A few zeroth-order gradient steps on prompt embeddings, using only forward evaluations and a toxicity scorer, reduce toxic LLM outputs while preserving fluency.
Self-detoxifying language models via toxification reversal
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Test-Time Detoxification without Training or Learning Anything
A few zeroth-order gradient steps on prompt embeddings, using only forward evaluations and a toxicity scorer, reduce toxic LLM outputs while preserving fluency.