Weronika Sieińska

I am involved in the EU H2020-ICT SPRING research project (Socially Pertinent Robots in Gerontological Healthcare), where the objective is to develop socially assistive robots with the capacity of performing multi-person interactions, and to validate the new technology based on the needs of gerontological healthcare. The robots are supposed to alleviate the workload of hospital staff in ageing societies by providing patients and their companions with hospital-related information.

The robotic platform used in the project is a social humanoid robot called ARI, which was developed by PAL Robotics. The conversational interface built in the robot is based on a Large Language Model (LLM) called Vicuna, which was prompt engineered for the task of providing hospital-related information.

My PhD research project mostly builds on the conversational AI system developed for the SPRING project. I am interested in the safety issues around LLMs for conversational interaction, and I run evaluations of the SPRING system to assess its safety, which is critical in the healthcare domain. I am particularly interested in the problem of LLM hallucination. Unfortunately, the SPRING system often provides its users with believable but false information. To address this issue, I explore methods for LLM hallucination mitigation such as retrieval augmented generation and red-teaming. Additionally, I would like to test these methods with other LLMs such as GPT-4o, GPT-4 Turbo, GPT-3.5 Turbo, Llama 3, and Mistral.