>OH N-
What a shock.>Can't post genuine stuff online in most places because of moderators pushing their ideologies>Can't post genuine ideas because people will take those ideas>That leaves only daily life slop from 60+ year olds and zoomer memes from 6-7 year oldsUnless we have a genuine uncensored internet in all facets (which means all countries with censorship need to be pulled away) AI will not learn much.
>AI models are choking on AI data
eat it up aijanny
>>108751398>these faggots have been saying this shit since 2022Two more weeks.
>>108751398Don't you just hate it when models choke on junk?
>>108751398As I often state.The West: "We need to invest everything into AI to give us the ultimate surveillance device."China: "We can just turn AI into a giant flak launcher and fuck their shit up."
>>108751420>6-7
>>108751422feedback loop>ai models are trained on literally everything made by humans>they generate a bunch of data>new data from humans is dwarfed by new data from ai>as this continues, ai will be trained on more and more other ai data (including hallucinations) doesn't take a genius to realize their training data will get worse and worse, barring massive improvements in the ability to segregate human data from ai data
John Searle debunked AI 25 years ago.https://en.wikipedia.org/wiki/Chinese_room
>>108751467It's a zoomer meme. I was trying to be funny
>>108751529that's why they want to increase internet accessibility in places like India. For that high quality jeet data
>>108751398>on junk dataBut that's what AI models are producing and flooding the Internet. So AI is choking on its own output.
>>108751398Ultimately, they'll need you in a pod, hooked up to the matrix to provide the data generation needed to keep the system on track
>>108751588but will i get my bug snackbars (tm) with happy pill?
>>108751420
>>108751543He also bonked his Chinese assistants.>In March 2017, Searle became the subject of sexual assault allegations. The Los Angeles Times reported: "A new lawsuit alleges that university officials failed to properly respond to complaints that John Searle ... sexually assaulted his ... research associate last July and cut her pay when she rejected his advances."[59][60] The case brought to light several earlier complaints against Searle, on which Berkeley allegedly had failed to act.[61][62]>The lawsuit, filed in a California court on March 21, 2017, alleged sexual harassment, retaliation, wrongful termination and assault and battery and sought damages both from Searle and from the Regents of the University of California as his employers.[63] It also claims that Jennifer Hudin, the director of the John Searle Center for Social Ontology, where the complainant had been employed as an assistant to Searle, has stated that Searle "has had sexual relationships with his students and others in the past in exchange for academic, monetary or other benefits".[63] After news of the lawsuit became public, several previous allegations of sexual harassment and assault by Searle were also revealed.[64]>On June 19, 2019, following campus disciplinary proceedings by Berkeley's Office for the Prevention of Harassment and Discrimination, University of California President Janet Napolitano approved a recommendation that Searle have his emeritus status revoked, after a determination that he had violated university policies against sexual harassment and retaliation between July and September 2016.
>>108751398Prove it
Is it that they're training on ai generated stuff people are posting on the internet, or is it that the labs are explicitly generating vast swathes of training material because they've already consumed the entire corpus of human creativity? I thought it was more the latter
"ai" should be able to generate responses from first principles, textbooks, the endless sea of instruction manuals and documentation, philosophical principals, simulations it itself executes and remembers the results of, and an actual understanding of mathematics and physics.
poopy hands caused this
>>108751398God, I wish they would choke on my dick instead.
>>108751600>esl nonsense
>>108751398IGGER
>>108751467Digits check out.Want to see something retarded? Type 67 into Google search.
>>108751467Holy shmoley
AI was supposed to completely destroy stock photos. Turns out if they don't want to ingest their own vomit and collapse their models, labs have to source contamination free images of new things and concepts from somewhere.
>>108752169>The tool that's supposed to replace artists and photographers needs artists and photographers
MGS2 moment
"choking on junk data"yet they keep improving every release. luddie cope.
https://www.nature.com/articles/s41586-024-07566-y>AI models collapse when trained on recursively generated dataThe internet is full of so much AI generated garbage that it'll killing AI. This is the most predictable outcome imaginable.
>>108751630above average chinese chicks are great I don't blame him
>>108752674>2024Quick question are we going to have to read this shit all the way till 2030 or when will you faggots finally shut up about that failed cope attempt?
>>108752473they're not thoughthey're just throwing more compute at it
Who would'a guessed?> Garbage in ---> Garbage out
>>108751398fake problem based on a misunderstanding of talking points from 3 years ago, almost completely irrelevant to modern training paradigms
choke niggueh choke go head and choke
>not a single person in the entire thread has read the articlesasuga /g/https://fortune.com/2026/05/03/ai-models-are-choking-on-junk-data/this has nothing to do with LLMs
>>108753155it's fortune. do they ever write anything worth reading?
>>108751398Model collapse is a meme.
>>108751398>AI is the worst it will ever b-ACK
root@kl-874962 /var/log/nginx # du -sh access.log1.6G access.logroot@kl-874962 /var/log/nginx # grep openai access.log | wc -l730554I'm doing my part
root@kl-874962 /var/log/nginx # du -sh access.log1.6G access.logroot@kl-874962 /var/log/nginx # grep openai access.log | wc -l730554
yeah anyone who knew anything said this was going to happen, tech fags just coped and denied it
>>108753986https://arxiv.org/abs/2211.04325>Submitted on 26 Oct 2022So when will we see LLM collapse? Before or after everyone leaves the EU post Brexit?
My junk. Their models. Universal release.
>>108754018it's already happening, why do you think is anthropic kneecapping their top model 3 weeks before new model release (and then when they release is even hn simonw says deepseekv4 is pretty much on par but 17x cheaper kek)
>people who predicted this were cast out as 'schizos' and doomersHappens every time, kek
>>108753962doesnt that mean openai is accessing your server?
>>108751422oh, ho hum, looks like humies are useful after all and human-generated information is more valuable than ai slop
There are 8 billion humans, how many do you think have actually original thoughts.
>>108754403about 4 billion
>>108754403At least they actually have thoughts, AI doesn't think.
Enjoy my spunk data, ChatGAPED
>>108752098no way google shakes its ass for you
Does AI algorithmic-inbreeding explain the piss filter on AI-generated images.
>>108751398>multitrillion dollar industry that will replace workforce globally>Rajeesh you coded in that failsafe that we dont use AI generated content as training data>Rajeesh
>>108754374my server is a tarpit, it's semantically correct junk that looks like technical documentation.if I search site:mysite.tld on google, I get 100+ pages of junk results.https://iocaine.madhouse-project.org/you can also whitelist search engine bots, but I don't.
>>108751463China had the whole "mass surveillance" gig figured out well before AI. AI just made it easier to search the surveillance and format it in a way thats easier to digest.
>>108752674Middle one is different from the other two.>click next>click post
>>108752473They aren't thobeit. If you actually follow the literature, you'd know the actual improvements have been with harnesses and supporting architecture like RAG and MCP.The models themselves have actually stagnated for a while, which is why we keep seeing surprising (but not really) shit like Gemini or Deepseek in the Claude Code harness suddenly jumping from being shit on coding to being neck in and neck in a tight race at the top of the boards.Its all a fugazi.
>>108753962>>108754636LMAOthat's awesomei'm not really pro or against ai, but this type of thing is why i still come to /g/
>>108751422That, and temporal blindness. It pulls data from a source that was accurate, 2 years ago, but never got updated. It doesn't understand that things such as online service games get patches every few months, so the way classes, skills, and items used to work isn't the way they currently work. Ask it to describe systems in a game such as Path of Exile or Diablo 4, and you regularly get answers that haven't been true for several patch cycles, because the AI is pulling from some shit, 3rd-party wiki site that never gets updated.
>>108751398
>>108751420There's nothing genuine about the sewage you spam, you bark the same approved nuspeak to fit in with the most embarrassing people on earth.
>>108751529AI doesn't need to be fed shit to hallucinate, that feature comes built in. It is full of goblins by design.
>>108751398>NOOOOOOO AI IS BAD AND ITS GONNA EXPLODE (MODEL COLLAPSE COPE) Keep seething trans, vibeGODS are here to stay and to replace you.Me? On the right. (Fast, strong and Bosnian cat).You? On the left (slow, weak, and luddite South African snailcat)
>>108756332>this spammer againGo to bed grandmas,the boss fired you, putin will nuke your country and the ai bubble will pop anytimeYou? a faggot (low-iq fentanyl huffer libtard)Me? pic related (a chad that says the truth)
when do we get to the part where AI does all the shitty jobs and humans can be creative?
This could have been avoided if the AI introduction was gradual. But the 2030 mass extinction event is coming and they are speedrunning techno medieval ages.
