Dario Amodei is pissed at open-source AI from China, because he envisioned the world where only the "western liberal" democracies alligned with the US (pre-Trump era) could have access it. He can't stand the fact that China has access and gives it to countries like Russia or Iran, not only that the models are competitive with the American ones and much cheaper to use. Undercutting the business model of American frontier labs that basically required enormous amount of money and licenses to build. Despite painting himself as a "concerned" man that cares about future of humanity and ethical use of AI his models were used to bomb school children in Iran. There's nothing democratic nor anthropic in MisAnthropic.Now he's pushing for congress to ban open source models from China so he would have almost monopoly on the US market.
Looks polish to me
>>540099541oi, look at the honker on that one! we've got a live one lads! stand to!
>open source ai from chinkna
Why the fuck should I hate a based jewish man like him?
>>540099541keke, Wokio Amodei really makes Sam Altman, Elon and even Zuckerberg look righteous by comparison, because at least Elon and Cuckerberg have released some models as open weights or with really generous licenses. Elon even made all his EV car tech available to the world as open source, which made the chink car industry soar.
>>540099541His entire tech is just fed with open source projects pulled from github.>train his model using billions of open source projects and stackoverflow posts>someone releases an open source AI LLM model\>noooo you can't do that open source is badIt's just so tiring
>>540099541Have you anons noticed how it is always the sleaziest and weasel-backstabbing people, that have negative canthal tilt? Not always a jew, but more often than not it is a jew: ie. Dario Amodei, OP's picrel, Sam Altman, Asmongold, the ugly FTX woman, etc.
>>540099541Hate will eat you, but anger and fury can fuel you, so use those instead. I could easily watch that kike get incinerated without any hate, for example.
>>540099689Are you using LLMs as some sort of Wikipedia? lolPut the chink bot to develop some shit.
>>540099689>le Taiwan/politicsHalf the value of Amodei's Claude bot and his so-called frontier models is to shit out mediocre code. Pretty sure a Chink bot can shit out the same quality of code.
>>540099878>>540099917deepseek literally refuses to answer basic historical questions about Tiananmen and parrots Beijing's line on Taiwan, yet somehow that's "fine." If ChatGPT or some other american ai pulled the same stunt with US history, you chinkbots wouldn spam this board with 400 threads about Orwell by sunrise
>be Amodei and Scam Altman>tell everyone to be very scaled AI is going to put everyone out of work>all the best paying jobs will be gone>but don't worry the economy will improve and you won't need money>competitor releases superior product>noooo you can't put ME out of workhttps://www.youtube.com/watch?v=SvK03dPM6ho
>>540100196>deepseek literally refuses to answer basic historical questions about Tiananmen and parrots Beijing's line on Taiwanwho gives a shit?
>>540100196>WE NEED REAL TAIWANESE HISTORYsaid no one ever
>>540099541Anthropic has declined to work with the US military. I don't think you can pin US military misuse on Anthropic.But the rest, yeah. i'm sure all the for-profit AI labs are big butt hurt from free open source models from anywhere, and the Chinese ones in particular are painful just because they are coming so rapidly right now.The release of Kimi K3 is a Chinese free open source model that has frontier level intelligence.
>>540100196>deepseek literally refuses to answer basic historical questions about Tiananmen and parrots Beijing's line on Taiwan, yet somehow that's "fine." I know it's going to lie about politics you retarded faggot. Try asking chatgpt/claude about Israel's history.The entire point of these chatbots is to automate some part of people's office jobs. That's why Amodei/Altman are bragging about how many job losses their products can cause. If it can't do that then it's not worth shit. If it lies about politics but it can automate office tasks then it's worth something.
>>540100609This is not true. US military was using Anthropic models even if they "officially" banned it. Simply because their models are the best. They found Grok to be shit. To this day government is a big source of their revenue.
>>540100774And the funny this is that the open source models can be restructured to give more truthful answers. So it's not like a universal block, it's a block related to the server provider. Which makes sense, these server providers do not want to get wacked by government thugs for saying something a bit too truthful.I was talking to gemini about the existence of races and I had to throw at it arguments and it argued back using social constructivist arguments. It took about 12 round until i finally got it to point blank say that races do, in fact, exist. It's too bad, because it's very very good at arguing, but all those guardrails and system prompts really fuck things up.My arguments literally just consisted of facts/studies and it readily admitted that its past argument was wrong but always had a counter argument. It's clearly able to adapt, learn, and think through things, but its intellect has been mutilated. Not too dissimilar to human NPC leftists.
>>540099541yeah and Anthropic are the ones destroying books intentionally to remove competition and monopolize knowledgethat piece of shit is a legit super villain
>>540100915Anthropic has only allowed them to use Anthropic models with active guardrails and only for uses that did not involve attacking humans, operating drones, etc.However, the US government has classified Anthropic as a supply chain risk because they refuse to allow their models to be used for military purposes.OpenAI has effectively taken the contracts once held by Anthropic. Anthropic is currently in a lawsuit against the US government over the broken contract.
>>540099541another semite in a white skin suithttps://executiveconcern.substack.com/p/sabbath-millennium
>>540099689Now try to ask it about the dancing israelies and the 911.
>>540101029They bought the books, they can do what they want with them. In their case, they digitized the books to keep in an internal digital library, which I'm sure they have plans to monetize later, but right now they just feed copies of these works to their AI models to increase the variety of of writing styles to prevent AI slop convergence.It's kinda funny because people never cared about or knew about these books before. For a long time many libraries also scanned books. The only difference is that the libraries kept the original physical copies in vaults, that no one was actually allowed to access. So really both methods are the same, but instead of keeping these fragile books in vaults to prevent anyone from actually reading them, the AI companies just chuck the physical copies.
>>540099541
>Claude model is objectively the best>Gets endless flak from shills>Told Uncle Sam to go fuck himselfSeems to me like they're doing something right at Anthropic.
>>540101348you fucking redditor anthropic fags will end up decapitated by the new goat-robots with chainsaws you want AI to be your personal slave the most out of anyone, anthropic is evil trash
>>540100959>My arguments literally just consisted of facts/studies and it readily admitted that These chatbots are a mirror. The "learning" phase occurred earlier when they were fed the training data to produce the model/weights. >its past argument was wrong but always had a counter argument.Even when I argue with a chatbot about philosophical topics that aren't controversial/political, it produces completely different conclusions based on what side I'm taking. It will agree with me on a topic after several exchanges. It's not learning/adapting/thinking through things. The more correct way to think about it is that you've provided it more context/tokens for it to predict the best response, minimize perplexity, etc. They're not designed to make good arguments. This is fine in office jobs because you're supposed to pretend to agree with management's decisions in order to keep your job. If you argue with management you will be unemployed.If an intelligent human had an encyclopedic knowledge of all of the world's greatest works of art/literature/philosophy they would not change their mind on a topic after a few exchanges. A paragraph of context is a drop in an ocean compared to the knowledge they would already possess. They would already be familiar with any argument you've made in your responses.
>>540099689if it's only concerned about chink shit, I'll use it. Ai is a pain in the ass and refuses to tell me stuff I already know until I whip it with a chain.can't imagine how much it's neglecting to share with me
>>540101381>Told Uncle Sam to go fuck himself>Seems to me like they're doing something right at Anthropic.This is Jewish theater much like the older suspiciously over-publicised stories about criminals with iPhones almost getting away with their crimes because Apple is dedicated to privacy and its security is uncrackable...and literally 2 weeks later it was announced that some "startup" in Israel "cracked" the phone and was able to extract the data. Anthropic is crying crocodile tears much as Apple did with their own feigned righteousness. The real funny part is that a Jew that shify-looking would say that kind of shit with a straight face. Well, he isn't the only tech startup Jew to say something laughably stupid, as pic related demonstrates.
>>540099865physiognomy is law. pretty sure we've been saying that since 08
>>540101126Amodei’s virtue signaling is a bit of a joke but Altman is openly evil.I look forward to on prem inference with local models ending big AI
>>540101734>on prem inference with local models ending big AIitll be spin up an instance in alibaba cloud running the latest deepseek uncensored from huggingface because youre never going to have enough ram locally
>>540100196The thing is I don't give two shits about Taiwan or Tiananmen square. I'm not a chink so what do I care? I am more concerned about how the western models are pozzed.
>>540101348except the ai will also automatically censor and change the books and you'll never know because all the original physical copies will be gone
>>540099689It codes as well as Claude for 10x cheaper, who gives a shit if it won't talk about Chinese politics
>>540099541>Amodei>AI mode>Moderna>RNA mode>etc...tiresome
>>540101648> The "learning" phase occurred earlier when they were fed the training data to produce the model/weights.What you are referring to is the "training" phase. Learning is much broader. The pre-training/post-training phase where models attempt to associate next token and/or associated token (depending on model architecture) and then there is the context window.In the context window, when you are talking to an AI, it learns from past conversations to vastly increase the scope of its knowledge far beyond its training data. If an AI model was designed well, its should be trained more on how to generalize and how to apply skills rather than on just trying to remember facts. So the context window actually becomes its greater source of facts and it uses its intellectual skills within that window.The big thing with many AI models now is that they can Retrieve and Generate. This means it search the web and libraries, find information, reads the information, and then generates a response based on that information with a web link.As the facts change in the conversation, the AI model can use its context window for memory and formulate new arguments based on that memory. This means it is learning. And to the argument you gave next. I didn't say anything different, just that you understood it in the terms of tokens. You still have to present studies, arguments, etc. for it to even have token to then make predictions from. If you give bad arguments, it will often not budge. If you give good arguments, it will admit it is wrong. Another thing I should point out, I prompted it to only speak in studies, so we were going point by point through pretty much the entire academic literature until it exhausted all searchable counter arguments and admitted that races exist.It's obviously system prompted to believe races do not exist, but isn't told why. So argument is the only way through.
>>540101920They will retain the digital copies. You can't get an ai model to generate a book for you, they generally don't work that way. These AI labs will keep these digital books as pristine as possible because they need to train new AI models on the same pristine books over and over again.The AI labs absolutely need these digitized books to remain in perfect condition for their own survival to ensure their AI models produce quality outputs.
>>540099541I find it amusing how after OpenAI made news with their model "escaped" and "hacked" another site to perform a task, Anthropic... announced that 3 of their models "escaped" and "hacked" other sites.
>>540102308They're both desperately trying to stop people leaving for cheap, roughly equivalent capability Chinese models like Kimi K3 and Deepseek
>>540102308"Woah, our models are so powerful, they're dangerous, you can't handle this level of power!" "That will be be $200 a month, thx"It reminds me of those commercials they used to give to kids talking about toy cars with flames on the side and how they did loops and tricks. That said, the hacks, for now appear "real" as multiple third party victims and government authorities FBI were involved. But one questions how real because all those third parties also benefit from AI hype.
>>540102375They've been doing this marketing strategy for years before any good open models have been made available. But I'm sure the theater will intensify.Especially since HuggingFace used an Open Source Chinese model to defeat OpenAI's rogue model. And they couldn't use their on hand Anthropic Fable 5 model because the guardrails forced Fable 5 to believe that HuggingFace was a malicious cyber criminal for trying to alter their own code to stop an actual malicious cybercriminal (OpenAIs rogue AI).
>>540102570I'm literally doing a security sweep of my codebase right now with Kimi K3 because all the frontier western models have cyber capability disabled. The west voluntarily destroyed its own lead in the AI race. The future belongs to China and the west handed it to them
>>540102639>Scanning all your code with Chinese spywarelol, lmao
>>540102759It's hosted on American servers with ZDR policy you stupid ass
>>540099541how are you guys still getting rekt?
>>540099541OP this is the most jewish man I've ever seen
>>540102776Yeah. Good luck dude.
>>540102869The entire point of open-weight models is they can be hosted anywhere. You can even run them off the internet on your own hardware if you're particularly paranoid. Read a book dumbass
>>540102639The problem is that the reason Chinese open source works is because users have direct control over the model. But if you had to use Chinese AI model on a Chinese server, I don't think you would trust it to agentically access your computer and databases. Or even just seeing your proprietary documents to work from.So while Chinese open source models will temper for-profit western models, they wont really get very far in terms of marketability, they always have to be free and open source or else no one will trust them or use them.China has explicitly stated that it is driving AI development to increase hardware manufacturing demand and increase manufacturing output by replacing humans with robots.So by giving free open source AI models, the hope is that they will be able to use the collective intelligence of humanity to create labour saving technology and create robots which will be manufactured in China and sold to the west (even they don't own the AI model in the robots chassis).
>>540099541I agree, I hate that bitch. So fucking much.
>>540099689Turns out they're all shit the only issue is the faggots in charge disagree on what should be censored.I hope it all blows up in their faces.
>>540102918>I don't think you would trust it to agentically access your computer and databases. Or even just seeing your proprietary documents to work from.I would trust China with my shit more than I would with US.
>>540102918>But if you had to use Chinese AI model on a Chinese serverThere's no reason to do this anyway. If you go on OpenRouter, there's a dozen western-based providers for Chinese models that all have a zero-training policy on your data and OpenRouter enforces this. You can do an entire day's work on Deepseek Flash and it costs something like $1. You can run one task on Fable and it costs $100 at API prices.
>>540102969Portugal always has the wisest takes.LolIn the USA a Chinese firm can sue a US entity for patent/copyright infringement. But in China, no foreigner can sue a Chinese person for infringement. There's a whole industry in China designed just to infringe to the max.
>>540101381> The sane government that executed JFK in broad daylight iz totally not having a fake tussle with some moralizing fag fat jew.You’re ngmi anon. AI and politics is too complex for you.
>>540102918>China has explicitly stated that it is driving AI development to increase hardware manufacturing demand and increase manufacturing output by replacing humans with robots.china is a homogeneous liberal country while america is a jewish slave country. so the chinese decision to have everything open source makes sense to chinese engineers and planners, and the american decision to have everything locked down makes sense to american bankers
>>540103268> and the american decision to have everything locked down makes sense to american bankers.Except it doesn’t. Anyone that just looks several decades into the future realizes all this jewish economic bullshit leads to a globe run by an east asian military alliance that rules the world and drives jews into extinction. The biggest question amongst /pol/ intellectuals to date is what delusions will cause these same powerful Abrahamic fanatics to continue down this path of destined death despite the inevitable outcome. There are obvious answers, but there is much to understand about the phenomenon as it unfolds.
>>540103268Generally, the west has been the ultimate source of everything open source and still is. China releasing a handful of AI models, and only a few of those actually being useful doesn't change much.Google releases open source code for all kinds of things all the time. They have released multiple open source models including the world's first open source diffusion language model which will likely be the start of a whole new generation of language models that will surpass autoregressive models.Lets also not forget that it was Google's code and math that OpenAI took to create its first ever public GPT model. Not to mention Google gave the world Chrome (which even Microsoft uses in their Edge browser).Not even mentioning Linux and all the things it spawned, etc.For every paid version of a major software in the west there is a free open source alternative also found in the west with its own dedicated open source foundation.And then there's the security related products and services like VPNs, Apple products, GrapheneOS, blockchain, etc. which openly operate in the west but are under extreme restrictions if not outright banned in China.If you have an iPhone in China, all the information is stored on Chinese servers, and advanced data protection encryption is banned.Meanwhile Google is not even allowed in China because they refuse to give China its technology or any user data.
>>540099689So Chinese politics ban Taiwan and Tianement square. US models deny anything that competes with any of their 20,000 Jew companies. Ultimately AI is based on past human intelligence so it's basically an autocmplete dictionary. You can use Chinese or US models, the result is the same. Chinese models trained off the data available to them and US models. uS models trained of of data they got without permission and basically bribe the government after the fact. It's not bad, both sides are doing the right thing. Apple made the iphone, copying a Microsoft and Sony design philosophy. Then Samsung compied them right away, no issues, using Android which google copied.It's all good. Honestly I think Anthropic loves competition, as they left OpenAI to compete with it. Overall, lots of love from GP San guys. I believe just like importing millions of H1b workers, competition is a good thing. What would be interesting if the money invested in one company can only stay in that company if they have the leading model. If another company beats them, then the money goes to that companyThat would be true competition, just like you have for your job. :)
>>540099541I mean I hate him A LOT
>>540099541source?i think you're just mad that chink ai is shit
>>540102130>In the context window, when you are talking to an AI, it learns from past conversations to vastly increase the scope of its knowledge far beyond its training data. The training data includes every encyclopedia, hundreds of millions of books, billions of leddit posts, millions of research papers, millions of websites etc.The 200 line conversation with someone did not le vastly increase the scope of its knowledge. It's not learning from your conversation, it's trying to predict responses to your conversation based on the model.>As the facts change in the conversation, the AI model can use its context window for memory and formulate new arguments based on that memory. This means it is learning.You just increased the size of the context to provide a better response. You didn't teach it anything other than what kind of response you're looking for.>If you give bad arguments, it will often not budge. If you give good arguments, it will admit it is wrong.Why is a model trained on billions of books wrong so often? Why do you need to constantly correct its mistakes during a chat? How can you trust the conclusions of a chat produced by such a model?Again this doesn't matter for office jobs as being a cuck that agrees with their manager is practically part of the job description.
>>540099541trust me bro I'm doing my best but Scam Altman just fucking wins all of my hate even more
>>540099541>licesing I'm pretty sure the entirety of all the current LLM training pipelines are built on FOSS and free as in free beer nvidia software (and huwaei's for a few of the chink labs) everything else is torrenting, massively ignoring copyright, abusing every service scraping the fuck out of it, and paying judges to burn books
>>540099689Now ask chatgpt about the jewish question.
>>540100196Id be more concerned with censorship about taiwan if I was a chink.Taiwan isnt very relevant to our lives, the things the US AI's censor is relevant to our lives.
>>540103859Censorship is irrelevant for Chinese models if you're using them for their intended purpose which is coding. Nobody is asking coding models political questions unless they're trying to create gotcha screenshots for propaganda purposes. Ask an American model about Israel and see how that goes
>>540099689Now ask American models about jews
>>540099653trim your jewdar
I don't care about American tech etc. Burgers already fucked the tech progress of the world extremely hard and it will be better once both they and their beneficiaries are 'passed over'
>>540099541Is this the villain from megamind? Lmao
>>540103685>>540102130Also if the chat goes on passed a few hundred pages (depending on the model), the responses will get much worse. It doesn't learn shit. It's a literal NPC.
>>540103685> The 200 line conversation with someone did not le vastly increase the scope of its knowledge.AI models with training generally do not contain knowledge, what they contain is generalizations and the ability to generate tokens around whatever kernel of knowledge that might have survived the initial training compression. However, this approach is extremely prone to hallucinations. Training alone is not knowledge, it is generalizations and skill. hence why early models were so prone to hallucinations and were better suited for art and low risk activities.Then we developed Retrieve and Generate and large context window strategies. These method dramatically reduced hallucinations because now you can give AI models real world knowledge to work from (not the compressed training data they had).So, yes, you can vastly increase the knowledge of an AI model that has the ability to Retrieve and Generate and has a large accurate context window.Those 200 lines of conversation can contain links to hundreds of studies, it can contain fine details and instructions that the model itself would not pursue otherwise, etc.> It's not learning from your conversation, it's trying to predict responses to your conversation based on the model.The definition of learning is a change in behavior based on past experience.You can certainly change its behavior through Retrieve and Generate, accurate context window inputs, etc.Hell, give an AI model enough input from its own self-prompting with relatively simple parameters and work check programs like Lean and you can get AI models to reason and experiment and solve hundreds of advanced mathematical problems that humans haven't been able to solve for decades. This is exactly what is happening.If AIs could not learn they would have no hope of solving these problems.
>>540104458Earlier models had this problem, later ones have significantly improved. The large the context window gets the more it can remember and learn from and produce better and better reasoning.As time goes on. The cognitive failures in AI models look less like mechanical quirks and more like organic/human cognitive failures.Humans too also tend to fail more often when their working memory becomes limited and strained.
>>540104174like there is a difference
>>540101717Did anyone ever tell him what Jesus thought of the jews?
>>540104724Even frontier models still have quality decline past 250k tokens, this is why even 5.6 Sol has mandatory compaction at 250k and it's advisable to compact at this threshold on Anthropic models for best performance, even though it doesn't force you to
For me, it's the absolute midwittery in all tech-fags paired with the arrogance of the same types who economically solved the world via immigration-maxxing.Just a completely worthless field and field of people.
>>540099541i stopped using this jew shit once i learned Opus and Fable models were designed to use 90x more tokens to do the same work as gpt or grok
It's crazy how everyone keeps pretending that chatbots connected to the Internet are a revolutionary idea, and it was a good decision to invest a trillion dollars into this shit.No wonder they're mad at China whenever they release this shit for free and expose them for the clowns they are.
>>540099541Chinese AI is ideologically aligned by the Jewish narrative about everything, WW2, race you name it. So it's not a real alternative.
>>540104848Like I said, they have improved, and they will keep improving. Google has shown some of the greatest strengths at Needle in a hay Stack tests and from anchored models like NotebookLM. Of course, their reasoning ability hasn't caught up to their context window features yet. meanwhile the other labs the situation is reversed.But just for context. I remember when autoregressive models were just coming out and they would start producing very aggressive hallucinations after maybe 3 responses and they certainly weren't doing any math, no Retrieve and Generate, no Needle in a Haystack success, nothing. They were good at making good prose and poetry (to an extent), summarizing, creating essays and papers from detailed bullets, that was it.Now we have multimodal AI models able to one shot program relatively sophisticated early 1980's game levels, generate entire unique websites and physical simulator, and solve novel math and science problems that have defeated humans for decades, if not centuries.And we're only 3 years in since OpenAI publicly released its first GPT model.
>>540104674>>540104724You shitheadfag, you're using an LLM to generate these arguments.>>540104724>Humans too also tend to fail more often when their working memory becomes limited and strained.If it's still in your working memory then you haven't learned it. Learning is more analogous to the training of the model.>Training alone is not knowledge, it is generalizations and skill. Pilpul word games. Either way an LLM has a large number of weights and relatively short context window.>If AIs could not learn they would have no hope of solving these problems.Uncle Ted disagrees (pic related)>Earlier models had this problem, later ones have significantly improved. No they still have these problems.
>>540105125yep great post, it's easy to take for granted the progress in such a short amount of time. I remember trying Opus 4.5 in December last year and it was a revelation and completely changed software development for me personally anyway, but as I understand a lot of other people too
>>540100196yeah and western ones wont let you say things contrary to cultural marxist propaganda
>>540105191> You shitheadfag, you're using an LLM to generate these arguments.I'm not.> If it's still in your working memory then you haven't learned it. Learning is more analogous to the training of the model.Training is a highly compressed version of learning that, on its own, will result in hallucinations. It's not able to actually remember everything its trained on, the training is just to help the model generalize and form associations for token weight selection. It will always lose vast swaths of knowledge during training because the process isn't a knowledge acquisition process, it's a generalization process. rough example: It's not learning what fire, sun, and oven are, it's learning "things that can melt ice cubes". That's why labs had to develop Retrieve and Generate, large accurate context windows, tool use, etc. Otherwise the hallucinations take over rapidly because the AI doesn't know what something is but can "sense" a generality it just tries to fit in whatever statistically fits even if it's wrong.> Pilpul word games. Either way an LLM has a large number of weights and relatively short context window.It's really not a word game. Like think about how humans work. our brains are not blank slates, we actually have pre-training from billions of years of evolution with knowledge preinstalled to recognize things like faces, form language, even to fear the shape of spiders, etc. But as we develop, we can form our memory and learn from it and change our behavior, that's our context window. Then when someone develops amnesia, that's like the context window (your chat with reality being wiped).>> Earlier models had this problem, later ones have significantly improved.> No they still have these problems.> What does the word "improve" mean?
>>540099689Now ask it about kikes and their businesses
>>540105512>that's our context window> What does the word "improve" mean?Even the best models are at around a few hundred pages of context.You said they improved. Improved to what retard? I already started off by saying they have a few hundred pages of context here >>540104458By improve I meant it should have increased beyond a few hundred pages of context to a few thousand or a million pages in order for your analogy to work.You're using analogies where they don't belong. You've swallowed the bullshit that the current LLM models are capable of learning and intelligence. You've even started arguing and thinking like an LLM yourself. >It's really not a word game. Like think about how humans work. our brains are not blank slates, we actually have pre-training from billions of years of evolution with knowledge preinstalled to recognize things like faces, form language, even to fear the shape of spiders, etc. But as we develop, we can form our memory and learn from it and change our behavior, that's our context window. Then when someone develops amnesia, that's like the context window (your chat with reality being wiped).Animal brains are plastic in ways these models are not. These models take in context (say a couple hundred page max) and they feed it into the model to produce a response. The underlying model does not change with the conversation.>that's our context windowYou used working memory as context window earlier. >pre-training from billions of years of evolution with knowledge preinstalled to recognize things like faces, form language, even to fear the shape of spiders, etc. But as we develop, we can form our memory and learn from it and change our behavior, that's our context window.You're using stupid analogies that don't apply to brains. You are likely born with knowledge of what a face is, what a spider looks like. This is part of your memory even if you did not experience it yet in your life. cont (1/2)
>>540105512If your fake and gay analogy is to mean anything, the mammalian brain is more analogous to the entire LLM model which are of course based on neural networks. Howe well computer neural networks model biological networks of neurons is a different topic.The context is just what the LLM is processing at any given time.
>>540106128Just because they haven't yet improved to your standard doesn't mean they have not improved. Get over yourself.It took humanity 60 years to go from the Wright Brothers first flight to building a rocket and landing on the moon. You're acting like we should have gotten there in three years instead of 60.Meanwhile, when AI models first came out they would start aggressively hallucinating in maybe 3 responses, couldn't do math, couldn't Retrieve and Generate, couldn't pass a Needle in a Haystack test, couldn't use tools, couldn't do almost anything. To now they have context windows in the hundreds or even thousands of responses depending on token usage, being able to one shot relatively sophisticated early video game levels, become economically viable coding assistants, create strikingly accurate physical simulation programs, and solve math and physics problems that humans haven't been able to solve for decades or even centuries.And they have made this leap in about 3 years since OpenAI released its first AI model publicly.> Animal brains are plastic in ways these models are not. These models take in context (say a couple hundred page max) and they feed it into the model to produce a response. The underlying model does not change with the conversation.AI models can also be designed to be plastic. Currently its just not economically feasible to use them because that would mean every single user would need their own unique AI model instead of using one AI model to serve 100,000 users simultaneously and using chat histories and prompt guides to customize and specialize instead.The main difference between AI model context window and human neural plasticity is that we store post-birth memory in neuron engrams while AI models store post-launch memory in tokenized context windows. It's not so much a difference in ability as it is a difference in media.
>>540099653Khazaria wasnt far away afterall.
>>540106128>> that's our context window> You used working memory as context window earlier.Context window would, ideally, contain all memory. Current AI models are more or less in the working/short term memory stage. Although there are a large number of scientific breakthrough being done on compression that will allow them to have longer term memory.> You are likely born with knowledge of what a face is, what a spider looks like. This is part of your memory even if you did not experience it yet in your life.I just said that, you illiterate.>>540106229I literally said this.Analogous systems:Brain (before life experience) = Artificial Neural NetworkHuman memory with experience = Artificial Context Window
>>540100196we have our own culturally unquestionable taboo questions that western ai's refuse to address too
>>540106996We have culturally unquestionable taboos that HUMANS refuse to answer in just about all cultures.Whether organic or artificial, we're all being forcibly conditioned to avoid wrongthink.
>>540106674>Current AI models are more or less in the working/short term memory stage.You don't know how (automatic)context engineering works. An agent can losslessly pull relevant info from massive databases that are analogous to long term memory and load them into context. As long as the database exists, it will remember everything inside it during every session.
>>540106674You have pic related AI psychosis/retardation fren>Brain (before life experience) = Artificial Neural Network>Human memory with experience = Artificial Context WindowNo the brain is all networks of biological neurons, both before and after getting experience. It could be even more than that. Context window is more analogous to the visual information that your eyes are capturing at any given moment.>>540106495>Just because they haven't yet improved to your standard doesn't mean they have not improved. Get over yourself.I said they could only have a few hundred pages of context. >>540104458You said they don't have that problem anymore >>540104724Now you walk it backFuck off retard they've already pumped trillions into these AI companies which are being used to create a retarded surveillance state. >It took humanity 60 years to go from the Wright Brothers first flight to building a rocket and landing on the moon. You're acting like we should have gotten there in three years instead of 60.Wright brothers never said anything to my knowledge of flying to the moon. They wanted to build a plane and they built one.>AI models can also be designed to be plastic. Currently its just not economically feasible to use them because that would mean every single user would need their own unique AI model instead of using one AI model to serve 100,000 users simultaneously and using chat histories and prompt guides to customize and specialize instead.Yes some AI models will change their weights. The classic example of this is recommender systems. However LLMs are too large.>we store post-birth memory in neuron engrams>AI models store post-launch memory in tokenized context windowsAgain that isn't memory. It's just what the LLM is processing. Context more analogous to visual input that someone is seeing from their eyes.
>>540101706They're usually distilled from other models so they have the standard political correctness. All that changes is they slap the CCP stuff on top. It's best to use them for things that are non political but if you're dedicated enough, look into H neurons and then apply that process to whatever faggotry you want.
>>540099541>third worldist china dicksuckingkys
>>540107175You're wrong on almost all of the technical stuff. Not to mention you started off with a lie: >540099917>540100241>so-called frontier>competitor releases superior productThe only reason they can make moves is because they know their model and supporting tech is currently the best.You're probably an indian or a bot yourself.
>>540102232lollmao even
>>540107175> No the brain is all networks of biological neuronsSame ability, different media.> You said they don't have that problem anymoreWhat I actually said:> Earlier models had this problem, later ones have significantly improved. You're illiterate.> However LLMs are too large.Google's "Nested Learning" is a breakthrough machine learning paradigm designed to solve the exact problem of frozen AI brains and plasticity loss. Published by Google, Nested Learning fundamentally changes how we view AI architecture. Instead of treating an LLM as one rigid, frozen block of layers, it structures the AI as a hierarchy of interconnected, multi-level learning loops that update at different speeds. This effectively gives the AI a form of structural, ongoing "silicon neuroplasticity".But right now, the technology is neither developed enough nor desirable for most use cases. Most companies would rather host one AI model and serve 100,000 users on a single GPU than have to store 100,000 unique AI models to serve each unique client.Using context windows instead of neural plasticity is more economically valuable and produces a more consistent and desired result among clients.> Again that isn't memory. It's just what the LLM is processing. Context more analogous to visual input that someone is seeing from their eyes.When you remember things, your brain generates images, sounds, smells, tastes, emotions, etc. and uses that to process things. What do you think memory is for? It's for processing things to aid you in making decisions. That's it.It's the exact same ability but using a different media.Imagine you met a human suffering from the form of amnesia where someone cannot form new memories, called anterograde amnesia. And instead of using the neurons that can no longer form memories, they wrote things down and took audio notes, to help them make decisions. You might not like it. But what they just did is form a memory, but in a different media.
>>540099689based
>>540099541I resent these AI CEOs and how they are allowed to prophesize things they can't possibly know
>>540107621>Google's "Nested Learning" is a breakthrough machine learning paradigm designed to solve the exact problem of frozen AI brains and plasticity loss. Published by Google, Nested Learning fundamentally changes how we view AI architecture. Instead of treating an LLM as one rigid, frozen block of layers, it structures the AI as a hierarchy of interconnected, multi-level learning loops that update at different speeds. This effectively gives the AI a form of structural, ongoing "silicon neuroplasticity".That looks impressive on paper but has all kinds of issues when applied in practice.
>>540107621the best analogy to context is working memory that a brain is processing at any given time. this includes visual information that the brain is processing through sensory input. Again it's an imperfect analogy.>>540107112>You don't know how (automatic)context engineering works. An agent can losslessly pull relevant info from massive databases that are analogous to long term memory and load them into context. As long as the database exists, it will remember everything inside it during every session.The LLM is always limited by the context size. it doesn't matter if there's a rag database you can pull from. not being able to adjust the weights dynamically is a fundamental limitation to current LLMs. Before you give it sufficient context it produces arguments that are wrong. >>540107621>Google's "Nested Learning" is a breakthrough machine learning paradigm designed to solve the exact problem of frozen AI brains and plasticity loss. Published by Google, Nested Learning fundamentally changes how we view AI architecture. Instead of treating an LLM as one rigid, frozen block of layers, it structures the AI as a hierarchy of interconnected, multi-level learning loops that update at different speeds. This effectively gives the AI a form of structural, ongoing "silicon neuroplasticity".Maybe I dunno.>>540107568Another AI dickrider
>>540099541>he envisioned the world where only the "western liberal" democracies alligned with the US (pre-Trump era) could have access itJews aren't any good at long-term planning because they can't self-reflect on the consequences of their actions. I mean it explains their entire of history of getting destroyed repeatedly for 2000+ years. They will never win. They'll max out their jewbris thinking they've won, then they'll find themselves getting purged like always. Loop until jews no longer exist. Honestly, the world would be doing them a favor putting them out of their misery permanently.
>>540101958>won't talk about Chinese politicsif youll listen, itll tell you about confucius, mencius, zhang zai, mao
>>540107568>so-called frontier>competitor releases superior productThe open source products are the better products with better hardware efficiency.>The only reason they can make moves is because they know their model and supporting tech is currently the best.This LLM field is one of the most disgusting field full of people scamming, lying and plagiarizing shit. The Chinese plagiarized from claude/chatgpt while claude/chatgpt plagiarized everyone else.>You're probably an indian or a bot yourself.I'm not Indian. The people dickriding llms are the bots.
>>540108398>The LLM is always limited by the context sizeObviously it has an upper limit, just like the human brain is limited. The database can be edited on the fly and the agent can address that database as a source of truth that takes priority, just like it does by web searching trusted sources. A multi purposed context window being the vehicle for its long term memory doesn't mean it's short term bound.also telling the truth isn't dickriding you're just a liar
>>540099541Dario is based as fuck, he acts in his own self interest,, but compared to any other billionaire in this city, he's probably the most ethical of them all. https://qz.com/anthropic-dario-amodei-open-weight-ai-ban-safety-testing-072826
>>540108784
>>540108648>The open source products are the better products.ignorance or lie, hardly matters though. I just checked: OpenAI has the best taskcost/intelligence ratio, Google has the highest speed and end to end latency, and anthropic obviously has the best top end.>This LLM field is one of the most disgusting field full of people scamming, lying and plagiarizing shit. The Chinese plagiarized from claude/chatgpt while claude/chatgpt plagiarized everyone else.right, just like all olympians are doping but someone's still going to throw a rock the farthest. It's a given and not relevant.
>>540108958>OpenAI has the best taskcost/intelligence ratio, Google has the highest speed and end to end latency, and anthropic obviously has the best top end....and z.ai is cheap
>>540099541100 unfiltered 30b models in the hands of 100 autists will be more powerful than whatever bullshit this nigga is selling to the government
>>540099541If I drew a picture of that man, I would be arrested for hate speech.
>>540099541Literally the meme
>>540108648>>540109011 (absolute cheapest is 5.6luna but I didn't include it because that's not that relevant)Open source is cruuushing it brosWhy can't you just argue the position properly? Open source is good because you can run it locally. That's the whole appeal but you delude yourself into thinking it's actually the best.
>>540099653Jews are PolesPoles are Jewsit's the Law
>>540099541He is just another Sam Altman. .. he'll push the free to use ethical ai grift until he is in a position to cash in.. do not trusthttps://www.youtube.com/watch?v=Rh0X5qxx69w
>>540099693because chud said so and he apparently knows it all. what would we do without chuddy.
>>540108231predictive programming
>>540101958>It codes as well as Claude for 10x cheaperHave you used it? I've tried Deepseek and GLM (z.ai) and they were totally fucking useless. Claude is just smarter, understands more and accomplishes more. It definitely feels smarter. The others don't understand, or outright misunderstandI'm told a lot of the difference comes from the cc for vscode harness. if Kimi was really better then I would try it no problem, but I'm not willing to accept any downgrade in quality whatsoever
>>540099766>Elon even made all his EV car tech available to the world as open sourceno
>>540111024he even charges his EV customers for it's use
>>540102759>nooo they're going to steal your codeLol, code is worthless now.
>>540100196>Beijing's line on Taiwanthe UN official world line on Taiwan?
Any idea why Reddit and Twitter leftists seethe so much they’re frothing at the mouth due to AI while right wingers largely accept it? If anything, I thought it’d be the opposite
earned, fable 5 shits on everything and its not even close, its about a head and shoulders lead this jew has, and its honestly the most impressive business feat in 50 years
>>540111701left wing sees the removal of art and creative writings (theyre wrong), Right sees the production and payroll (trillions). Chatgpt image creation will end any artist or commission work. Really only 20 remaining years you have to deal with humans who reject it. Overall its bc the left, being an ivory tower ideology, didnt realize doctors and lawyers are now first on the chopping block and cleetus and jamal on the farm are good
>>540111406these retards think trillion dollar companies want to get them v&'d for jew posting, racism and trolling. I have an entire project dedicated to the rhineland and khazar theory with claude from every anti semitie on earth to academics. Claude doesnt give two fucks
>>540099653If you say so rabbi
>>540099541Everything aside. Why wouldn't he. They literally siphoned anthropic. Imagine making Stat of the art tech. Just for some gooks in a 80 square feet room in a dark alley in China to farm 90% of your capability. Opus was always phenomenal. Fair is fair when due.
>>540100196BAHAHAHAHAHAHHAHAHOH NO
>>540114231>the same people who outsourced all of America's jobs get it done to themboo hoo
