India Artificial Intelligence Mission (AI, LLM, ML): News, Updates & Discussions

Azaad

Revenge Is The Purest Emotion ~ The Mahabharat
Member
Joined
Jul 1, 2024
Messages
2,475
Likes
10,833

Cybercoolie see Cybercoolie do !

For all his accomplishments , Nilekani is no risk taker nor are his colleagues. He prefers earnings thru arbitrage ONLY. No wonder Infosys is the way it is & InfoMurthy goes on his senile rants at the frequency of once every month / alternate month.
 

Summary:

In an interview with AIM, founder Pranav Mistry revealed that the company generated $4 million in revenue this quarter and expects $20 million for next year.

TWO AI’s SUTRA, a series of multilingual online GenAI models, added a new feather in its cap. The company claims it outperformed GPT-4o, Llama 3.1, and Indic LLMs, including G42’s Nanda, Sarvam’s OpenHathi and AI4Bharat’s Airavata, and leading in over 14 Indian languages.

Unlike other startups, TWO AI targets only big enterprise customers instead of pursuing the consumer market. “Jio is one of our major enterprise customers, and we also work with clients like Shinhan Bank and Samsung SDS in Korea,” Mistry said. He further revealed that the company has started partnering with companies like NVIDIA and Microsoft from a technology perspective, and is working with them as well.

“We are targeting India, Korea, Japan, and some parts of Southeast Asia, like Vietnam, specifically the central region. APAC (Asia-Pacific) is one of the key markets that we are always going to focus on,” Mistry added.

Earlier this year, the company launched ChatSUTRA, a ChatGPT-like chatbot. Mistry shared that the platform currently has over 600,000 unique users.
 

View: https://twitter.com/Gautam_A_k/status/1882238816813031455?s=19


View: https://twitter.com/vaibhavbetter/status/1882245545504498060?s=19


View: https://twitter.com/PrasadSatya10/status/1882419088368279820?s=19


View: https://twitter.com/m_saharia/status/1882426139815432416?s=19

Let me give a ground view from where academia meets industry in a specialized domain. I had a great meeting with an NVIDIA team today, who wants to partner to scale up our Deep Learning models for SAR and Optical satellite applications. As well as some LLM-specific work in climate domain. We are developing a small foundation model for geospatial applications, which piqued their interest. They came to our campus, and we had very detailed technical discussions, which we enjoyed. For the past 1.5 years, we have also been working on a project funded by SONY on LiDAR and Deep learning applications. We meet every month, assess technical progress rigorously, and its a great avenue for mutual learning.

No established Indian company has ever approached us till today (TCS, Reliance - all have presence in geospatial domain). All I see is domestic industry leaders and VCs pontificating on Twitter and fests, and looking for handouts from the government in R&D spending. For all its limitations, government ministries and agencies actually have tremendous appetite for risk.

Lots of vendors want to partner with IITs for working on regular projects, but domestic appetite for cutting edge R&D is minimal. Indian VCs and companies need to have a more realistic understanding of where they stand vis-a-vis Silicon Valley. We need to ask ourselves if we really need another food delivery or makhna packaging company.
 

india/indic-aware LLM in 10 months? very agressive. but even if delayed and ultimately available, it will be a significant breakthrough.


building an Indic LLM will be challenging. Singh says high-quality, labelled datasets in Indian languages are scarce. Data collection and curation, he says, can be both time-intensive and expensive.

R Systems director of AI, Samiksha Mishra, also says building an Indic foundation model tailored to India’s linguistic diversity would be a monumental task. “While datarelated issues are significant, deeper systemic and infrastructural barriers also complicate the devel opment process.” She points out that an Indian LLM must accommodate over 22 official languages and up to 1,600 dialects—an especially difficult prospect.

Samiksha also notes that Indic scripts are morphologically rich and have complex ligatures, which make building models difficult. “We also need to ensure linguistic inclusivity, so that dominant languages don’t overshadow smaller but culturally significant ones,” she says.
 
1738224965314.webp
 
india/indic-aware LLM in 10 months? very agressive. but even if delayed and ultimately available, it will be a significant breakthrough.


building an Indic LLM will be challenging. Singh says high-quality, labelled datasets in Indian languages are scarce. Data collection and curation, he says, can be both time-intensive and expensive.

R Systems director of AI, Samiksha Mishra, also says building an Indic foundation model tailored to India’s linguistic diversity would be a monumental task. “While datarelated issues are significant, deeper systemic and infrastructural barriers also complicate the devel opment process.” She points out that an Indian LLM must accommodate over 22 official languages and up to 1,600 dialects—an especially difficult prospect.

Samiksha also notes that Indic scripts are morphologically rich and have complex ligatures, which make building models difficult. “We also need to ensure linguistic inclusivity, so that dominant languages don’t overshadow smaller but culturally significant ones,” she says.
They should first build a proper English Original LLM like ChatGPT or DeepSeek or Claude instead of all the Scam Wrappers. Those Gandu Dehati Indic lAnGuaGe Bullshit can come later.
 
They should first build a proper English Original LLM like ChatGPT or DeepSeek or Claude instead of all the Scam Wrappers. Those Gandu Dehati Indic lAnGuaGe Bullshit can come later.

isn't it the case that training with english as base and later trying to use it for desi language training and applications takes exponential number of tokens?
 


folks who are interested in public-private partnership model related stuff might want to watch the video with extra care. what's being said there is execution of a desi model, which is different from both murican and european models of digital infrastructure. a techno-legal model with clarity on guardrails already in place from get go.

these are sort of things that get into text books and gyanis write their thesis on later.
 
I would not discard it like that. A ~19k strong GPU cluster is nothing to sneeze at. Hopefully, more to follow soon.

was talking about the llm ... 'our own foundational model'

these people took a open source OS project , then simply changed the name text in source to bharos . thats their level of vision.
 
was talking about the llm ... 'our own foundational model'

these people took a open source OS project , then simply changed the name text in source to bharos . thats their level of vision.
Bruh, we all knew it was nothing more than a mere hobby project of a bunch of IIT guys (the Minister was visiting the IIT M, met a series of startup founders/interns and let his PR team craft social media posts on them - he met some 'genuine' guys too). The mediawallahs (even the gora ones) and the Indian netizens picked one of those posts up, spread all sorts of BS and made a mess of it.

I think it is important to differentiate between PR blitz and actual gormint $$$ backed initiatives. Apparently, they are planning to provide some viability gap funding for start-ups and looking to procure more GPU units.
 
Bruh, we all knew it was nothing more than a mere hobby project of a bunch of IIT guys (the Minister was visiting the IIT M, met a series of startup founders/interns and let his PR team craft social media posts on them - he met some 'genuine' guys too). The mediawallahs (even the gora ones) and the Indian netizens picked one of those posts up, spread all sorts of BS and made a mess of it.

I think it is important to differentiate between PR blitz and actual gormint $$$ backed initiatives. Apparently, they are planning to provide some viability gap funding for start-ups and looking to procure more GPU units.

ok then .

thing about bharos is that media didnt hype it . iit madras made those claims. only reason they got exposed is because a developer made the source code accidently public. honestly have a hard time beliving any of these people now.
 
tweet about hardware needed to run single instance of deepseek

5-7 lakh cost , 768 gb ram , 2 cpus + deepseek is a 700 gb download :oops:

chatgtp probably has same hardware costs , and to think its free


View: https://x.com/carrigmat/status/1884244369907278106

the above spec is true for the highest capable variant.

i have the deepseek 8B variant running perfectly fine on my 16 gig RAM VM occupying just 5 gig disk space.

so, depends on the variant you want to use.
 
ok then .

thing about bharos is that media didnt hype it . iit madras made those claims. only reason they got exposed is because a developer made the source code accidently public. honestly have a hard time beliving any of these people now.
My rule of thumb is that I am generally long on gormint $$$ backed focused initiatives like PLI, Semicon India, India Stack etc. Incumbent gormint has a good past track record of delivering on this front.

If IIT M indeed started the PR blitz then it was just plain stupid. They had a number of hard tech start-ups to boast on (Ather, Agnikul, Hyperverge, Uniphore, QuNu Labs came out of their incubator).
 

Latest Replies

Featured Content

Trending Threads

Back
Top