Table of Contents
OpenAI CEO throws shade, but India’s AI Uprising! Can desi tech giants build the next ChatGPT? We dig into the quiet revolution brewing in India’s LLM labs.
Bhavish Aggarwal, the co-founder of Ola, has ventured into the Indian LLM arena with his latest AI venture, Krutrim Si Designs. Introducing their first product line, Krutrim, a collection of multilingual AI models that derives its name from the Sanskrit word for artificial.
Krutrim: Spearheading India’s AI Uprising with Multilingual Mastery
Introducing Krutrim, available in two distinct sizes: the standard model, which has been meticulously trained on an impressive 2 trillion tokens and exclusive datasets, and the remarkable Krutrim Pro, a larger multi-model version set to launch in the upcoming quarter. The Krutrim Pro is specifically crafted to tackle complex problem-solving and execute tasks with utmost proficiency. Get ready to experience the next level of advanced India’s AI Uprising.
LLMs rely on tokens as the fundamental units for handling and expressing textual information. Imagine tokens as the vital components that empower LLMs to comprehend and produce language that closely resembles human communication.
Krutrim is not just known for its massive size, but also for its remarkable proficiency in multiple languages. With the ability to comprehend 22 Indian languages and produce outputs in 10, including Marathi, Hindi, Telugu, Kannada, and Odiya, it truly stands out. In fact, the company boldly asserts that India’s AI Uprising models surpass OpenAI’s GPT-4 when it comes to handling Indic languages.
OpenHathi-Hi-v0.1: From Llama Spark to Indic Language Revolution
Sarvam AI, India’s AI Uprising startup, caused quite a stir recently when they unveiled OpenHathi-Hi-v0.1, the inaugural Hindi LLM in the OpenHathi lineup. This groundbreaking creation, developed using Meta AI’s Llama2-7B framework, boasts an impressive performance comparable to that of GPT-3.5 for Indic languages. The release of OpenHathi-Hi-v0.1 has truly set the stage for a new era in India’s AI Uprising.
Sarvam AI takes a distinctive approach by extending Llama2-7B’s tokenizer with a 48,000-token extension and implementing a two-phase training process. In the initial phase, the focus is on embedding alignment, which aligns randomly initialized Hindi embeddings. The second phase involves bilingual language modeling, where the model is trained to attend cross-lingually across tokens.
Vivek Raghavan and Pratyush Kumar, the founders of a young startup that is only five months old, went above and beyond in assessing OpenHathi’s capabilities in Natural Language Generation (NLG) tasks. They made sure that its performance was not limited to just standard tasks, but also had practical applications in the real world.
In order to enhance its base model, the startup joined forces with KissanAI and utilized conversational data collected from interactions between farmers and a GPT-powered bot in various languages.
BharatGPT: CoRover Makes Indian Languages Sing through AI
CoRover.ai, a Bengaluru-based company, recently introduced its very own language model called BharatGPT. Developed in partnership with Bhashini, a National Language Translation Mission (NLTM) under the Ministry of Electronics and Information Technology (MeitY), BharatGPT is designed to cater to the needs of over 12 Indian languages.
CoRover is a leading provider of India’s AI Uprising virtual assistants, offering a wide range of chatbots, voice bots, and video bots to various esteemed organizations like IRCTC, LIC, IGL, KSRTC, NPCI, and the Government of India. With an impressive user base of over 1 billion, CoRover is revolutionizing the way people interact with technology.
CoRover asserts that BharatGPT enables developers and business users to effortlessly craft multilingual Virtual Assistants with text and voice capabilities in a matter of seconds.
Zoho, the Indian software-as-a-service (SaaS) powerhouse, made headlines in June when it unveiled its ambitious project to develop its own LLM, a cutting-edge technology akin to OpenAI’s GPT and Google’s PaLM
Zoho Forges Small but Mighty AI: Solving Domain Dilemmas for Customers
At the CNBC-TV18 and Moneycontrol Global AI Conclave held on December 16, Sridhar Vembu, the visionary founder of Zoho, revealed an exciting development. Zoho is currently dedicated to crafting smaller AI models, ranging from 7 billion to 20 billion parameters, with the sole purpose of tackling specific domain challenges faced by their esteemed customers. This innovative approach ensures that Zoho’s solutions are tailor-made and highly effective in addressing the unique needs of their clientele.
The SaaS titan has just revealed a collection of 13 cutting-edge AI extensions and integrations for its applications, all fueled by the incredible ChatGPT technology.
