4 papers
Subliminal Signals in Preference Labels
Isotta Magistrali, Frédéric Berdoz, Sam Dauncey +1
As AI systems approach superhuman capabilities, scalable oversight increasingly relies on LLM-as-a-judge frameworks where models evaluate and guide each other's training. A core as…
You Can Learn Tokenization End-to-End with Reinforcement Learning
Sam Dauncey, Roger Wattenhofer
Tokenization is a hardcoded compression step which remains in the training pipeline of Large Language Models (LLMs), despite a general trend towards architectures becoming increasi…
Double Descent as a Lens for Sample Efficiency in Autoregressive vs. Discrete Diffusion Models
Ahmad Fraij, Sam Dauncey
Data scarcity drives the need for more sample-efficient large language models. In this work, we use the double descent phenomenon to holistically compare the sample efficiency of d…
Approximations to the Fisher Information Metric of Deep Generative Models for Out-Of-Distribution Detection
Sam Dauncey, Chris Holmes, Christopher Williams +1
Likelihood-based deep generative models such as score-based diffusion models and variational autoencoders are state-of-the-art machine learning models approximating high-dimensiona…