Finn Charlotte Thomas
Mapping ‘profundity’ in large language models: A geometric analysis
This project investigates how Large Language Models (LLMs) represent the abstract concept of profundity. By applying multidimensional scaling (MDS) to high-dimensional embeddings from the Llama-3 model, and comparing these with similarity judgments elicited from its post-trained assistant variant, the study examines whether pseudo-profound, profound, and mundane statements align along a latent geometric dimension in semantic space. The work integrates applied mathematics with data science methods, uniting geometry, computation, and cognitive theory to advance understanding of how complex semantic judgments are structured in artificial intelligence.
Finn Charlotte Thomas
The University of Newcastle
Finn Charlotte Thomas is a third-year Bachelor of Psychological Science student at the University of Newcastle with strong interests in computational cognitive science and quantitative methods. She applies mathematical modelling to understand human cognition and is preparing for postgraduate study in psychology and machine learning. She also holds a bachelor’s degree in audio engineering and maintains active interests in music, fashion, and design.
You may be interested in
Thomas Hanyang Zheng
Hidden Markov Models for Analyzing Stress Levels in Working Dogs: A Comparative Study of Data Collection Devices and Frequencies
