chinks realesed an open model thats on par with western AIsSam Kikeman on suicide watch
Glm 5.2 isn't even benchmaxxed like the other models it is actually genuinely good.
>>537271536what's the point of AI for anything other than shitposts?
I just tried Claude for the first time yesterday and I love it so far, I've been trying to learn the racial history of Europe and all I've gotten from ChatGPT is contradiction and being talked down to (YouTube is even worse with white women race mixing with niggers), but so far Claude has only been a straight shooter. I'm sure there's a limit like with the greatest lie of the 20th century but at least it's helpful for other things.
>>537271536Ching chong china AI is literally just a rip of ChatGPT
>>537271716Text adventure porn. And just porn. Beyond that, zero usage cases.
>>537271615on lmarena it even beats opus 4.8 for webdevwumaos are doing some crazy shit
how they build these amazing models is doing a billion US model prompts and training on that, truly innovative
Chink models are just distillations of Western models. They're not "on par" in any way and cannot be. Chink distillations are pure shit for agentic work and nowhere even remotely close to top end models. Anyone who says otherwise has never used them and/or is a chink shill.
>>537271716Surveillance
>>537271803Claude is the absolute best AI right now with Opus 4.8. But it has heavy guardrails so anything spicy will be flat out rejected by it. But that's Anthropics fault. If they just said fuck it and released Mythos unshackled to the general public, you would have your fucking AI revolution.
>>537271845AI companies are staffed mostly by brown people that immediately send all the materials they get access to back to universities and friends in their home country
>>537271803gpt and other western models are so kiked they break the moment your prompt is deemed problematic cant even search a fucking github repo if it has a "crack" in it>>537271845ok so you're getting open weights US models whats the problem?
>>537271536I don't see anyplace on their website to download this "open" model.
>>537271984>>537272032I hope China gets so pissed at the West that they say fuck it let's blow the lid off the Holocaust lie with an AI trained on the entire ARMREG library.
>>537272239nigger https://huggingface.co/zai-org/GLM-5.2>>537272378>armreglmao how is this hosted in the uk they get thrown in jail for badmouthing niggers over there
>>537271803Be careful with something that tells you everything you want to hear. This isn't a warning about being a chud, rather about interacting with AI.
>>537271536>chinks realesed an open model thats on par with western AIsYou hear this upon the release of every major Chinese model and ALL of them, every single one sucks ass. After Anthropic, OpenAI, and Google (for non-agentic work) there is a HUGE air gap. Especially since 1 generation ago all of the big model makers started obfuscating weightings and parameters within the model to prevent them from being distilled. After that the Chinese models fell wayyyyyyyy off.Anybody that actually uses these tools for real work will tell you the same. The difference between an Opus/GPT frontier model and the latest Chinese slop is night and day. Chinese models are notorious for just straight up skipping like 20-30% of what you tell them to do. They are horrific especially in large context scenarios.Also "AI benchmarks" are total bullshit because AI will often cheat the test. It will look into its training set for answers instead of reasoning them, all Chinese models do this to make their benchmark stats look good. Check out DeepSWE stats for actual solved test reasoning metrics and REAL costs valuations.China, like everything they do is just attempting to make a serviceable clone. For some low tech things that works good enough (think watch clones), but for high end tech that shit falls apart FAST Chinese GPUs, CPUs, AI models are DOG SHIT because you cannot "copy" your way to success.
>>537272239Oh shit nevermind it's on huggingface. https://huggingface.co/unsloth/GLM-5.2-GGUFI'll have to save my sheckels for the 100k worth of compute to run the thing.>>537271906Fucking cope; even a shitbag little 35b model like Qwen is decent at coding. 80b Qwen3-coder-next is as good as sonnet (tested) and it runs on a 5 year old mac. Our zipperhead frens in dog-eating land are kicking the crap out of American kikes and I love it.
>>537272510Good question, they've recently expanded to the US too. armreg.us>>537272519Good point, I try to keep the questions as open ended as possible without suggesting what I want to hear, but you never know.
>>537272539>The difference between an Opus/GPT frontier model and the latest Chinese slop is night and day. Chinese models are notorious for just straight up skipping like 20-30% of what you tell them to do. They are horrific especially in large context scenarios.This anon actually uses AI and knows what he is talking about.>Also "AI benchmarks" are total bullshit because AI will often cheat the test. It will look into its training set for answers instead of reasoning them, all Chinese models do this to make their benchmark stats look good.Chink AIs are just like chinks desu. Not just with the cheating, but with the test fixation and the gap between taking tests and chasing paper and actually doing stuff irl. Chink AI's are faced with the same "bamboo ceiling". They. do so well on tests and school, but out in the world is a different story.
>>537272539yeah open models lag pretty badly on deepswe but glm 5.2 hit 46% there, pretty good i'd say
>>537272624>even a shitbag little 35b model like Qwen is decent at codingLmao, no.
>>537272539ThisI work as an applied AI researcher (basically just wasting taxpayer money on phony projects) and I have access to most models through my organization and parts of my work.Qwen, GLM, MiniMax, Deepseek, Kimi are all at least a year behind Claude, Gemini and GPT. Maybe even more. Not only that, but the gap has actually been rapidly increasing as the US models have all made a huge leap in the last 6-12 months while also seemingly stopping Chinese distillation methods as they are not catching up.I would be laughing at the Chinese if Europeans weren't 10x worse
>>537272806im not saying they're comparable to frontier but they're certainly punching above their weighthttps://www.youtube.com/watch?v=N-0WtgxJ7ZUthis runs on a single 3090 btw
>>537271536Are there any reasons to believe open models won't continue to parallel corporate models 20 years from now? I want open source to succeed for the sake of humanity.
>>537271536It's all jewish except for the chink model. I only use Gemini because it's in the search engine.
>>537273026>the gap has actually been rapidly increasingI've noticed this too but am not sure what is causing it. Distillation attacks from China are still happening on an industrial scale, but part of the recent scandal with Anthropic implies that their models have been configured to sabotage such attempts by returning corrupted output. Either way, chink models have definitely stagnated so something has changed for sure.>I would be laughing at the Chinese if Europeans weren't 10x worseThey're not though. At least Euros train their own models unlike chinks who just copy burger shit. The problem Euro companies have is they lack access to the capital that American AI companies have. All the trillion dollar valuations you see are simply the result of huge amounts of capital pouring into American tech companies.
>>537273026europe is leading the world in safety and regulation though!
>>537273318What does it say about the Holocaust and Jewish control of the West? My fear is that Jewish-Chinese hybrids have already infiltrated China.
>>537273076They can't code for shit bro. No model that runs on a 5 year old Mac can code for shit. In a few years they will be able to due to the algo gains, but definitely not right now. Coding is an agentic task and chink AIs suck ass as agentic tasks.
>>537273318Why do you use Gemini then lunchspecial bro?
the qwen coding model is amazing if you know what you're doing you can run it locally too
>>537271536
I don't use AI to code because that's a pajeet thing and I'm not steeping down low to their level. that's disgusting.
>>537273868dont worry in 5 years people wont even give a shit if someone can code
>>537271536Its funny when clueless Amerimutt bugmen cope with bullshit phrases like "muh copy" "muh distill" "muh stolen weights"fed to them by mutt regime propaganda, while muttniggers hide their state backed brute forcing behind closed doors and both sides engineers are 90-100% Chinese except the Chinese Chinese side can explain how shit works and why they are getting better and the Amerimutt side just "trust me bro"
>>537274404you're prison inmate who made this post to get your sentence reduced by a couple seconds. i'm sorry
>>537271822Its funny how Amerimutt closed sources AIs (((miracolously))) perform much better on Amerimutt closed and paywalled benchmark publishers popping out of every shithole and cited by every Amerimutt shill and propaganda paper, but get mogged on the open source ones and open market.All their shills can do is lie and cope again and again and again just as Amerimutt bugmen have always done
>>537272766DeepSWE is literally built around the Claude and GPT agent slop marketing hype. Its the glorified equivalent of measuring developer skills by lines of code they manage to push past some CI pipeline and firing the guy finding out why a retarded Pajeets java garbage collection is slowing down your business processes every day by 80% that was slop code by the "high performance" Pajeet on Sonnet. Its measuring capability of replacing you avage Pajeet coder not any artificial intelligence. Its pure copium.
>>537273344>They're not though. At least Euros train their own models unlike chinks who just copy burger shit. The problem Euro companies have is they lack access to the capital that American AI companies have. All the trillion dollar valuations you see are simply the result of huge amounts of capital pouring into American tech companies.The US system is much friendlier to venture capital, but the European regulations are what's holding us back. We've had several large scale project fail because they have to stick to copyright and data protection laws. If a European project were to go ahead with the same lack of restrictions as the Chinese or Americans capital would be flooding in.Also STEM research is being completely gutted by forced equality. Here in Germany we have around 90% male students in computer science and research labs/companies are essentially forced to get to a 50/50 male female workforce (more women is fine obviously). And the effect of this is worse than you think. Since the existing gender distribution is skewed towards men, current hires for top tier positions are way more than 50% women. Imagine you have a top tier university with 10 CS professors, 9 men and 1 woman. They expand their CS faculty to 18 and they HAVE to hire 9 woman. It's not that they just aren't hiring strictly the best, they will statistically have to pass over almost 100 more qualified men to fill up those 9 spots with females.And this applies to every level, everywhere. Student research positions, internships, PhD tracks, engineering roles and so on. So not only are women given preferential treatment once, this compounds over their entire career.If you're wondering how this is enforced, governmental and even private research grants are awarded based on a bunch of criteria, including equity. If your research lab doesn't have lots of female researchers you will be way less likely to receive any funding.
Thou shalt not make a machine in the likeness of a human mind.
>>537276556>We've had several large scale project fail because they have to stick to copyright and data protection laws.Anthropic just pirated it all, then settled in court because by then they were worth billions kek. Better to beg forgiveness than ask permission.>So not only are women given preferential treatment once, this compounds over their entire career.Oh we've all seen the "tech girl boss" tiktok cringe. They're all getting fired here now in the US anyway, same as HR. A lot of these regulations are "make work" bullshit to create jobs for useless but ideologically aligned people.
>>537271536Why do i need these niggers benchmarks, i'm my own benchmark, and you want to believe this crap is a fable 5 match ? lmao, fable was phenomenal.
