1 paper
Christina Lu, Jack Gallagher, Jonathan Michala +2
Large language models can represent a variety of personas but typically default to a helpful Assistant identity cultivated during post-training. We investigate the structure of the…