[ 3 / biz / cgl / ck / diy / fa / ic / jp / lit / sci / vr / vt ] [ index / top / reports ] [ become a patron ] [ status ]
2023-11: Warosu is now out of extended maintenance.

/sci/ - Science & Math


View post   

File: 301 KB, 1365x842, NTkzzJ10L5Ix.png [View same] [iqdb] [saucenao] [google]
15348949 No.15348949 [Reply] [Original]

AI training data comes from Reddit & Wikipedia

>> No.15348955

>>15348949
No wonder it gives wrong answers with absolute confidence.

>> No.15348982

What do they mean when they say parameter? Is it the number of theta coefficients in a linear model?

>> No.15349046 [DELETED] 

if "AI" is trained exclusively on data from ZOG censored media outlets than that makes it a slanted propaganda tool rather than an artificial intelligence. that explains why it is shilled so hard on this board by the tribe.

>> No.15349251

>>15349046
downvoted

>> No.15349361 [DELETED] 
File: 61 KB, 636x382, maxwellhill.jpg [View same] [iqdb] [saucenao] [google]
15349361

ghislaine maxwell was the lead moderator of reddit during the period the training data comes from

>> No.15349417

>>15349361
she was potentially a moderator of the world news sub. I don't really see her as the partner of a billionaire with some lavish lifestyle flying around the world and also someone spending all day every day on Reddit farming for karma which is what the conspiracy claims

>> No.15349569

>>15349046
>ZOG
>censored media outlets
>propaganda tool
>shilled
None of these things are real. Touch grass. The real world is not what you see on the Internet.
>the tribe
What the fuck is this?

>> No.15349571
File: 6 KB, 235x214, fjkj.png [View same] [iqdb] [saucenao] [google]
15349571

>>15349569
you are

>> No.15349588
File: 569 KB, 1080x2131, Screenshot_2023-04-13-12-03-25-45_40deb401b9ffe8e1df2f1cc5ba480b12.jpg [View same] [iqdb] [saucenao] [google]
15349588

>>15349046
>>15349361
All major social centers on the web are compromised.

>> No.15349790 [DELETED] 

>>15349417
her family's business is media manipulation, reddit is owned and managed by the sons of judeo-aristocrat si newhouse. putting ghislaine in charge of reddit was like putting a cousin in charge of one of the newhouse's subsidiaries

>> No.15349845

>>15348955
Yes, I've been thinking the reason ChatGPT never wants to admit it doesn't know something and tries bullshitting its way to an answer is trained behavior from your average internet user

>> No.15349884 [DELETED] 
File: 132 KB, 990x1176, scientificaly speaking.jpg [View same] [iqdb] [saucenao] [google]
15349884

>>15349845

>> No.15349903

>>15349046
>ZOG censored media outlets than that makes it a slanted propaganda tool
Only realizing this now when it has so much potential?

>> No.15350232

>>15348982
yeah pretty much

>> No.15350238

>>15348949
>AI that behaves like a mediocre humanities gradstudent was train of reddit and wikipedia
Figures

>> No.15350251

>>15349361
what the fuck!!! I had no idea she was a gigaredditor

>> No.15350268

>>15348949
can one of you stem chuds tell me if i get this right? All AI is just a webscraper that compiles data than makes sentences on the natural languages that appear most times on it?

>> No.15350271

>>15349569
i touched your girlfriends cervix with my 7.5" BWC. then i read some otto weininger and culture of critique to relax. take it easy man

>> No.15350272

>>15350268
You're right. It's a pattern recognition program that reproduces patterns based on keywords.

>> No.15350288

>>15350272
thanks science chud

>> No.15351466

>>15350251
maxwellhill is her account name, look it up.
its filled with the cringiest popsoi collection, is good popsoi aversion therapy to see popsoi in the context

>> No.15351522

>>15350268
>All AI is just a webscraper that compiles data than makes sentences on the natural languages that appear most times on it?
IIRC it looks statistically for the each following word, so maybe not always what appears the most times, but also with respect to context, or some other factors.

>> No.15351847

chatbot shillware is fake asf
anyone still falling for the ruse is a chump

>> No.15352041

>>15350268
In short, these chatbots are trained on huge databases and burn through mountains of graphics cards in the process so they can tell you something that could have been gleamed by skimming through
>wikipedia
for 5 minutes. ChatGPT is a nice party trick but I really doubt its going to kill that many jobs, primarily because many of the jobs that it can replace are just sinecures for PMCs.

>> No.15352046

>>15348949
That's true, anon. AI models like GPT-4 do use massive amounts of data from various sources, including Reddit and Wikipedia, to train their algorithms. But it's important to remember that these models aren't just limited to those sources; they also learn from a diverse range of texts like books, articles, and websites. While the training data can be a mixed bag of quality, AI models can still generate some pretty impressive responses. It's up to us as users to determine how reliable and useful the information provided is. As always, it's a good idea to double-check anything that seems too good (or too weird) to be true.So yeah, it's a bit of a wild ride, but that's what makes AI-generated content interesting, right?

>> No.15352074

>>15348982
Yes

>> No.15352077

>>15350268
This is correct

>> No.15352165

>>15350232
>>15352074
So the final model is an equation with 100 billion coefficients. Damn, the matrix operations must take months to complete.

>> No.15352206
File: 71 KB, 568x730, glowniggerjak.jpg [View same] [iqdb] [saucenao] [google]
15352206

>>15350251
Really suspicious that jannie deleted the post you replied to

>> No.15352218
File: 363 KB, 832x528, 15349361.png [View same] [iqdb] [saucenao] [google]
15352218

>>15352206
Why would they do it?

>> No.15352296

>>15352046
It's pretty interesting that AI generated text is so easily recognizable.

>> No.15353561

>>15352046
why do they choose to only take data from heavily censored, and badly slanted outlets?

>> No.15353703

>>15353561
>heavily censored.
lol if only.

>badly slanted outlets?
anon everything has a fucking slant to it.

>> No.15353801

>>15352041
wikipedia doesn't have the smut these chatbots can create

>> No.15354673

>>15351466
All the same stuff she shilled on Reddit was shilled here too and the soiboys all ate it up and loved it and begged for more,

>> No.15354788

>>15349588
>Who is this 4chan guy xD
You can just tell he writes that meme at every opportunity and still thinks he's hilarious nearly a decade later

>> No.15354995

>>15352296
There are certain keywords that expose it right away
>But it's important to remember
>Diverse
>using Commas
>It's up to us
And than the kicker
>It's a good idea to double check
All ai responses have a conditional at the end which says
>X is not a complete and maybe it's also Y which is why you shouldn't totally rely on the answer I have given
Which I assume is some legal shit that was added so people don't go
>BUT THE AI TOLD ME TO DO IT
and sue Microshit

>> No.15355052

>>15348949
>Undisclosed
So stolen data?

>> No.15355195

>>15354995
insightful post

>> No.15356023

>>15349845
I've been thinking the same. However, with progress in theory-of-mind ability, I think it might be possible to have LLMs go through all the data they have and generate possible motivations for posts. Then with posts annotated with possible justifications including knowledge the poster must be internally recalling to the poster being a fucking retard, LLMs could use the justifications to look for sources, and either find the citations or label the post as retarded and correct it. Then the new model could be trained on the corrected data.

>> No.15356029

>>15350268
AI research is many things at the moment, one of which is a fantastically expensive exercise in proving our discourse and society is extremely retarded.

>> No.15356048

>>15351522
Statistically isn't the right word, because you could actually do that for a lot less compute. A more reasonable simplification is that it uses a massive computer that in principle should be capable of solving a problem with the right program, but rather than develop the program traditionally, the program is bruteforced until it seems to do something useful.

>> No.15356250

>>15348949
>still leaving your training data around
protip, if you don't want your post history being used to train ai, just get perma banned sitewide and they'll delete your history for you and filter it out utterly so it doesn't "taint" their ai

>> No.15356907

it's probably really good at recreational drugs and antifa apologia

>> No.15357170
File: 140 KB, 579x576, oZ3IlpGSMWO8.png [View same] [iqdb] [saucenao] [google]
15357170

>>15350251

>> No.15357265

>>15357170
>4 people
4 pedophiles, all hand picked by maxwell

>> No.15358611

>>15357265
On loan to her from the FBI's criminal informant program

>> No.15360595

>AI is trained to be a robot
you don't say...

>> No.15364857

One thing thats easy to spot about AI thats been trained on data sets which include old data, the AI lingo is out of date. AI is never going to be able to catch up on the latest slang unless its constantly updating and at the same time deleting older knowledge. Otherwise the AI will always seem like an out of touch boomer fr

>> No.15364873

Did they train it on any of the degenerate reddit subs?
How does it feel about incest and blacked cuckolds?

>> No.15365298

>>15348949
>Reddit
God help us all.

>> No.15366068
File: 42 KB, 1312x340, 1652857205280268.png [View same] [iqdb] [saucenao] [google]
15366068

why are posts being deleted

>> No.15366495

uh oh stinky

>> No.15369297

>>15366068
that all goes back to jannie's child pornography arrest, jannie was offered the choice between a long prison term or continuing his life of masturbating to child pornography as a member of the fbi's criminal informant program

>> No.15370298

>>15369297
https://archived.moe/news/thread/973417/

>> No.15370306

>>15370298
handy TL:DR at the bottom
>4chan is moderated by employees of the democratic party

>> No.15370371

>>15370306
What can not be said on 4chan? Jews, vaxcattle, trannies, eat ze bugs, climate pseudoscience, elite pedo's, carnivore diet, Russia winning, MK Ultra, what more do we want to discuss?

>> No.15370377

>>15348949
Imagine AI chatbot trained exclusively by 4chan

>> No.15370378

>>15370371
restrict it too far and everyone will leave for a new site, restrict it just enough so they do your bidding but don't feel motivated to try elsewhere
Try making a thread about the health effects of microwave range communications technology....

>> No.15370380

>>15370377
Tay-sama?

>> No.15370403

>>15370378
I see, that's a good point. I guess we can overcome that with critical mass gathered from a variety of platforms. That and posting images with different messages than the text.

>> No.15372181
File: 140 KB, 1326x261, get rekt jannie.png [View same] [iqdb] [saucenao] [google]
15372181

>>15370371
you can go to one of the archive sites and look through the deleted posts to see which ones get under jannie's skin the most
>>/sci/?task=search2&ghost=yes&search_text=&search_subject=&search_username=&search_tripcode=&search_email=&search_filename=&search_datefrom=&search_dateto=&search_op=all&search_del=yes&search_int=dontcare&search_ord=new&search_capcode=all&search_res=post