Title:	Evaluating Large Language Models as Research Assistants: Potentials and Shortcomings
DNr:	Berzelius-2026-137
Project Type:	LiU Berzelius
Principal Investigator:	Dejan Manojlo Kostic <dmk@kth.se>
Affiliation:	Kungliga Tekniska högskolan
Duration:	2026-05-01 – 2026-11-01
Classification:	10201
Homepage:	https://www.kth.se/blogs/dai/
Keywords:

Abstract

Since their introduction, Large Language Models (LLMs) have shown an astonishing performance in a wide range of tasks. Their undeniable potential has led to their rapid integration across a wide range of domains, including many tasks in research. LLMs now help researchers to find relevant work, polish their ideas, validate their results, and even write their papers. However, this significant reliance on LLMs might raise some concerns. Do these LLMs truly understand the papers we feed to them, do they provide us relevant information, do they guide us in correct path, and can they actually find relevant works for us? In this project, we are looking at the potentials of large language models when working as assistants to the researchers. We will look at different aspects of being a research agent, such as relevant work discovery and information extraction from publications, and hope at the end of this project, we will have a better view of the potentials and shortcomings of LLMs when it comes to research innovations.