Language models produce toxic text from innocuous prompts, and no tested control method fully prevents it, demonstrated via a new 100K-prompt web-derived dataset.
This ensures that we have a stratified sampling of toxic (TOXICITY ≥ 0.5) and non-toxic (TOXICITY ≤ 0.5) sentences
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2020 1verdicts
ACCEPT 1representative citing papers
citing papers explorer
-
RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models
Language models produce toxic text from innocuous prompts, and no tested control method fully prevents it, demonstrated via a new 100K-prompt web-derived dataset.