conftrace_

harmless generation

1 paper

Explore in graph

Co-occurring keywords · papers mentioning both

attention layer (1) attention manipulation (1) toxicity reduction (1) language model detoxification (1) representation steering (1) steering vector (1) model safety (1)

Papers

Self-Detoxifying Language Models via Toxification Reversal EMNLP 2023