Enhancing and Adapting LLMs for Societal Applications

Speaker: Rohini K Srihari, Professor and Associate Chair, Department of Computer Science and Engineering, University at Buffalo, The State University of New York

Date: 07 January 2026

YouTube link: https://youtu.be/5wHgN6wadYg

The talk on ‘Enhancing and Adapting LLMs for Societal Applications’ by Rohini K Srihari was organised jointly with the Department of Computational and Data Sciences (CDS) at IISc. The attendees were from different departments in IISc, government organisations such as NIELIT, and industry such as Ola Krutrim AI and IBM. A summary of the talk is provided below.

The evolving proficiencies of large language models (LLMs) are leading to their use in a multitude of applications ranging from business solutions, scientific discovery and even in the creative arts.  In this talk, Rohini K Srihari focussed on research challenges related to their use in two specific societal applications.  The first involved assisting users with motor neuron disease such as cerebral palsy or ALS. These users typically use Augmented and Alternative Communication (AAC) devices to communicate due to their physical limitations in speaking. She discussed how LLMs have been integrated into their communication system. The second application involved mental health support where this an increasing need spanning different demographics.  A viable solution to these applications must address key technical challenges within LLMs.

A major focus of this talk was personalisation of LLMs; she discussed personalisation across individuals and groups. A second challenge is strategic planning during goal-oriented conversations. Mental health support conversations tend to be longer, spanning 30 turns or more.  It is necessary to steer LLMs during long, empathetic, multi topic conversations while maintaining strategic conversational goals established by the counselling community. The speaker’s solution involved examining the output of attention units within the models. She discussed the creation of synthetic data sets for both training and evaluation in such solutions as well as a new paradigm for LLM based evaluation of conversations. Finally, she discussed the use of computational resources provided by the Empire AI consortium established by New York State.

MENU