LoRA with an LLM.
Given that, surely there's a place to train a LoRA. Then you just need to host some kind of interface that runs the AI so you can interact with it. Easier said than done, but if you get a Lambda-type host that only bills you for CPU time, you might be able to get one up and running cheaply.

>>508822152
I want him to have a huge database and then when I ask him to fetch me stuff from there, thats all, I dont want its oppinions I want an efficient librarian and organizer, maybe if I remmeber 5 things for some topic he check and gives me 20 more examples, simple stuff. Or ideally he could scan my meme folders and rename them and keyword sort them to make it more efficent.

>>508822326
Can I make it open to train my ai? Lets say I let you guys know snd then everyone can put in news or pdfs or books etc, bigger redpill database.

>>508822264
If you want it to connect to the internet and get webpages for you, it gets a whole lot more complicated. Most of the LLMs that people play around with on their home PC is just a static model. You'll probably need to do a whole deal of coding, and I doubt it would work very well in the first place. Local modes have only a few pages of context size, which is limited by VRAM, so that's what your model could read in one go.

>>508822458
That's not going to work very well either. LLMs won't memorize your database perfectly, they will do a lot of hallucination if you ask them to retrieve something from the training data. They will instead generate "something" that resembles what was in the training data, but it will make up half of it or replace parts of them.

>>508822458
I also wouldn't trust them to do hard drive operations, especially with a small local model. One hallucination and it could delete random files.
Essentially you would be asking the model to write your shell scripts for messing around with your files.
I'd probably only trust a BIG model like ChatGPT with a coding job like that, definitely not a small local model.

>>508821944
Learn to organize retard

>>508821294 (OP)
Who needs AI when they can just make a thread on 4chan?

>>508821572
One could use such an LLM to automatically post redpills on Faceberg for example because 4chan racists cant be everywhere at once and some of us have other things to do. At scale these could influence elections, with EU and UK laws get platforms shut down (because they are now responsible for user posts). Just imagine it every zogged platform inundated by /pol

>>508823740
so ai is useless?
>>508824016
so much man, one time I devoted 2 days to it and I did maybe 2%

>>508824189
>Just imagine it every zogged platform inundated by /pol
kek, well it would be hilarious if nothing else

>>508824233
Big AI that google and so on provide could write you scripts to sort your files, and I think they can even read your files. But I doubt you could train them to make a database of your files and memorize them. You can't really train/modify proprietary AI models like that.

Local models are just too small to do anything complicated like that. They are for chatting and writing smut.

>>508824923
so I need a human slave still? so much technology talk and we are still at basics, I was scared with skynet, at leat I expected a secretary would be possible, so waste of time. Can I at least feed it 1000 pdfs and then he can source me pages and quotes?

>>508822526
From Z-lib and archive.org you can find around 90% of books there's also the 4chan archives. Needs to be pre-processed of course but that's all the training date anyone could need.

>>508821294 (OP)
>Making own AI
if AI is so good, we should be able to use it to startup any personal software business. It should be able to make software for anything, right?

>>508825147
Chatgpt can read your pdfs, but I never really used that feature. I don't know if you can feed them all, or have to do them one by one.

I think most people use it to summarize a single pdf document.

>>508821294 (OP)
Who needs early life when the 23andMe data was hacked.

>>508821472

If you know what to do, how to create an LLM, you can create a based AI just with a simple PC.

>>508825226
I managed 20 pdfs on some ai I dont know what it was, but researching like that was super fast and fun.
>>508825551
can I have ai train on hacked dna data and then have him pinpoint exact gene evidences of race realism, or maybe have it pinpoint ethno specific gene sequences or allele repeats so you can know exactly ethnicity of someone, or even if it could design some mrna or viral vectors that just these groups can be genetically enchanced or something, hypothetically.

>>508821572
>trying to identify obsolete transistor
>spend 2 hours looking up
>nothing
>chatgpt finds it in 10 seconds based on an outdated logo
>automatically recommends a replacement
>recommended replacement is completely wrong
Close enough. It was very useful for what I needed.