I don't understand the hype around Chinese models at all. I tried using them and they are significantly worse than the OpenAI and Anthropic and Google ones, in particular at understanding intent.
>>541860896OpenAI and Gemini are dogshit. Using that as a performance standard isnt saying much
skill issue
Just prompt ChatGPT to create the best AI
lollmao
>>541861018they're so neutered by kikes that they became basically useless, I ended my subscription to claude once he demanded me to explain what's the intent of me setting up ssh server on my pc and demanded proofs that I own the deviceanyone that still pays for this muttkike garbage instead of going for chink models is clearly sub 89iq nigger
>>541860896My understanding is they're pretty good but still noticeably a level or two below Openai and Anthropic. The Chinese do a lot of viral marketing for their stuff
>>541860896of course they are, they're significantly smaller and are distilled from western frontier models. china is still playing copycat
>>541860896>flashdistilled models are inherently worse than the models they were distilled from seehttps://www.youtube.com/watch?v=QBG6bXiOq40
>>541860896qwen is utter dogshit, all versionsdeepseek is usable, the v4 flash version is decent for its priceopenai etc are way too expensive for their capabilities will be trialing muse soon, we shall see but its a toss up really
>>541860896Try typing in "ching chong ding dong" after each prompt
>>541861660Then it'll answer you in Chinese and you'll have to beg it to translate that to English
>>541861298>copycatof shitty western models? ok lol
>>541860896bro they are communists. they will eat the ox cart full of shit never forget thatif you cant tell some ideologue has written a piece of propaganda i dont know what to tell you. especially chinese are not good at this. theyre always over the top almost kitsch
>>541860896Chinese models are easier to "break" and can provide information you normally wouldn't get from Claude or ChatGPT. This whole which ai is better debate really only comes from consumers since most serious people use multiple AIs
>>541860896No wonder it's worse if you used the flash version. It's been deliberately downgraded to make it more compact.
>>541861913>and can provide information you normally wouldn't get from Claude or ChatGPTthats because they are trained on the outputs from claude and chat gptif claude and chatgpt once gave those answers and then later were patched, the chinese distillations would still have the pre-patch outputs to train on so it doesnt imply that they know more than the american models, its just that whoever made them doesnt care to update their data set to align with the americans censorshipthis is why the white house ordered anthropic to cut access to china and patch their opus model , because amazon alerted them that they were able to extract dangerous outputs from it and those could be used to train dangerous models
>>541862251I totally agree. Didn't mean to make it sounds like Chinese agents know more i was just saying people really only use them because they are easier to break and you can get some restricted info from them since they were trained on Claude (this has been shown in multiple prompts from people who do break them because it starts identifying as Claude)
>>541861211BASEDDeepSeek
So, if they are shit, why are Amerigolems so afraid, literally fear mongering?
>>541860896`>a Claud agent wrote this post
>>541860896>understanding intentmaybe learn to speak english and express your desired better in prompt form. my local deepseek runs like a champ, no kiked data center required. none of the shit AI you named could tell you how many times nigger appears on the average daily stormer article archive.
>>541860896Kimi is a good anthropic alternative and is capable of being based.
>>541866879Fucking kek
>>541860896I think they're more uncensored
Spicywriter already has all the models broken
>Chinabad>Open source bad>tech oligarchy GOOD>dystopian centralized information control GOOD
>>541860896>Falling for chang propagandaDon't be that guy.
This was my very first AI when the China one was free last year. Still blows me away.
>>541861018Gemini is absolute dogshit but chatgpt is fire now
>>541860896Chinese models are obviously aimed at Chinese speakers LLM's that are English speaking focused perform better for English speakers .. it's just common sense
>>541861298Yeah distillation has hard limits especially with actual reasoning. The reality is that we are still 2 years away I think from truly useful local AI.
>>541862251You can finetune models to be dangerous regardless. Straight up post train the refusals out while feeding raw textbooks.
>>541860896The exciting thing about them is they're cheap.The idea is: you use these for your bullshit daily tasks and experimentation. Then when you need the big dick models, they're freely available because you haven't even touched your tokens yet.
>>541861211taiwan is part of china. I don't even need AI to know that. I welcome my new yellow overlords.
>>541866879soldso tired of soulless Ai agent telling me to become better men. I'm sure deepseek got issues with winnie the pooh but my problems are with kikes and jeets
>>541860896They were not trained for 2digit iq users.
>>541861211
1pbtid op makes a threadFuck Ai
>>541872242Fuck you
>>541861846Yes, idiot. Distillation is always behind and always reduces the brainpower of the distilled model.>>541860896It's cheap and can be run locally. For a lot of things, especially if it can do web search, it's good enough and that's sort of the appeal: at some point at some time you could have a Sol or Mythos locally for cheap.I think it's good the chinks are creating innovative pressure on the frontier model companies.
/thread
>>541860896>>541861018gaymini is trained on fucking reddit kek
