Knowledge representations in large language models
Dnr:
NAISS 2023/22-716
Type:
NAISS Small Compute
Principal Investigator:
Richard Johansson
Affiliation:
Göteborgs universitet
Start Date:
2023-06-30
End Date:
2024-07-01
Primary Classification:
10208: Language Technology (Computational Linguistics)
Allocation
Abstract
In this project, we will investigate knowledge representation in large language models. We are going to extract representations from open-source LLMs such as Falcon, Orca etc. and carry out various statistical investigations on these representations. This specific research is a sub-project of a larger WASP project involving researchers from GU, Chalmers, and Linköping university.