AI Safety & Evaluation

Project One Liner: Coming up with evaluation frameworks and understanding model learning representations w.r.t safety

Status: ongoing

Project Theme: safety

Project Areas:

Team: Gokul S Krishnan, Balaraman Ravindran

Collaborators: Precog Labs, IIIT Hyderabad

Short Description: As Large Language Models (LLMs) have been increasingly used in various sectors and applications, it is important to evaluate and improve safety aspects. Current evaluation methodologies for AI models/systems standardized mechanisms that can be applied across diverse stages of AI lifecycle. Several gaps hinder the capacity of both developers and deployers to assess the performance, fairness, and other safety aspects of AI solutions. For solving these issues, we work on designing evaluation frameworks for AI models and systems. We also focus on understanding model learning representations w.r.t AI safety aspects and we expect that this understanding will help us improve AI models in terms of trustworthiness.

Back to Projects