Leo Schwinn

New methods improve language model safety while keeping functionality intact.

A study on the resilience of Graph Transformers against adversarial attacks.

A look at activation steering for improving AI memory handling.