India's LLM Moment is Here

Can't read or see images? View this email in a browser

THE WEEKLY NEWSLETTER OF AIM. Monday, Dec 18, 2023 | Was this email forwarded to you? Sign up here By Amit Naik This week witnessed a slew of announcements from Indian companies and startups launching their large language models (LLMs), including Google-backed CoRover’s BharatGPT, Khosla Ventures-backed Sarvam.ai’s OpenHathi, Microsoft-backed Kissan AI’s Dhenu, and Ola’s Krutrim. Of all the announcements made this week, Ola’s Krutrim stood out like none. Unlike other companies, which are mostly building LLMs on top of GPT-4 or Llama 2, Krutrim is going the extra mile to, not just build proprietary LLMs from scratch, but also develop the silicon and the infrastructure layers – all in-house.  The first version of Krutrim supports over 20+ languages, including Hindi, Marathi, Bengali, Tamil, Kannada, Telugu, Odia, Gujarati, and Malayalam (not Malayali). The model is trained on 2 trillion tokens. AIM estimates the model size to be between 7-13 billion parameters.  Bhavish Aggarwal claimed that the team built the first version of Krutrim in just three months. This is quite unrealistic, compared to other players in the space. Elon Musk, for instance, took nearly four months to build Grok and xAI is still grappling with GPU shortages.  But Ola’s Aggarwal, a rebel with a cause, promised everyone that Krutrim would be followed by their first multimodal Krutrim Pro by the next quarter. However, there was no mention of buying additional GPUs, besides sharing its plans to build everything locally. Zoho waits for the dust to settle: Zoho chief Sridhar Vembu previously told AIM that it is working on proprietary LLMs, preferably smaller and domain-specific models, without delving into the timelines.  “We are waiting for the dust to settle,” said Vembu, in a recent interview, sharing that they will continue to increase their GPU capacity to run AI models. He said, for now, Zoho is working on developing its infrastructure in-house instead of relying on hyperscalers, as it would work out cheaper in the long run.  Tech Mahindra, which was the first company to lock horns with the likes of OpenAI with the launch of Project Indus, is expected to release its LLM for Hindi and its 37 dialects by the end of December or early January.  Meanwhile, OpenAI eyes India at a time when India-focused LLMs began to rise. The company is looking at setting up an Indian office soon and has appointed several Indian-origin leaders in advisory roles, bringing OpenAI execs to the country. Read the full story here. AI Forum for India. Our Discord Community for AI Ecosystem, In collaboration with NVIDIA. Join Today >> Top Stories of the Week >>  OpenAI to Release GPT-4.5 Soon While Sam Altman has put an end to the rumours surrounding the release of GPT-4.5 with a simple ‘nah’ to the query, ‘gpt4.5 leak legit or no? on X, there is a stronger reason to believe that OpenAI cannot afford to take a break, with competitors (the likes of Google’s Gemini, Anthropic, Mistral AI, xAI’s Grok, and others) constantly breathing down its neck. What will happen to OpenAI if the competitors beat GPT-4? Read to find out.  “Kids, Papa’s Here,” Says Intel Chief Pat Gelsinger Intel is back with a bang! The company that provided the world access to personal computers is now on a mission to bring AI to everyone.  Super excited, Pat Gelsinger was all over the place at its recent event ‘AI Everywhere’ held at the Big Apple, New York City, flaunting the latest Gaudi3 processor, alongside unveiling next-generation AI PCs and upcoming nodes, with Intel 7, Intel Core Ultra and more. Check out Intel’s silicon prowess here. Meet the Researcher Building Real Robots for Real Household Chores Lerrel Pinto, the 31-year-old computer scientist, is working on making robots cool again. In an exclusive interview with AIM, Pinto discusses the challenges and slower progress in robotics compared to other AI fields and highlights the importance of data diversity and relevance in the development of robotic technologies. Read the full interview here.  AIM VIDEOS >> India boasts a staggering Redis user base of approximately 12 million downloads per day, securing its position as the third-largest adopter on the planet, closely trailing behind the technology giants, the United States and China. “Every month, half a billion people use Redis Enterprise in India alone,” co-founder and CTO Yiftach Shoolman told AIM during his recent visit to India. Click here to read the full interview. AIM EVENT >> DevPalooza 2.0 After the overwhelming success of the first edition of DevPalooza, AIM is back with another edition of the most-awaited, invite-only gathering, scheduled to be held on January 6, 2024, at our vibrant office in HSR, Bengaluru. The event is a beacon for tech enthusiasts, developers and AI aficionados.  What are you waiting for? Book your slots today. AIM SHOTS >> IISc and Bengaluru City Traffic Police recently collaborated to leverage data-driven AI solutions to analyse and alleviate traffic challenges in the bustling metropolis. Microsoft released Phi-2, a 2.7 billion-parameter small language model showcasing exceptional reasoning and language understanding abilities. Learn more here.  After AMD announced at its Advancing AI event that its MI300X is faster than NVIDIA’s H100 for inference, NVIDIA has rebuffed AMD’s claim. Stability AI has officially announced its Stability AI Membership plans.  LangChain now lets you integrate the Gemini Pro API into its platform, enabling multimodal functionalities within its environment.  Download our mobile app Access our stories, video content and events on the go. Stay Connected info@analyticsindiamag.com © 2023 Analytics India Magazine Pvt Ltd and/or its affiliates. All rights reserved. For more information, email info@analyticsindiamag.com

This email was sent by support@analyticsindiamag.com to alexvarboffin.abbb@blogger.com Not interested? Unsubscribe | Manage Preference | Update profile Analytics India Magazine | 280, 2nd floor, 5th Main, 15 A cross, Sector 6, HSR layout Bengaluru, Karnataka 560102