Krutrim SI Designs, the new artificial intelligence venture founded by Bhavish Aggrawal, co-founder of Ola, has introduced a family of multilingual AI models specifically tailored to address the distinctive requirements of the Indian ecosystem. The models, named Krutrim Pro, are designed to operate in 22 Indian languages, emphasizing cultural connect and accessibility at India-first cost structures.
Understanding Krutrim
Krutrim, which means ‘artificial’ in Sanskrit, comes in two sizes. The base model, Krutrim, is trained on an impressive 2 trillion tokens and unique datasets. Its larger counterpart, Krutrim Pro, is set to launch next quarter, boasts advanced problem-solving and task execution capabilities. Aggarwal highlights the importance of India-specific training data, cultural context and cost considerations for successful AI implementation in the country.
Multilingual Capabilities
Krutrim is a groundbreaking AI model capable of understanding and generating content in 22 Indian languages. With a focus on inclusivity, it covers languages such as Marathi, Hindi, Bengali, Tamil, Kannada, Telugu, Odia, Gujrati and Malayalam. The model’s training data includes over two trillion tokens for Indian languages, making it stand out in terms of its extensive linguistic understanding.
Deployment and Early Access
Interested users can now sign up for the base model, with early access scheduled to roll out in batches. The full open release of Krutrim APIs from February 2024. Aggarwal mentions that Ola group companies are already utilizing Krutrim for various internal tasks, including customer support, voice and chat interactions and customer sales calls.
Performance and Comparison
According to senior executives, Krutrim outperforms OpenAI’s GPT-4 in Indian languages, showcasing superior timing and computation. In English, Krutrim reportedly performs better than Meta’s Llama 2 chat but trails behind GPT-4, Google’s Bard and Gemini. The model operates in multiple modes, including text and voice, catering to diverse user preferences.
Future Plans
Looking ahead, Krutrim SI Designs aims to launch a supercomputer within the next two years. The company is also developing AI infrastructure, including indigenous data centers and eventually, server-computing and supercomputers. Prototypes are expected by mid-2024, with a production roadmap for rollout by the end of 2025.
Beyond Models
In addition to the AI models, Krutrim SI Designs plans to create its own chiplets and other AI-focused hardware for data centers. Aggarwal emphasizes that while the model is the soul of their work, infrastructure and silicon form the body, showcasing a holistic approach towards AI development.
Separate Entity and Collaboration
Aggarwal clarifies that Krutrim is a separate entity and not a subsidiary of Ola Cabs or Ola Electric. However, he hints at potential collaboration in terms of data sharing and usage between the three companies.
Important Questions Related to Exams
Q1. What is Krutrim SI Designs focusing on with its AI models?
Sol. Tailoring AI models for the unique needs for the Indian ecosystem, emphasizing cultural connection and India-first cost structures.
Q2. What are the two sizes of Krutrim models introduced?
Sol. Krutrim and Krutrim Pro, with the latter offering advanced problem-solving and task execution capabilities.
Q3. How many Indian languages can Krutrim understand and generate content in?
Sol. 22 languages, including Marathi, Hindi, Bengali, Tamil, Kannada, Telugu, Odia, Gujrati and Malayalam.
Q4. Which month will Krutrim APIs be accessible to developers?
Sol. Krutrim APIs will be accessible to developers in February 2024.
