Capabilities and limitations of LLMs: from viewpoint of our research in dialog system, medical, legal and political applications and in human-like NLP system

by Prof. Yoshinobu Kano (Faculty of Informatics. Shizuoka University, Japan)

Thursday 15 Jun 2023, 15:00 → 17:00 Europe/Lisbon

131 (CES)

Bio:
Yoshinobu Kano
Associate Professor, Faculty of Informatics. Shizuoka University, Japan
BS in physics (2001), MSc (2003) and PhD (2011) in information science and technology from the University of Tokyo, respectively.
A research associate in University of Tokyo (2009), JST PRESTO researcher (2011), an associate professor (PI) in Shizuoka Univerity (2014-).
He is interested in human-like natural language processing and its applications in medical, legal, political and conversational issues.
http://kanolab.net/kano/index.en.html

 

Abstract:
Recent advantages of the LLMs (Large Language Models, such as ChatGPT) affect not only to the NLP community but the global society of the ordinary people.
A question is that the LLMs could solve what sort of NLP problems, though it is still under investigation. Another question is that to what extent the LLMs are "similar" to the humans -- they apparently differ in its architecture and the training data size.
I introduce a couple of related NLP projects that I organize, such as automatic medical diagnosis, AIWolf project (dialog system in conversation games), Legal NLP (legal bar exam solver), and SNS analysis project for public opinions, discuss the ability and limitations of the LLMs with future directions.