Safety Beyond the Interface: Detecting Harm via Latent States in Large Language Models
Lightweight 12.6M-parameter probes on LLaMA-3.1-8B activations detect harmful prompts with F1 up to 99%, matching guard models 1000x larger at lower cost.
1 stories, newest first · one card per story
Lightweight 12.6M-parameter probes on LLaMA-3.1-8B activations detect harmful prompts with F1 up to 99%, matching guard models 1000x larger at lower cost.
Today 1.6K readers · 2.5K views · 785 reads · all time 189K views