ChatGPT has transformed the world from what it used to be. The extravagance of the use cases today proves we are amidst a technological revolution. ChatGPT roots back to Large Language Model (LLM), that enabled businesses to automate tasks.  

To analyze the scope of large language models in India and evaluate the current use cases, Time of India organized a panel discussion on ‘Large Language Models: Does India need to build one.’ Abhishek Singh, President and CEO of NeGD; Monish Darda, Co-founder and CTO, Icertis; Srikanth Velamakanni, Co-founder of Fractal Analytics; and Abhinav Agarwal, Co-founder and CEO of Fluid AI represented the panel. The discussion, which went live on Facebook, was mediated by Times of India’s Sujit John and Akhil George. 

The term AI was coined in 1956. Even after 50 years, AI was about creating and following the rules. However, times began to change when Deep Learning gained popularity after 2010. Unlike earlier models, which function using supervised, unsupervised or reinforcement learning, LLMs work with semi-supervised learning. For instance, when a text is provided, it will analyze the first few words and tries to predict the next word. LLMs can do this with large volumes of text and generate output. “They are largely parameterized with an incredible amount of data”, said Srikanth Velamakkani.  

LLMs in India 

LLMs exist in the world in the form of applications. However, in an Indian context, cultural diversity is one of the biggest challenges we face. “The services should be available in all Indian languages, even for people does not have access to technology”, said Abhishek Singh, talking about LLMs in India. “To ensure digital inclusion, we need to enable services through voice in all languages”, he added. 

Using LLM tools, we can train the models through which the public can communicate. “We are now implementing a project called Bhashini in which we crowd-source large datasets. People are contributing to the datasets in their language. These datasets are used in LLMs to build tools for translations. These models are published in our platform, which will be accessible to the public”, said Abhishek Singh. 

The developers of Bhashini have integrated it with ChatGPT and WhatsApp APIs to develop a tool called Jugalbandhi, which is now being tested. With Jugalbandhi, a query can be raised on WhatsApp voice notes in their native language, and LLMs can provide answers in the same language.  

These models will be a huge win for India’s governance. However, ensuring sufficient number of datasets in all languages is one of the bottlenecks the developers face. 

The use cases 

Models like ChatGPT provide significant help in contract management. “No transaction in a company happens without a contract. When the intent is realized, we can optimize the outcome. Every parameter that makes up a business goes into a contract, making it difficult to understand. LLMs starting with basic NLPs have been a cornerstone for contract management”, said Monish Darda. 

 The LLMs can summaries, translate, categorize, intent and sentiment massive, consuming contracts in a simple manner which is useful for the public.  

“LLMs can act like a co-pilot to the employees in an organization”, said Abhinav speaking about how Fluid AI uses LLMs in their activities. From trainees to experienced salesperson, there is one AI assistant who safeguards all relevant knowledge about the organization. Moreover, with multimodal capabilities, LLMs can take in images and visuals along with images. “It is like LLMs have eyes and ears”, Abhinav added. 

“After using ChatGPT, efficiency of our coders has gone up”, stated Abhishek Singh. However, there will be jobs that will become extinct. Therefore, we need to find means to reskill and upskill them. 

Accessing data to develop a model is one of the biggest challenges for a country like India. From the issues of legalities, knowledge of what to access and what not to access, and even the availability of required datasets, persists. “Projects like Bhashini are phenomenal because you are getting consent from the people for using the datasets for LLMs”, stated Abhinav. 

Just like the wide benefits of LLMs, there is indeed another side which causes negative implications. “AI is like nuclear energy, it has to be regulated, but it also has to do public good”, added Monish.