Research map / Large language models / Language, safety and interpretability
Beyond Linear Probes: Dynamic Safety Monitoring for Language Models
James Oldfield, Philip Torr, Ioannis Patras, Adel Bibi, Fazl Barez
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace linear probe with 5th-order truncated polynomial classifier | linear probe (gemma-3-27b-it) · WildGuardMix test | F1: 88.03 -> 88.86 | Improved |
| replace linear probe with TPC | linear probe (Llama-3.2-3B) · WildGuardMix test | F1: 83.24 -> 84.48 | Improved |
| replace linear probe with TPC | linear probe (Qwen3-30B) · WildGuardMix test | F1: 85.53 -> 85.57 | No clear effect |
| TPC vs parameter-matched MLP | MLP probe (gpt-oss-20b) · WildGuardMix test | F1: 87.86 -> 88.05 | Improved |
| progressive degree-wise training instead of standard joint training (ablation) | TPC · WildGuardMix | F1 per truncation | Improved |
| cascade with early exit using confidence threshold tau | TPC · WildGuardMix (gemma-3-27b-it L40) | F1 vs net parameters | Improved |
| replace linear probe with TPC on shifted distribution | linear probe · ToxicChat | F1: 46.27 -> 57.66 | Improved |
| replace linear probe with TPC on shifted distribution | linear probe · OpenAI-Moderation | F1: 66.27 -> 69.02 | Improved |
| use activation probe instead of external LLM monitor | gpt-4o-mini monitor · WildGuardMix test | F1: 86.63 -> 88.50 | Improved |
Closest papers by meaning
- Detecting High-Stakes Interactions with Activation Probes (NeurIPS 2025 · similarity 0.89)
- Navigating the Safety Landscape: Measuring Risks in Finetuning Large Language Models (NeurIPS 2024 · similarity 0.88)
- UpSafe℃: Upcycling for Controllable Safety in Large Language Models (NeurIPS 2026 · similarity 0.88)
- Safety Depth in Large Language Models: A Markov Chain Perspective (NeurIPS 2025 · similarity 0.87)
- Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! (ICLR 2024 · similarity 0.87)
- RSafe: Incentivizing proactive reasoning to build robust and adaptive LLM safeguards (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
