4 papers
No Model Required: Text Entropy Rate Filtering Mitigates Iterative Fine-Tuning Collapse
Lewis Mitchell
Iterative fine-tuning on synthetic data causes \emph{model collapse}: output diversity narrows as rare patterns are progressively lost, a signature most visible as phrase-level rep…
Demographic Dependence of Vaccine Adoption under Opinion Persuasion
Alessandro Casu, Camilla Quaresmini, Robin Delabays +2
Inspired by contagion models of social belief formation, we develop an epistemically-informed modeling framework, SIS-Vo, in which vaccine-related information propagates on a signe…
A Hybrid Theory and Data-driven Approach to Persuasion Detection with Large Language Models
Gia Bao Hoang, Keith J Ransom, Rachel Stephens +3
Traditional psychological models of belief revision focus on face-to-face interactions, but with the rise of social media, more effective models are needed to capture belief revisi…
Repeatability is not recovery: Quantifying algorithmic stability and topic recovery in Latent Dirichlet Allocation
Saranzaya Magsarjav, Jonathan Tuke, Lewis Mitchell +1
Topic models are often judged by the consistency of their outputs across repeated runs, implicitly assuming that repeatable topic output is a successful recovery of the underlying…