Search Results for author:
Found 3 papers, 1 papers with code
Auditing language models for hidden objectives
First, we conduct a blind auditing game where four teams, unaware of the model's hidden objective or training, investigate it for concerning behaviors and their causes.
Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teaming
Large language models (LLMs) are vulnerable to universal jailbreaks-prompting strategies that systematically bypass model safeguards and enable users to carry out harmful processes that require many model interactions, like manufacturing illegal substances at scale.
Sparse Autoencoders Find Highly Interpretable Features in Language Models
One hypothesised cause of polysemanticity is \textit{superposition}, where neural networks represent more features than they have neurons by assigning features to an overcomplete set of directions in activation space, rather than to individual neurons.
