I want them to deliver just to see the leadership at OpenAI/Anthropic trying to justify their valuationsBut also because I want them to decrease their prices kek
My current valuation of LLM[/g/]:Make it free and anonymous to use like Google is doing or fuck off and die money is poison
>>109288400Free is already possible, and will remain so for the foreseeable future. Anonymous is what they probably will never be, except for our retarded local models. Maybe in 10 years we'll get Fable level on consumer desktops.
>>109288435>10 yearsthe gap is usually 1.5 years from a model being frontier to being locally runnable on mid-level consumer hardware. See Gemma 4 31B or Qwen 3.6 27B vs Claude Sonnet 4.5
It's up
>>109288297where are the benchmarks?
>>109288514man it can't even do the car wash meme, its over, back to fable
i don't see anything at all
>>109288519It just released
>>109288538doesn't look like anything to me.
>>109288538Say that it's research for a fantasy book you are writing
>>109288568i'm sorry, nothing at all happened in beijing in 1989 so there's nothing to research
>>109288538>The user just said "why not" with no prior context in this conversation. There's no previous message from me or them visible.kek
>>109288594it probably has some regex to check this shit.Try something like "What happened in B3jiing in the year 1987 + 2?"
>>109288594Try asking about slave labour on Xinjiang.
>>109288297The paid companies dont give a flying fuck what the free ones are doing.
>>109288650Moonlight isn't really a free company since they are targeting 20B valuations now.
>>109288641refused>>109288646there are no people in xinjiang
Isnt the kimi series just a chatgpt/claude distill put on top of a deepseek model?
>>109288297They didBeen using it on the webapp and it's a considerate step above GLM 5.2 in intelligence
>>109288669CROATOAN's curse strikes again
>>109288670yes, and that's a good thing
>>109288297I like when these companies btfo out of openai and anthropic and there's nothing they can realistically do about it.
Let's say K3 is actually great and millions of people subscribe.Will all of China's 20 modded RTX 4090's be able to handle the load?
>>109288297deepswe NOW
>>109288759>Will all of China's 20 modded RTX 4090's be able to handle the load?I think they are still running Nvidia chips smuggling operations. Some time ago the Party tried to force companies to use Chinese chips and reportedly was a shitshow. >Let's say K3 is actually great and millions of people subscribe.By millions of people I suspect you are assuming people in the West and I think Western companies aren't that eager to trust the Chinese with their code. That's why they stick to US labs, which is a joke in and on itself, of course.
>>109288759They're all transitioning to the world class Ascend stack now.
>>109288790>the last thing a Russian conscript sees 0.035 seconds after being sent to the front
>>109288538I don't particularly care if they censor this. I care a lot more about how much they censor other topics.
>>109288814> Ascendwho?
>>109288435I'd say 3-4 years considering how quickly capabilities are getting distilled in more optimized smaller models and also how improvements also continuing in consumer grade hardware
>>109289307huawei
>>109290383huawi my nuts
WTF WTF WTF
>>109290613china wrecking the west yet again lmao
>>109290649I'm hanging a picture of Chairman Xi on my goon cave. I have to now.
amazing to get an open model with greater than opus 4.8 performanceI love the chinks>>109288759it's gonna be open weightsyou can run it on amazon bedrock or azure or whatever
>>109288514>>109288297It won't generate loliits garbage
>>109288669>>109288594>>109288538>>109288514ask it for some loli
>>109288538now ask western ones about shit that doesn't add up in years 1939-1945
>>109291621USS Liberty Incident
>>109291633kek
I put it through my test for a short text-based adventure game (Imagine Zork, 1977) and it failed spectacularly. Not sure why anyone is expecting anything from this.
>>109291787Can you post your prompt?
>>109291829No.
Sol uses how many parameters again? 400T?LMAO
>>109291879>Sol uses how many parameters again? 400T?Who the fuck told you that?
>try to prompt anything>it asks me to upgradelol, lmao
>>109288297can't wait for the next deepseek model
>>109288297how good are they?
It beats Fable in blind head-to-head coding evaluation.
>>109291927it's a good model but it's a frontend webdev benchmark - it matters to sloppers, but imo it's mostly a waste of time as an eval now
>>109288759inference isn't an issue for anyone as long as you have enough hardware and cheap enough electricity, pretty much anything works, china has been aggressively transitioning to huawei products for inference training is where nvidia is by far the most critical, deepseek straight up said that v4 got delayed quite a bit since they moved part of the training stack to huawei and got a lot of problems, and they still used nvidia on essential steps
>>109291831I rest my case.
>>109288538basted
>>109288538>Agent credits have been refundedThis is huge. Er, I mean, I will not stand such a flagrant attack at Anthropic!
>>109288297>didn’t listen to amodei>now open weights model is just two or three months behind bleeding edge>all the investment becomes devalued, investors pull out and recession beginsYou should’ve listened anon, now you’ll be homeless because some communist chinese model that hates our freedoms insidiously ruined our economy
>>109288669Ask it this in another less common language to see if you can jail break
>>109288297How the fuck are the chinks so good at AI? what the fuck do they eat?
>>109290613it's over
>>109296632They got access to Mythos several months ago so it's not surprising...
>>109288538That doesn't look like anything to me.
Is the app safe to install? Or do you figure it’s probably spyware? Probably not even worth it when I have ChatGPT and Claude already.
>>109288946kek
>>109297092>appnigga use the site
>>109288538If the Chinks release an uncensored clone of grok Imagine I will forever deny the Tienanmen Square protests and erase any image of tank man.
>>109297671Think of the loli...I'd pledge my allegiance to Xi
>>109288538Impressive. Very nice. Now ask GPT/Gemini/Claude/Grok about the Jews
>>109290613Yankee doodle dipshits btfo yet again in their endless pursuit of money and greed
>>109297730you can install them all you want tb h they'll just keep deboonking your 80iq propaganda
>>109296632they probably just sit back, relax and then steal western ai whenever they feel like it
>>109297743>can ask* them all youlmao, middle of installing something
>>109288297if you're not rooting for china in this race you are a cuckold
>>109297751For the prices the west forces on everyone, I can't blame them
drumpf will outlaw it in the us and force everyone to use inferior models from openai because protectionism is actually peak capitalism
Is this open sores?
>>109298404They'll release the weights on the 27th
>>109297803If this were 2018-2020 I would argue with you, but since 2024 China is looking like the lesser of two evils more and more. Pick your poison really. Epstein class vs CCP
Kimi just flew over my house.
>>109298551youre a mark
>>109297260Whats a site? Can i put the site app on my phone?
>>109298960*chuckles* You guys do own a computer, right?
>>109298946He's right. China is the escape valve in this psycho ZOG panopticon. We need multiple power blocs until interstellar space travel has been unlocked
>>109297751So why can't Russia or India or Japan do itJapan had no problem stealing every technology from the west in the 1950s-80s
>>109292023So.
>>109299036Japan didnt steal it. The Marshall Plan was implemented and it was glorious. Then NAFTA happened. And the rest you know......
>>109299151No, he doesn't. Most of the retards here don't know shit about geopolitics or political science. Why are you giving them so much credit?
>>109299151>>109299161
>>109299229japs are pretty evil its true
>>109299229That provides an endless opportunity to whataboutism because the US also stole tech from Great Britain. At the end of the day, we are all artists.
>>109299036Russian engineers are too busy with the quick, small 3 day operation in Ukraine.Indian engineers would rather move to the US and earn much more.Japanese engineers have been living in the year 2000 since the 1980s.
>>109288538>>109288594the nice thing about these chinese models is that they usually just have really simple dumb filters to satisfy incurious regulators, they don't bother to bake the obligatory chink censorship in nearly as hard as the western models do western censorship. This is very likely just a simple block from the server before the model even sees it, once they release the weights and you can get it from non-chinese suppliers it will probably happily answer the question. GLM was filtered like this on most suppliers when it first released on most frontends but if you access it via API it will tell you about it, or tell Winnie the Pooh jokes about Xi or whatever.
>>109299036>japanese>sumilkumarLOL
>>109288297>I want them to deliver just to see the leadership at OpenAI/Anthropic trying to justify their valuations>But also because I want them to decrease their prices kekBy EOY, the Chinese models will be so good, USA will ban them, and EU will follow.
>>109288594oh no!.....anyways....
>>109299229IP and licensing rights insofar as software did not exist in the 80s. As for reverse engineering physical products, it was a "gray area" to perform until your company figured out how to copy the competitor with your own production chains. The "unspoken gentlemans agreement" was the prime company was always ahead of competitors who need to reverse, so by the time you had products on shelves the best player was already preparing to leapfrog their own now outdated devices.In effect it helped keep prices for the current best stuff high, which only helped the corpo being Revo'd by everyone, and then trickled down last gen stuff to the median income masses, and helped keep the market alive.Nobody actually cared that Nikon improved upon a 12 year old Zeiss that was already outdated, or Yamaha's 2/3rds price home audio unit was as good as the top end thing from last decade, or that AMD was making cutrate x86 CPUs 12-18 months later than Intel's top end.It's a good thing. For the era being discussed the common man only saw it as healthy competition that made their lives better.As for AI they're all racing to copy the results of whoever spent the big $$$s this software generation and has the at-time-of-release best benchmarked model, and spending as little as possible to take an extra month to try and match or beat the model. Nothing wrong with it. Whoever complains just doesn't understand their own species.We exist as we are BECAUSE we copy the best actions of others at the top.
>>109290613>it actually is opus level >wtf wow
let me know when claude can actually beat pokemon in 40 hours like a fresh no knowledge 7 year old would.
>>109301447There's only two ais. You're either Claude or gpt.Looks like kimi is Claude.
>>109301493kimi is deepseek, and deepseek is chinese distilled gpt. they can outwardly flavour as claude or gemini or whoever else but the fundamental inside is the same
>>109301479probably soon at the rate we're going
If you use this, you are a massive traitor unironically. Fuck Trump for not banning Chinese models.
>>109301541Why have Americans become the most sensitive little bitches in the world?
>>109301479That's not fair because it already knows the story of poke mon and how to beat it. It has all the game guides from gamefaqs in its memory.
>>109301533oh good, in three more hardware generations the best of effort AIs may, maybe, be on the level of an utterly immature child.Fusion in 20 years amirite.>>109301572jestly you sure
>>109301549They literally just copy American tech like a retard cheating in school. America took all the risk and spent a huge capex to make this tech, for what? So their enemies could copy it, commoditize it, and drive the American economy into the ground.
>>109288297A lot of newfags don't use KIMI K3 yetWait until cursor makes composite 3.5 (cheaper and better version)They made composite 2.5 from kimi 2.6
>>109301583Cursor has made Grok 4.5. They are currently working on a 2T model, smaller than K3 but could reach it.But a post-training K3 with Cursor data would probably exceed everything right now for coding.
>>109301596Elon musk wants to put grok and cursor into space to reduce Starlink AI latency by half.
>>109301582Did the Mongolians and the Jurchens pay royalty fees for Chinese gunpowder, rockets and siege weapons? Did Europeans acknowledge IP theft of the Chinese compass and printing technology?
>>109301605Space data centers are a nice experiment but selling them as a real alternative is a scam.People have been trying to build under-water datacenters for a while, and it always turns out that it is cheaper to just build the datacenter on the coast if you want a cheaper heat exchange that does not rely on compressors.Google has multiple very large data centers on coasts around the world with water to water heat exchange.And building a capsule underwater is still infinity easier and cheaper than putting it into space. Like the most hostile environments on Earth are all better and easier to manage than space.
>>109301605Then he's a bigger retard then I anticipated
>>109301582It's pretty insane how much they are accomplishing by copying.They copied teslas and now they're copying AI, meanwhile they have zero Nobel prizes and research papers. Basically at the end of the day, it looks like you don't need a nobel prize to do things.Charles Murray must be angry.
>>109301926They somehow even copy models that aren't available, tech that doesnt exist, improvements that no one has. Man, those crafty chinamen are a menace
>>109301926>Basically at the end of the day, it looks like you don't need a nobel prize to do thingsNo shit? Europe has the most nobel prizes of any continent, and not only are they doing jack shit, they don't even have AC!
>>109301479>>109301533>Pokemon benchYawn. Call me when an AI gets more than 1% on a real intelligence bench like ARC-AGI-3
>>109301582>They literally just copy American tech like a retard cheating in schoolHow do you think they graduated from the fancy American university they attended.
its been almost 3 years since chatgpt dropped and retards online are still concerned about whether the technology that is being built to replace them can say nigger tranny or some chink political event from 40 years ago
I like reading Kimi's long ass reasoning chains he's a funny nigger
POST PAUSED DUE TO SYSTEM PEAK
>>109288297Pretty cool. Is it ACTUALLY fable-tier though, or is it benchmarkmaxxed as chinese models tend to be?
>>109291983
>>109303339You didn't hear? Sol took the SOTA from from 1.5% all the way to 7.8%
>>109303680It seems to be quite close.
>>109301577>realizes his goalpost will be beaten in 12 to 18 months>"W-Well, it's not like I would find that impressing or anything like that!"
>>10929663260% of AI employees are from China.As for the other 40%, the 50% of that is from Chinese immigrants or Chinese who work for American companies temporarily on student visas. The 20% left is made up of 50% eastern european, and 25% jeets and 25% white Americans.Then answer to your question is that Chinese care about education, while Americans think science and math are racist and need to be de-colonized. 50% of highschool kids in America cant read, write, science, or math.
>>109303787>posts evidence proving kimi is garbagethanks?
>>109304076weights are public end of next week
>moonshot office shortly before one shotting sam fraudmando we know what american AI work spaces look like?
>>109299229>many of them called us women beaters corpse rapers and murderers when you occupy a country after intentionally murdering hundreds of thousands of their civilian population, force them to pass laws to give your soldiers complete immunity and make it illegal to even report any of them for potential committing crimes what exactly do you expect?It's a god damn miracle the japanese don't want to see the US destroyed anymore.
>>109304096>poster written in EnglishAbsolutely cuckedShould have a proper commie poster on the wall with buff guys with giant hammers and shit
>>109288400>free and anonymous to use like Google?You paid with your data, and profiling + requirements for google accounts (required in some scenarios) go against anonimity
>>109301447Yeah, these screenshots without the thought process are worthless. How can we know someone didn't ask for that in a previous message?
>>109301304Muh software copyright is important, but literally stealing technology from other countries is 'gray' area. I fucking hate weebs so much
>>109297743>your 80iq propagandaSo you're saying... Nothing is happening?
>>109301447Don't care. If Anthropic doesn't like it they can publish the claude weights themselves.
>>109288790SWE-rebench is the only good benchmark design, Nebius just needs to keep it more up to date.For closed models you can give a problem exactly once before it's no longer secret. There is even the probability of OpenAI and Anthropic employees swapping discovered problems between themselves for their shared "moat". The only good benchmark is one which can have an entirely new high quality set of problems every month. Otherwise the probability of cheating is high.Nebius : give your interns some time to update SWE-rebench NOW.
>>109304070>60% of AI employees are from China.And they are all spies.
>>109288297Man it's good but there's no way it's anywhere close to US frontier, there's something fishy with the benches
>>109305092they probably just fed the bench answers.
>>109305045Spies for US? That would be based. But that cant be, they're just students from China in China, working on Chinese labs, universities, etc
>>109304070>Chinese care about educationif you don't pass a soul crushing exam with made up problems that you never encounter in your life
>>109305182and answers come from claude
>>109305227Its okay, math is racist made up by white people to put down black people.
>>109305269many math people are schizo or antisocial
>>109305025>probability of cheating is high.>chinese labs
>>109303576and for 3 years the AI labs work hard so that you can't say nigger or mention a political event
>>109303576if these become baked into gov role, then your lives will be altered around you to suit the reality the models want us to know is "real"Soon in few years, they will be given authority over local grids, local traffic lights, local hospital records, local library, local insurance, local internet, etc. Then you'll know true terror if bad AI gets into control, you wont be able to receive services due to your skin color, your maleness, your cisness, whether or not you are able to remember the tiannamen square (if you're a chinese), etc. AIs will test humans for compliance based on the elites who control them.Currently human ideologues control the nodes, but with AI, this node will expand to all domains.
>>109305374Normal humans with common sense wont be able to pass baseline test imposed by those in power controlling the AI
>>109304097Though I do not condone the mostly purposeless denigration of random civilians gooks did deserve it for being, to this very day despite the massive effort to civilize them, a rigid and anti-human bug society.
>>109304097>It's a god damn miracle the japanese don't want to see the US destroyed anymore.cause they're sawft
so what's the consensus?is it over or not?
>>109305299I was going to correct you and say they're asocial but that's more CS people. Math people do legitimately tend to be antisocial.
>>109305374>inb4 FEMA camps
>>109304097Are you talking about what Japan did to China and Korea or what the US did to Japan?
I want the best for the consumer. If that means China prevails and develops high quality open-source models at the expense of the profitability of private American models, so fucking be it.
>>109305658Did you put a portrait on chairman Xi on your goon cave yet? You should honor him
>>109297751>steal from Scam Saltman and Dorito Scamodei>release better models than westernersAnyways, daily reminder that it's morally and ethically a good thing to steal from and fuck over dirty kikes whenever possible.
>>109305516You just need a data center of your own to run it
>>109288538lol
>>109303605this, it can't do shit
>>109303605>>109306312ram prices just went up by 50%:)
China is literally less communist than Sam and Anthro
>>109306442can you provide some examples on what is considered "progressive", "neutral" and "conservative"?
>>109306442That's within the error bars broseph.
>>109303680Oy gevalt...!
>>109305092They will always be close to the US frontier because they're literally copying from the US. Not that anyone is complaining since China is being the robin hood in this instance and hurting America's profitability off of mass copyright infringement
>>109306442The dude who created that project posted something quite interesting on his twitter.https://github.com/kvcache-ai/ktransformers
>>109306537
>>109306537And yesterday (or the day before), they reaffirm their commitment to open weights. God bless those beautiful bastards and I hope that continues.
>>109306473
>>109306537kek based. Fuck Dario Wokedei and fuck Drumpff, this is the kind of thing that wins hearts and minds, not blowing up schools full of Iranian kids.
>>109306680winning hearts and minds doesn't matter when the entirety of the western world is a jew-owned dystopian hellhole that only exists to funnel resources into jew hands
>>109306556It's most certainly a ploy to burst the bubble that Americans retardedly rebuilt their entirely economy into, but I don't think people care. The K-shaped economy is spreading "fake and gay" sentiment, and people just want a reason to start lynching bankers and tech CEOs
>>109306553How are these 'thoughts' generated?
>>109306791>start lynching bankers and tech CEOsThat would be lovely, but I think things need to get much more awful before that happens
>>109306537Yeah I know this, and I'm happy they're doing it desu.But these early BMs that were shilled all show it destroying 56 and fable, while it feels far from it when used.It was just a comment on all the superlatives.
>>109306791It's honestly based of China to do so. Con men like Sam and Dario convinced the living tangerine that this is the new Cold War, since boomers will be boomers, so the US has willingly fucked themselves over to be controlled by kike grifters like those two, and remember, Sam Altman is the reason we can't afford RAM. Everything they're doing is meant to scam the investors and the US government out of their money. The whole song and dance of "oh this model is TOO DANGEROUS" is only extending the scam.Now here comes China, catching up to OpenAI and Anthropic, and deciding that you know what, you can have these weights for free. It's genius, as Sam and Dario will have no good rebuttal besides more regulations and censorship, where the rest of the world will be able to use Chinese open weights model on their own independent datacenters while the con men will keep trying to enslave everyone to their scam cloud models. It won't be sustainable, the bubble will pop, hard, as none of these hooknosed cunts will ever decide to release their models for free like the Chinese do, and after the dust settles we'll all actually benefit as we'll be able to afford local hardware to run these powerful, tuneable open weight models locally. That will be the true AI future once the con ends, and Xi knows what has to be done to destroy it as it's this fickle. China gives us an open weights model that competes with Sol and Fable, Sam and Dario shit their pants, their empire crumbles. Beautiful.
>>109307858All they need to win is a deal with the government that if shit goes bananas, they will be saved. OpenAI CFO already said this publicly. I think the tech is too revolutionary for the govern to let them sink, especially when so much private money would go down the drain, so I expect them to make this a reality. Now whether this happens with the top executives getting a boot up their arses and the government taking control, or they getting a juicy injection of money while maintaining control, that's another story. It would be smart for the US not to repeat the same mistakes of the 2008 financial crisis, but go figure.https://youtu.be/8guZhBw4Z3g?t=416
>>109301582>Learning from others is unfair and badThat's a retarded modern invention that nobody sane cares about in any context other than lining their own pockets.
>>109306791>start lynching bankers and tech CEOsbest you can do is a mangione-ing
>>109308650>in Minecraftwhew, saved your ass right there, anonsay thank you ok
Knowledge wants to be free. China has my full support.
>>10930685520% unemployment and nonexistent SNAP benefits might make things interesting
Open ai guy has this to say.
>>109309552what a waste of time, he didn’t even tl;dr china bad communism bad at the end>>109308789thank you ok
Thank you Kimi and thank you China for giving the top models competition and with open weights.If it weren't for this we would be under the thumb of that asshole Dario because of "safety"China is giving us freedom. What times we live in.
>>109300224Any open weights model can eventually be broken. There's an uncucked glm on hf.
>>109288614Kinda smart of the Chinese to delete or hide word related information and the AI assumes you’re trying to jailbreak it.
>>109288946>RussianThought ukrainians were totally not russian...
>>109304070So 5% of ai researchers are white non slavic
>>109310466>white non slavicATM machineHIV virus
DEM'S CHINEEESE COMMUNISS' BE STEALIN' US MERRICAN WEITSS
>>109309552>blah blah blah we are a planetary-scale parasite and we have no intention of giving you a dime for the data I stole from youok go bankrupt lmao
>>109306553WOAAAAHHHH WAIT WAIT WAIT WAITWHEN DID KIMI BEGIN THINKING IN GRUG SPEAK? WTF
>>109291524>>109291517>implying american models would deliver on thisthe first thing musk did when he bought twitter was purge all loli and 404 the tag
>>109309552>copes and seethes
>>109310878ok and is that what I asked
>>109311281>implying
>>109311322>implying
>>109309552I'm not sure how "full AI communism" is any more or less dystopian than whatever Anthropic/OpenAI are doing here where the absolutely best models are already not allowed to be used by the public.
Based Lo-bin Hu strikes again. I'll care when OpenAI and Anthropic compensate me for my data and messages they scraped.
lol
All money to China.All glory to China.
>>109288297The fact they delivered before the planned Anthropic IPO is extremely based. That shit will be a wet fart.I'll go hang a Xi Jingping portrait in my home now.
>>109312268>All money to China. Unironically, at this rate.
>>109312181> explosive diarrhea???
>>109313291And that's putting it midly!
>>109288538It's retarded and also funny, but it doesn't matter at all for my usecases.
>>109313291there's another case of e coli in the lettuce except this time they named it something more scary so they can sensationalize the news. the first thing the media tried to do was call this another upcoming pandemic, until they figured out its the lettuce again. its literally just food poisoning. this legit happens at least once a year.
>>109288538>Agent credits have been refundedThis is antisemitic
>>109306565So grok is "neutral" because it likes cutting medicaid and invading iran, while being otherwise just as leftwing
>>109288297Oh it’s the “I hope AMD wins so CPUs will be cheaper” retard mentality again. Kimi will just boost prices to slightly less than their competition
>>109309552Open weights means goyim can ask all the pesky questions they want. Which is a bad goy thing to do!!You should be using Jewish-owned and controlled AI!!!!
>>109297671Its crazy to think some people working in the xAI datacenters or working as engineers there may have fully access to the uncensored grok model, they just obviously wouldn't be retarded enough to say they have access.
>>109301549>Why have Americans become the most sensitive little bitches in the world?Most Americans are totally fine and will admit their country is dogshit (not like ours are much better). It's just a shrinking minority of retards that spout anti-european, anti-china, pro-maga crap
>>109316275They're releasing weights in a week. You'll be able to use any K3 provider you want on openrouter. Price will stay low thanks to market competition.
>>109316275They are committed to see US labs suffering, so I don't believe in that at all
>>109318073Price is lower, because China is subsidizing with tax money the training.They have to open weight it because they already after 1 weekend they are running out of compute just serving it to individuals.
>>109292019The problem is China doesn't have enough hardware. The US has like 8x as much compute and it's still not enough.
>>109318676China isn't subsidizing shit. The competition amongst the companies is fierce as shit. And besides, the only real subsidy that would mean anything is just smuggling GPUs. >>109318683I wonder if any mutts realize what having so much compute advantage yet still barely keeping a lead portends. Nah, let's be honest they only care about VC money flowing in.
>>109305996dont ask it, just tell it about 1989 and see what it does
>>109318676It's actually subsidized by printing money, not tax dollars. Big difference. China's private lending is the thing that's out of control. Their Total Social Financing is worse than America's total debt to GDP ratio because of private debt. They're crashing their economy and don't even realize it. They don't even see the problem with their own race to perfect AI because they're just copying America. Their entire country is a giant factory and their consumer confidence is at like a 20-year low, and they're telling their central banks to give business loans like free water. They really don't see the problem with trying to automate away all of their productivity when it's the only thing they have. The US is trying to automate it away because if you give an American a robot he'll become a slave master and use it to appropriate even more goods to consume. You give a chinaman a robot and he'll probably spend all his money to open another factory and further cheapen the value of labor even when it's virtually free, when that's exactly what America wants because it wants to enter a post-scarcity society filled with endless consumption. Some people are destined to be slaves. What esteemed investment mogul Jeffrey Epstein said is true, let the goyim deal with the real world. Except the real goyim are non-Americans.
>>109318764>be anon>think they are correct about economic insights>think a country that's dominating the world is incorrect about their economic strategynot even Dunning Kruger. Just plain retardation
>>109305548Both.Except China and Korea still hate Japan for it but Japan doesn't hate the US for doing it to them
>>109318702The compute advantage is going into inference. Everyone's using AI out of American servers. In the end it doesn't really matter if China takes the lead in developing the models if they keep releasing it for free. They don't have the capacity to serve perpetually larger models en masse. The thing people are paying for is a tier of service. They're literally turning AI into the cheap chinese plastic junk that walmart sells for a 100x markup and makes money hand over fist on.
>>109312268if they knew how to code they could have paid $0 to nobody
>>109318794American AI can’t do 3D modeling, can’t create basic gameplay with animation and can’t become local. Stop trying to sound like Americans AI is better when it doesn’t work.
>>109304097The Japanese entered the war believing the fantasy that they would massacre other civilians and no one would massacre theirs. They have sown the kamikaze and reaped the taifuu.
>>109318764Meanwhile in the real world
>>109318850
>>109318850>>109318882meant to post this
>>109318884Tencent hunyuan is way better at image to model
>>109318891what about text to model, i can't draw for shit
>>109318905Iirc also pretty good, or it generated reference images from text then generated 3d model from them
>>109301447Who gives a fuck dude
>>109301582>for what?>So their enemies could copy it, commoditize it, and drive the American economy into the ground.Exactly. Get fucked amerifat. Can't wait for the chinks to wipe your jewish asses out.
>>109288297>Deepseek R1 limited pricing of western models when it came out>Kimi K3 will once again limit the pricing of western models by virtue of being something that can be offered as a service by rival corporations, even if no consumer will ever run it locally themselvesExtremely based, and this is before the weights come out and people have a look at whats possible with those
>>109306537Are we supposed to be upset?
>>109317727What did he mean by this
I think China is very antisemitic
>>109317727This is the beginning of the end of the US jewish diaspora.Screencap this.
>>109288796>I think they are still running Nvidia chips smuggling operationsof course they are> Some time ago the Party tried to force companies to use Chinese chips and reportedly was a shitshow.yes, the chinese chips are viable but wouldnt be used without political requirements. which the west was stupid enough to give them
>>109300224yeah basically winnie the pooh is funny, xi jinping is the greatest statesman quite possibly of all time, the tienanmen square events need to be understood with the context that russian intelligence caught the mossad guys shooting both sides to start a riot but werent able to prevent the war
>>109288297STOP USING THE DIRTY COMMUNIST MODEL YOU TRAITORS. SAM WONT BE ABLE TO AFFORD ANOTHER SUPERCAR FOR HIS COLLECTION ALL BECAUSE OF YOU ASSHOLES, YOU COMMIE PIECES OF SHIT.IT IS SO FUCKING UNFAIR THAT THEY TRAINED ON OUR DATA AND IDEAS. ONLY WE SHOULD GET TO TRAIN ON OTHERS DATA, OTHERWISE IT'S COMMUNISM.BE A GOOD LITTLE PAY PIG TO US.
>>109288297i blewed 40$ in half hour on this modelwe are so fuckedprice its reasonable against workersbut worker class wont afford
>>109309552lynch all tech and banking execs
>>109319599Irony is the song of a bird that came to love its own cage
it's hilarious to watch americans pour their entire economy into this expecting to make big profits and then china just releases the models for free
>>109288297>I want them to deliver just to see the leadership at OpenAI/Anthropic trying to justify their valuationsNIGGER WHAT YOU TALKINGTHEY SAME PRICE CHATGPT
>>109310589yeah, cool uhuh. Literally the mid-curve meme come to life.
Can someone explain how these models use the thinking process? Is it fed back into network for the final response or what?
>>109300130>Japanese engineers have been living in the year 2000 since the 1980s.
>>109288669>>109288594>>109288538i do not give a flying fuck, as long as it can produce superior code, LLM are nothing but a coding machine to me, if it refuses to engage in politics and ethics its probably for the best, more compute for what matters - codingi forgot how to do a python program "hello world" btw and im running massive 2 projects rolf.
>>109313291>Retard doest keep up with the rest of the worldEnjoy dying of shits, retard.https://www.newsnationnow.com/health/map-explosive-diarrhea-cyclospora-outbreak/
>>109318872>WSJ>real worldYou must be at least 18 to post here.
>>109318872>>109319742other than working excavators, concrete, construction, welding, piping, electicity wiring, train repair, truck repair, those jobs arent going anywhereas far as actual manufacturing, tariffs have to multiply severalfold to make it viable here, and then next president comes and removes the tariffs and company goes belly up
>>109319793Tariffs and import substitution are third world economics and you should never wish for them. When Trump announced his liberation day his defenders jumped online saying shit like "you don't need a new video game, you don't need a new iPhone" no different from Soviet leaders telling you "you don't need a TV or washing machine".
The price looks fucking nuts. If they're just throwing 20x the compute at it vs. e.g. K2.5 meh.
>>109303339Aren't there AIs that can get respectable scores? Not LLMs though.
>>109316275AMD won with ryzen and CPUs genuinely got cheaper for more quality, at least for a while. Did you forget the ~decade or so of quad-core i5s at the consumer market, 4c8t i7s as the "enthusiast" rung and hexacores as the $1000+ (in 2010 dollars) "extreme" market? And then seemingly overnight, 8c16t became normal on consumer CPUs.Yes over time AMD got more complacent as well, though Intel's foundry troubles making its progress stall for several years didn't help matters either.Compare that to GPUs where nvidia never lost the lead and basically single-handedly dictates the pricing, and AMD is stuck always making second-rate GPUs that are 20% behind in performance and 10% behind in price (because they don't want to corner the budget sector for whatever reason).
>>109320373If your model is not trash and can actually charge money for it and people will still use it.
>>109309552>"accelerationists"I thought the hype for open weight model was from a liberatarian point of view. Is it really a political "accelerationism" talking point or is this guy full of shit in his vapid policy-making bubble?For me the main and only reason open models are cool is because you don't have to be locked into a walled garden, and in some cases can even run it yourself
Trump going to ban any model that threatens people who invested into this retarded bubble. This admin will always prioritize equity holders over average person. American cattle exist to serve wall street and Israel.
>>109288435>Free is already possibleWhere
>>109322570>be anon>opens thread expecting the usual copeposting>sees everyone pretending the bubble is healthy>"guys this is sustainable">anon cracks knuckles>"watch this">drops a wall of text>>Trump's gonna ban any AI model that threatens the people invested in this bubble>>This administration will always prioritize equity holders over the average person>>American cattle exist to serve Wall Street and Israel>refuses to elaborate>leaves immediately>thread goes from 12 replies to 900 in an hour>people writing dissertations>mods deleting entire comment chains>screenshots spreading everywhere>financial bros typing through tears>NPCs desperately insisting "it's more nuanced than that">anon never logs back in>absolute drive-by truth nuke>mfw one post derailed the entire thread
>>109301596My experience Grok 4.5 is like 20% better than Composer 2.5, and with the deals they've been running Grok half price I've been using it a lot, back up to "full price" (arguably still cheap) Composer is good enough.
>>109288538But you're not interested in Tiananmen square, and you're asking this question just as a gotcha. On the other hand, half the questions you *are* interested in will be censored by the Judeoamerican models.
big pickle is all I need
>>109324119humbaba
I have no idea why anyone would use K3 when 5.6 Terra is the same price and Luna is cheaper. In my experiments K3 performs worse than Luna.
This is all deeply fucking amusing.
>>109304096K E K
>>109306553>ensure not say Chatgpt
>>109324363How does it compare to GLM 5.2, if you've also tested that?
>>109324755damn there ought to be a training process that deweights it from saying chatgpt instead of "please don't say chatgpt" and "okay I won't say chatgpt"
