A 300M-parameter guardrail model reportedly matches or beats a 7B model on content-moderation benchmarks, but the proposed RL-guided adversarial training loop degrades performance in the paper's own experiments.
safe” or “unsafe
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Lightweight Safety Guardrails via Synthetic Data and RL-guided Adversarial Training
A 300M-parameter guardrail model reportedly matches or beats a 7B model on content-moderation benchmarks, but the proposed RL-guided adversarial training loop degrades performance in the paper's own experiments.