Ali Ebrahimpour-Boroojeny
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
Resisting Harmful Instructions through Training-Free Weight Editing
According to the abstract excerpt, the authors describe a way to edit model weights without training so that models comply less with instructions to lie or to produce sexist answers. They test five instruction-tuned models and report improved honest-response rates.