26-February-2024-Editorial

BharatGPT GROUP UNVEILS HANOOMAN

The BharatGPT group, led by IIT Bombay and seven other premier Indian engineering institutes, is set to launch its ChatGPT-like service named ‘Hanooman.‘

Developed in collaboration with Seetha Mahalaxmi Healthcare (SML), the initiative is backed by Reliance Industries Ltd and the Department of Science and Technology.

Generative Pre-trained Transformers (GPTs):

• Definition: GPTs are large language models (LLMs) utilizing transformer neural networks to generate human-like text.
• Training Data: They are trained on extensive unlabelled text data from the internet, enabling them to comprehend and generate coherent and contextually relevant text.
• Applications: GPTs can be fine-tuned for various tasks, including language generation, sentiment analysis, language modeling, machine translation, and text classification.
• Self-Attention Mechanism: GPTs employ self-attention mechanisms to focus on different parts of the input text during processing, allowing them to capture more context and enhance natural language processing (NLP) performance.

Large Language Models (LLMs):

• Functionality: LLMs use deep learning techniques to process large volumes of text, understanding structure and meaning through extensive training.
• Learning from Text: They process vast text datasets, such as Wikipedia and OpenWebText, to identify meanings and relationships between words, becoming smarter with increased training data.

ChatGPT:

• Definition: ChatGPT is an advanced natural language processing (NLP) model, a variant of GPT-3, developed by OpenAI for generating human-like responses.
• Training: It is trained on extensive text data to provide technical, jargon-free responses, fostering natural dialogue between users and the virtual assistant.

Hanooman:

• Languages: Hanooman responds in 11 Indian languages initially, with plans to expand to more than 20 languages.
• Multimodal AI Tool: It goes beyond a traditional chatbot, functioning as a multimodal AI tool capable of generating text, speech, videos, and more in multiple Indian languages.
• Application Fields: Designed for healthcare, governance, financial services, and education, it includes a customized version called VizzhyGPT, fine-tuned for healthcare using extensive medical data.
• Model Sizes: The series comprises AI models ranging from 1.5 billion to an impressive 40 billion parameters, indicating its diverse capabilities and scalability.