[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: 1000117911.jpg (666 KB, 1079x1797)
666 KB
666 KB JPG
Rekt
>>
>SARvam
>>
SAARVAM
>>
>>108187156
why paki ser shilling for india saars?
>>
>>108187156
Why is nobody using it then?
>>
Holy shit, India is winning the AI race now? Can someone post the benchmarks for SARVAM? Just how badly does it blow ChatGPT, Gemini and Claude out of the water? How can I use it?
>>
File: .png (1023 KB, 1072x1662)
1023 KB
1023 KB PNG
fake news
>>
>>108187234
Seems like it was unveiled only this Tuesday. Is it even available yet?
>>
>>108187270
Nobody has been using any of the LLMs that have come out of India, and the Indians are fine tuning other peoples shit anyway, and still failing.
They just dont have it in them.
>>
File: .png (150 KB, 960x551)
150 KB
150 KB PNG
>>108187239
>Can someone post the benchmarks for SARVAM? Just how badly does it blow ChatGPT, Gemini and Claude out of the water?
sarvam doesn't compete against those giant models. it only competes against other 100b models.
>>
File: .png (27 KB, 640x289)
27 KB
27 KB PNG
>>108187319
>>
File: 1744777635166222.jpg (66 KB, 1280x720)
66 KB
66 KB JPG
>>108187265
I got banned for asking him about the texture of his sister's vulva :(
>>
>>108187156
>saarVM
I couldn't have picked a better name
>>
>>108187231
Tribal mentality 101:
>me against my brother
>me and by brother against the benchod
>>
>>108187346
u make us proud
>>
>>108187156
All these alternative models train on the output of ChatGPT and Claude.
>>
>>108187696
nope, a foundation model is not a chatbot.
you don't rl a foundation model.
>>
>>108187319
>giving a presentation about a language model
>on-par than
>>
>>108187156
>it's not possible to do this with just millions
>india spends 50+ millions doing it
sounds like altman was right for a change
>>
>>108187319
>>108189535
>>on-par than
kek
>>
>>108187156
I genuinely don't fucking understand what the point of this technology is.
>>
>>108190806
it solves child abuse since everyone can create photorealistic cp without need for actual preteens
>>
>>108190806
automate "content creators"
all it has to do is be as good as tiktok
>>
>>108187327
Top fucking kek, India competing with Miqu 70b and gpt oss (gpt oss is shit)
>>
>>108190806
for LLMs neither does anybody else which is why they're searching for problems to fix that they can actually extract profit from (they will fail). real AI applications will have less to do with natural language processing and a lot more to do with biotech, finance, military, police, etc. basically anything that benefits from mass surveillance, the intention is to build a magic crystal ball they can see the future with. and maybe a few math and science discoveries but nothing revolutionary
>>
>Can you believe how cheap and easy it is to bake a cake if all you have to do is open the oven and take a slice out of another cake that's almost done baking?

I have no problem with distillation or any other use to build on existing models (and in fact, I think its a much more optimal process rather than every proprietary faggot dumping all of Libgen/SciHub/Wikipedia/Youtube etc in to their brand new model redundantly) but lets not pretend that
>We managed to catch up with your ultra expensive model on a fraction of the cost, by uh... basically having your ultra expensive model teach ours everything.
is legitimate
>>
>>108191526
you don't know what distillation is.
distillation requires access to the innards of the teacher model, it’s not possible for a third party to sneakily distill data from a closed-source model like openai's.
>>
File: 1759396513955540.png (267 KB, 937x937)
267 KB
267 KB PNG
>>108187319
>On-par than
>>
>>108187156
The thing with Indians is that you have to publicly shame them in a way that makes them feel nationalistic that's the only way they ever get shit done
>>
>>108187156
Based
>>
The only Indian AI that will work is when it's Actually Indians™. A million monkeys at a million typewriters. Oops. I mean a million jeets at a million chrome boxes answering questions.
>>
>>108190806
Here you go
https://vxtwitter.com/charliebcurran/status/2024585477148627118?s=20

This is worth GPUs being $5k, ram shortage, and even hard drives going up in price
>>
File: waaaaahh.mp4 (1.19 MB, 480x640)
1.19 MB
1.19 MB MP4
>>108187156
in india, everyone works for free.
>>
>>108187327
Isn't GLM 4.5-Air comparable to GLM 4.7 Flash?
>>
>>108190814
Perfect, now we can get rid of everyone under 18 and we've solved child sex abuse
>>
>>108193294
how do you plan to kill all white children?
>>
newer models are cheaper to train because they can just distil from existing models no?
>>
>>108187239
>India is winning the AI race now?
SARVAM has another LLM pass it the answers to the test.
>>
>>108193350
>Training new LLM
>Give it internet access as a tool
>New LLM just learns to ask Claude
>>
>>108190806
Mass surveillance and deepfake propaganda
>>
>>108193303
Poison all dog dicks.
>>
>>108191970
you don't know what modern "distillation" is like, they really do just take traces and train on that finetune style
>>
>>108187346
Ask chatgpt about it
>>
>>108187346
Why would you do that to a gay man, you piece of shit? He must've been traumatized.
>>
File: x screencap thread.webm (1.7 MB, 696x522)
1.7 MB
1.7 MB WEBM
>>108187156
>>
It's probably just deepseek finetuned/retrained on poo



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.