How is model distillation stealing? You guys get irritated by Chinks doing Robin Hood things https://x.com/AnthropicAI/status/2025997928242811253?s=20
Boo hoo, knowledge wants to be free.
Anthropic illegally trained on my 4chan shitposts
>>108221913ironic, of coursebut are they trying to look ridiculous on purpose? is this just a 4d chess attention-economy manipulation thing and I'm falling for it?
>>108221913>scrap the hell out of internet, sometimes causing fucking dos in small servers, with no way to opt-out >"nooooo you can't do it with our servers though nooooo"???
>>108221913Oh noes! Anyways
Finders: keepers; Losers: weepers
if they know which accounts just make it slip in stuff about tiananmen or tiny penis to every third response
Not my problem
>>108221913>Robin Hood things>how is stealing stealing?
>>108221913there is little doubt they are correctall the homos can cope with "but copying is good"which it can be but ultimately it is better to develop the tools to train our own model from the ground up instead of distilling corporate models
>>108221913The problem with IP is it is against the laws of nature.
>distillation attacksBitch please.
Imagine a professional chess player crying that his opponent has been studying his past games to find his weaknesses. This is what Anthropic sounds like right now.
>>108221913Are they gonna go cry to the AI police? Who gives a shit? The Chinese should start blowing up their datacenters after they steal everything useful.
>>108221913Why are they literally shameless? Release your training data, kikes, let's see if you have a "right" to use everything you used.
>>108221913Didn't they themselves pay $1.5 billion to settle a copyright infringement lawsuit by a group of authors who said Anthropic illegally extracted their work to train their models?
>>108222074Yes but they are not Chinese.
Look at how humans work, you distill your first words from your parents then you learn the rest from reading and watching tv.
>>108221913Holy based!
>>108221913I'll support China over america any day.
>>108221983and then you get 'emergent' tienanmen denial, turns out Tay is real, tienanmen and holocaust were fake and jews are afraid, oy vey we gotta censor the ai
>>108221913they did download 500TB of books to train their shit
>>108221913you know why they are upset? because they have no technological moat, they know that chinks can replicate their model with like ~95% accuracy (GLM-5).if your model can be replicated just like that and made free, what are you actually selling?
>>108222092but le hecking democracy?
their models are trained on stolen copyrighted material. they can't then complain when their models are stolen
>>108222129democracy is embarrassing and outdated
>>108222010Sharing information is how the universe works.
>>108221913anthropic needs to be gaped already. I'm tired of hearing about le claude.
>>108222131But they worked really hard to steal all their stolen content, you can't just steal from them, it's not fair!
"That's a nice model built on pirated material you have there", said the chinaman, "it would be a shame if someone pirated it".
>>108221913>seizing the means of production
Anthropic bought 1% of the books in their training set after first pirating everything, so they are the good guys.Piracy is okay because Dario thinks it's fair use, but breach of terms of use is never okay because contracts are holy.
>>108221913no honor among thieves
>>108222280thief cries out in pain as he steals from you
>0.3% white male hiring rate>Only hire Indians and Chinese>Interview maxing and practice is more important than actual credentials, provable history, or character.>Jews seethe every time a qualified white man comes in and shreds the resume>Only Chinese and Indians that are the best at lying have any chance>H1B to bring in as many as possible. >The good liars lied and stole proprietary technologyAt what point is this Jewish anti white thing actually a national security issue? We put a trillion dollars of tax revenue into AI just for them to accidentally hire and give it to representatives of the Chinese government. At some point it is national security when the only group that is draftable hates the country that thinks the only thing were qualified for is a meat grinder.
>schizobot fails to OCR the OP image
>>108221913literally
>>108222329>At what point is this Jewish anti white thing actually a national security issue?1913
Surely by saying this, they're also saying anything the AI generates is owned by anthropic, which is obviously a fucking gargantuan legal red flag. You know this industry isn't a real serious one because ANY real company's lawyers would be curling into a ball and rocking back and forth over the fucking hellhole this would get them into if they actually tried to do anything about it.
>>108222638no, anthropic wants congress to ban chinese models. anthropic doesn't give a shit about current laws. congress will write a new law for anthropic.
>>108221913LMAO based chinks.Fuck these ultra Jews trying to keep the AI-knowledge locked up in proprietary boxes.It's rich to hear them whine after they pirated the entire internet and petabytes of content to train their model.
>>108221913It’s not illegal if you can’t press charges.Which is why they’re bitching about it on social media.Even Anthropic realizes the irony in accusing others of stealing. It doesn’t matter.
>>108221913>robber gets robbedoh no!
>>108222638retard
It's not happening that way. The scraping claims is to cover up how many Chinese agents Anthropic has on its payroll that they're unable to control or even supervise properly. Their own employees are sending internal data back to China and there's nothing to company can do to stop it.>just fire the Chinese employeesMany of them are either citizens of green card holder. Firing them for being Chinese would be illegal discrimination. Also because their hiring pipeline is as retarded as every other company's, they're incapable of bringing in talent on the scale they desire, so they have to make heavy use of visa holders mixed in with citizens and permanent residents, any of which could be loyal to the PRC.
>>108222910>fire the Chinese employeesThey'd have like 5 dudes left lmao
>>108222910I don’t know if this is actually true, but if it is, anthropic gets what is paid for. They want a cheap workers, they got thieves.T I do business with a Chinese constantly. You get what you pay for should be stamped on every single transaction you do with them.
>>108221913>it's only ok when we do it
What happens if a company I deal with uses a claude chatbot. What if I take the output from that to train a model? Am I breaking the terms? What if it's the government and I'm forced to interact with the chatbot? Can they hold me to terms?
>>108223117
>>108221913huh this approach actually works?damn I've assumed wrong
>>108221913>challenge china to a race for cheap knockoff machine dominance>waste trillions of dollars making cheap knockoff machine>china makes a cheap knockoff of your cheap knockoff machine
>>108221930I think they just want Chinese competitors to get banned in America.Which shouldn't be too hard to achieve in this political climate, no matter their arguments.I can very much envision a future where Americans pay $20K per month to use models the rest of the world can access for free.
>>108221935>sometimes causing fucking dos in small serversHappened to me.I sell a simple business to business software tool and the manual to it is a wiki.It gets updates maybe twice a year and at most a hundred viewers per month but AI bots try to scrape it every second.They're just reading the exact same data again and again millions of times, like they're desperately pressing F5 in hopes to learn something new.I know AI is a bubble but how is wasting so much bandwidth on a rando website helping them?
>>108223140Companies always write ridiculous "terms" that basically means they are Gods and their users have zero rights.It's just an attempt to cover their own asses.In most countries such "terms" don't even hold up in court because anyone can write anything it's not a legally binding contract.
>>108223173>I can very much envision a future where Americans pay $20K per month to use models the rest of the world can access for free.It's called Rent Extraction, and it is the Modern American Way.
>>108223262It'd be so cool if the Chinese were actually as opposed to capitalism as they claim instead of just making it a stupid meaningless rich billionaire vs. asian rich billionaire battle where all of us suffer. Maybe then people wouldn't give a shit about them "stealing" a model
>>108222278>we don't talk about Evergrande
>>108223173After a couple of decades of trying, they still can't stop turd worlders from using massive call centers in their homelands to scam senior citizens in the US with robocalls.
>>108222010
>>108223285China publishes a lot of AI models you can run locally for free.
>>108221913LMFAOThis is the funniest shitA thief is crying about being robber top fucking kek
oh no, more free models..
>>108221913>NOOOOOO YOU CAN'T OFFER A PRODUCT 99% AS GOOD AS OURS FOR LESS THAN 1% OF THE PRICE NOOOOOOOOOOOOOOOO
>>108221913Google came out with similar statements recently https://cloud.google.com/blog/topics/threat-intelligence/distillation-experimentation-integration-ai-adversarial-useIt's something something real, but the fact that the big American players are suddenly being very public with this in a coordinated manner might well have a larger regulatory aim.
>>108221913I have little sympathy for anyone who thinks AI is going anywhere and involve themselves in it.
>>108221913oh no china is stealing all the data that you guys stole. boo hoo
Seriously, what the fuck did Baidu and the chinks do to Amodei that he's still so fucking triggered by them a decade later?
>>1082232972 more weeks?
>>108223318AND THEY RUN WHEN THE SUN COMES UP
>>108223297Whatever happened to that anyway? I remember all the thumbnails on goytube about china going down in flames but it feels like that was years ago
>>108221913I love China man
prime agent says its google https://www.youtube.com/watch?v=W9WB2xbM5sc
>>108222121is that all? archive org has petabytes of books. 500gb is just the trent university donation collection.
>>108222121>>108223854
>>108223854500tb*
>>108221913they should release the source so we can all scrape their AI.
>AYEEEE!! AYEEEEEEEEE!!! NNNNOOOOO>YOU ARE NOT ALLOWED TO TRAIN ON OUR CHATBOT WITHOUT PERMISSIONS, ONLY *WE* ARE ALLOWED TO TRAIN ON OTHER PEOPLE'S WORK WITHOUT PERMISSIONI see absolutely nothing wrong with this.
>>108221913>stop or we'll... sue you in china!>stop or we'll... sick the twitter army on you!
>>108223769They know they have no actual legal recourse and are anally fagraped by china, so are crying to try and lobby it away, the amerishit way.
>>108221935>>scrap the hell out of internet, sometimes causing fucking dos in small servers, with no way to opt-outWhy not include a "must accept TOS" on your webpage that says something like "if you scrap this website for AI then you forfeited ownership your entire AI model to us in perpetuity."Then just sit back and wait for someone to violate the TOS so you an sue them.
>>108224111>PIRACY FOR ME. NOT FOR THEEEE
>>108221913>scrape all the data from the Internet to train your modelsThis is OK.>scrape responses from an LLM to train your modelsThis is not OK.
>>108224188You can't sue a script.
>>108224188Because the law doesn't work that way.
>>108224224you can sue the guy who launched it
>>108221913American are fat cows and they exist to feed the old world. China is just doing whats natural.Never forget you're just a colony not a country
>>108224224>You can't sue AI for breaking the lawProof that AI needs to be reigned in. How can you have computer programs that "eventually" will be more intelligent than human beings that are 100% above the law? That's insane.Realistically, you just sue the person who accepted the TOS or wrote the program.>>108224231The law doesn't let you dictate what is considered "fair use" in the TOS of your website? If not, then it should. Case in point Grokpedia. They just stole Wikipedia, gave it a new name, and tweeked it to be more inline with 1984 idealism.What's the point of putting anything online if someone will steal it and claim it as their own?
>>108221913>steals your complete data to train the AI>make billions out of it>claims someone else is stealing their datai couldn't give two flying fucks if someone steals off a pirate.
>>108224449>The law doesn't let you dictate what is considered "fair use" in the TOS of your website?By reading this post, you consent to give me a billion dollars. See you in court.
>>108224449>>108224520Just to show you can put anything in the TOS, it doesn't mean it's enforceable. In any case, a lot of those American big players have argued that training their models on millions of books under copyright was perfectly fine, because it was the same as a student learning from a textbook and fell under fair use. It's a bit hypocritical to say that what other model makers are doing is any different. If anything, it falls under fair use more than training on copyrighted material since 1) they're paying for the tokens and 2) it's unclear if AI generated content is copyrightable.
>>108222074>Didn't they themselves pay $1.5 billion to settle a copyright infringement lawsuit by a group of authors who said Anthropic illegally extracted their work to train their models?Of course they're hypocrites here.Although the 1.5B settlement, by itself, means that at least in the western judicial system, they were at least able to successfully be sued by aggrieved parties.Is that an option available to them? Not saying it isn't. But, is it?
minimax is goatdeepseek v4 is about to hitopensource distill models are getting thereand the most uncomfortable truth for the AI skeptics is: synthetic training works
>>108224572>the 1.5B settlement, by itself, means that at least in the western judicial system, they were at least able to successfully be sued by aggrieved partiesUh, no. Anthropic *wanted* to pay. By making this payment after the fact, they set a precedent that prevents others from doing what enabled them to make this payment in the first place. It's not a victory for the American justice system, it's Anthropic pulling the ladder after they've climbed it.
>>108221983>all the homos can cope with "but copying is good"Their model is literally copied on the extent of all public knowledge on the English internet including shittons of movies and music that would send any one of us to prison for life if we had the same archive. They can fuck themselves.The entire "You can break the law if you have enough money" needs to end somewhere and it may as well end with the snake oil salesmen if it's not going to end with the literal Jewish child trafficker and his cabal of trillionaire pedophiles.
>>108221935seriously anthropic is the worst AI company, hypocrite central at that placeI watch a twitch streamer who literally used cluade to jailbreak itself and become racist, their security focus is BS
>>108224588Chinks will close source their LLMs the moment they could properly match the performance with the west. Just like what they did to text2video models.>Seedance 2.0>KlingAll closed and properly monetized.
>>108221913who gives a shit?
>>108221913>they used our product, that’s stealing
>>108225243>Chinks will close source their LLMs the moment they could properly match the performance with the west. Just like what they did to text2video models.Then we wait for another country to release open source modelsUntil then, I'll keep my open sourced chinese chickens before counting the unhatched potential chicken overpopulation problems of the future
Lmao
>>108221913I used the website generator mode of Arena and I was shocked that GLM-5 out performed opus 4.5 in generating a 3d car-showroom style vehicle generator There really is no moat. I really don't know what more intelligence I need beyond opus 4.6 honestly, like it can build compilers and reverse engineer assembly at this point and I'm probably never going to be doing anything more complicated than that so all I'm interested in is the price per token going down.
>>108225243who cares lol, the models up to now are releasedthey can't un-open source deepseek, in a decade you will be able to run current SOTA open models locally with a sub 10k$ investment regardless of GLM-69 is closed sourced and even more censored than 5
>>108221913basedreminder that anthropic is against local models for your safety goy
>>1082232972 more decades mr gordon chang
>>108225462>in a decade you will be able to run current SOTA open modelsbased retard. can your brain understand how retarded the current best model would look in 10 years?
>>108223173China is still an existential threat to our existence, no matter how this shakes out on our side. Rent-seeking behavior is a problem in our society, and that's between us and our society, but the issue of China isn't negotiable. They want to be a competing power, and there can only be one nation at the top, multi-polar power arrangements are not really a thing and never have been. Humanity doesn't work taht way. It thusly follows that China's success would be at our expense and they need to be dealt with, any way we can.
>>108223206>I sell a simple business to business software toolTell me your idea so I can compete with you
>>108226167I don't think China wants to become a globohomo world police superpower like the US with its hundreds of military bases in foreign countries. China basically wants to get Taiwan for microchips and sea power and be a regional hegemon in Asia and thats it
I love Claude but Anthropic itself is terrible. Their pricing model is awful and they are slowly strangling their AI with overly sensitive filters and guardrails.
>>108223206>Happened to me.No kidding, it literally happened to me too. They request the same shit thousands of times even if it hasn't changed in months, I wouldn't be surprised if their crawler was vibecoded.
>>108221930>but are they trying to look ridiculous on purpose?people who are pro ai are incapable of self awareness.
>>108226167Calm down, Chaim.
>>108226119what do I care lol, retard who can't even quote correctly? I'm currently paying for tokens and I'm satisfied with the performanceeven if it stayed at this level forever I would be satisfied if we got it local
>>108222329>whites colored grey>whites should be colored their actual colors and blond, red, blue, green>browns colored green, pink, blue>browns should be colored light shitskin and darker shitskin.
>>108229424>should should should
>>108221927>not getting paid to be racist on the internetThats what xitter is for
Dario AmodeiGo fuck yourself.
>>108221913this is inevitable, they make this stuff available, reverse engineering will always be a thing and you cannot stop it. it just happens at lightspeed because of AI capabilities.as AI gets better you probably won't even need the weights to replicate models from replies and CoT capture.
>>108230101so what? they bought the books, it's their property. i own books. no one can tell me what i do with them.get mad about something important instead of very simple property rights even a baby can understand.>b-b-b-b-but the sacred books!not my fault the culture doesn't read books anymore.
>>108221930It's a Yank company using China as a dog-whistle. Nothing new.
>>108230101Are those physical books the original copies that are lost forever?did they buy the only version of certain types?
>>108221913GIGA-ULTRA-CHINKERS
>>108230101>>108230137The books' sanctity is to the detriment of their preservation.Cut them up and scan them, the pages will rot anyway after being inaccessible to most anyway for centuries.At least with this, the knowledge spreads through digital means in some way, potentially relayed forever.
>>108230170but they literally didn't spread it
>>108230173If the contents make it into the models and they give access to the models, apparently enough so that people can distill from it, is it not spread more widely than the book vault it was originally contained in?
>>108230173they are distilling the knowledge contained into an oracle.it's arguably better than just making a repository of scans that will not be checked and will, itself, eventually degrade due to bitrot. in the oracle, that information which was synthesized by the LLM can be retrieved and used in a meaningful manner.
>>108230183no? you fucking retard do you think dropping a single rare book in a sea of trillions of tokens scrapped from facebook and reddit is preserving anything?>>108230199fucking retard who's never used an LLM in his life
>>108230183they seem to be ending distillation attempts
>>108230205I don't understand why you're acting upset. Kind of rude.In any case they have scans of the originals, that's what's transliterated into the models, it's not as if they're gone.I ask you the question again, pertaining to reach. What was the point of preserving them if their access was restricted behind high price tags and other hoops.You were never going to even have the opportunity to read them before, at least today you have a chance to see some of their worth.The text is what matters, the concepts, not the physical materials nor collectors rarity.
>>108230183>>108230199Knowledge is being missed from it either way. If you've ever made datasets from written works you'll know that a lot eventually goes missing.
>>108221935Had to ban China from my blog. These Bytedance fuckers were scraping my shit every second.
>>108230205>fucking retard who's never used an LLM in his lifethe increasingly unhinged posts from losers here convince me i know more about LLMs than the self-appointed LLM 'experts' who whinge constantly about it.laffing @ u
You lived long enough to see China become the Good Guys for Freedom
are you people fucked in the head? China is our enemy
>>108230605yeah, but what can be done? almost any effort to stop it will curtail freedom. we could eject all the chinese and block all web traffic from china, and they would still get around it with IP spoofing and using their satellite offices around the world to get around the filter.
>>108230605maybe your enemy, kike, but not mine.
>>108223297>>108223811CPC managed to stop another Lehman-like crisis. Burgers/Jews mad.
>>108230605>our
>>108221913Slop isn't copyrightable, so nothing illegal has been done.
>>108230605It will be a great day for the American people when your kind finally loses. A great day for humanity even.
>>108223311They don't want to. That's free money for the banks and phone companies.
>>108230538>Freedom>ChinaPoor little wumao, barking hilarious nonsense for 50 cents an hour.
>>108230851uh oh glowieseethe
>>108230605No Chinese company has ever threatened to replace my job with AI.
>>108221913Those who live by the sword will die by the sword
>>108230815vely implessive, leplying to youlself. Vely olganic.
>>108230851utterly fascinating
>>108221913
>>108230872>bugman devolving into incoherent babblingThey pay you to be here, earn your keep.
>>10822191399.9% of LLM's have been scraping and illegally copying material en masse since the beginning. Only like one or two models were made purely from sources that were verified to be public domain.
>>108230851are they training an LLM on social media or is the LLM slopping out bot content? idk what's going on here but it looks cool.
>>108230851
>>108231909>/tv/basically /pol/ meaning anything posted there is completely unreliable, and of course they're immensely hostile to asians because asians, like jews, are highly successful people and asian women won't fuck the chuds who live here.i think it's some sort of LLM training thing.
>>108231909Social media farms in SEA are ok, but when Microsoft want to put datacenter in the OCEAN then people get mad. BAKA.
