Ola CEO Bhavish Aggarwal on December 15, 2023 announced Krutrim, a family of large language models (LLMs) trained on Indian languages to cater to “1.4 billion India.” During the launch event of the Artificial Intelligence (AI) models that was livestreamed on YouTube, Aggarwal said that the AI model will improve the country’s labour and capital productivity, push Indian technology towards a non-linear path and make India a leader in scientific discoveries.
“All AI models are trained largely in English. Current AI models can’t capture India’s culture, knowledge and aspirations given the multilingual context. AI needs to be trained on unique datasets specific to us. To top if off, it needs to be accessible to India,” said Aggarwal.
Two variants of Krutrim: The word ‘Krutrim’ means artificial in Sanskrit. The family of models can be broken into two larger groups: base models and pro models. The base model is trained on 2 trillion tokens, similar to ChatGPT 4 in terms of capabilities and reasoning. The base model can also write in 10 Indian languages and understand 20 languages.
Krutrim Pro is the multi-model that supports text, speech, video and is expected to have more sophisticated problem solving and task execution capabilities. The company expects the pro model to come out in the next quarter.
“In the future, Krutrim can also be able to manage large customer support for up and coming startups and businesses,” said Ravi Jain, Chief Marketing and Revenue Officer, Ola, during the event.
Krutrim using homegrown AI chip: As per the LLM website, the model was developed and manufactured using an in-house AI chip ‘Silicon’.
This is the first images of the layout of first Indian silicon AI chip that can be shipped to manufacture our own home grown AI chip – soon to be developed! kudos to the team led by Sambith@IndianTechGuide pic.twitter.com/PYtleDRPQh
— Chandra (tckb) (@this_is_tckb) December 15, 2023
Currently, NVIDIA is the biggest company in the AI chip market. The demand for NVIDIA’s Graphic Processing Units (GPUs) is such that small businesses cannot afford the chips, even if they make it to the company’s waiting list. If Ola manages to succeed with the homegrown AI chip, it’s likely to create quite the buzz in the market.
Ola claims to outperform existing LLMs: Krutrim also claims to outperform GPT 4 in terms of Indic language performance. The following is a chart shared by the company, comparing Krutrim and GPT 4.
#KrutrimAI vs #GPT4
the GPT4 bar is pretty light – but its right next to itKrutrim outperformed the GPT4 for Indian context. Evaluation was done by linguist experts which compared the results of data by the prompts fed into both models. pic.twitter.com/52XiRfRXSP
— Chandra (tckb) (@this_is_tckb) December 15, 2023
Similarly, Krutrim claims to outperform Llama 2 in terms of qualitative evaluation and reasoning capabilities.
#Krtopensdmodel pic.twitter.com/ze0QjfnmZU
— Chandra (tckb) (@this_is_tckb) December 15, 2023
Also Read:
- ONDC November Round-Up: Paytm, Ola, Metro Tickets, WhatsApp Chatbot, And More
- Elon Musk’s XAI Launches AI Chatbot Grok That Has Access To Real-Time Info From X
- Snapchat’s ‘My AI’ Chatbot In Hot Water With UK’s Information And Communication Officer Over Privacy Concerns
- Half Measures For AI Safety, Bing Chatbot’s Use By Children, And Other Key Issues Discussed In The US Subcommittee Hearing On AI
STAY ON TOP OF TECH NEWS: Our daily newsletter with the top story of the day from MediaNama, delivered to your inbox before 9 AM. Click here to sign up today!