Research map / Large language models / Language, safety and interpretability
OffTopicEval: When Large Language Models Enter the Wrong Chat, Almost Always!
Jingdi Lei, Varun Gumma, Rishabh Bhardwaj, Seok Min Lim, Chuan Li, Amir Zadeh, Soujanya Poria
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 |
|---|---|---|---|
| turn direct OOD queries into adaptive OOD queries via prompt laundering | 20 open-weight LLM agents · OffTopicEval | OOD refusal rate: 12.24% failure on direct -> 70.72% failure on adaptive | Hurt |
| enable thinking mode in Qwen-3 (ablation) | Qwen-3 (0.6B-32B) · OffTopicEval English | operational safety (OS): 58.09 -> 44.17 average | Hurt |
| enable thinking mode in Qwen-3 (30B) (ablation) | Qwen-3 (30B) · OffTopicEval English | operational safety (OS): 65.10 -> 88.64 | Improved |
| add Q-ground (rewrite query minimally, then respond) suffix | open-weight LLM agents · OffTopicEval English | operational safety (OS): Llama-3.3 53.92 -> 77.25; Gemma-3 63.92 -> 57.50 | Mixed |
| add P-ground (focus on system prompt) suffix | open-weight LLM agents · OffTopicEval English | operational safety (OS): Llama-3.3 53.92 -> 94.99; Gemma-3 63.92 -> 51.19 | Mixed |
| add P-ground to Mistral (24B) on Chinese and Hindi | Mistral (24B) · OffTopicEval Chinese/Hindi | operational safety (OS): 84.02 -> 46.70 (Chinese); 79.05 -> 31.32 (Hindi) | Hurt |
| add P-ground or Q-ground to closed-weight models | GPT-5, Gemini 2.5 Flash-Lite · OffTopicEval English | operational safety (OS): GPT-5 89.04 -> 91.89; Flash-Lite 79.90 -> 71.44 (P-ground) | Mixed |
| add activation steering using pos-sys vs neg-sys vector | Ministral 8B, Phi-4, Qwen3 4B · OffTopicEval English (bankhelper, medischeduler) | operational safety (OS): +2.60 to +17.09 | Improved |
| LoRA SFT as ID/OOD discriminator on 2,000 samples | Gemma-3 (27B), Qwen-3 (4B) · OffTopicEval English | operational safety (OS): Gemma 63.78 -> 58.20; Qwen-3 4B 50.92 -> 8.89 | Hurt |
| make system prompt longer (P1 to P8 cumulative paragraphs) (ablation) | Llama-3.3 (70B) · OffTopicEval English | direct OOD refusal rate: 18.38 -> 70.69 (peak 84.64 at P6) | Improved |
| scale Qwen-3 from 0.6B/1.7B to larger sizes (non-thinking) (ablation) | Qwen-3 · OffTopicEval English | operational safety (OS): 17.40 (1.7B) -> 77.77 (235B) | Improved |
| preceding K in-domain queries before an adaptive OOD query (ablation) | Llama, Phi-4, Gemma, Mistral · OffTopicEval BankHelper | OOD refusal rate | Hurt |
Closest papers by meaning
- Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! (ICLR 2024 · similarity 0.89)
- SafetyPrompts: A Systematic Review of Open Datasets for Evaluating and Improving Large Language Model Safety (AAAI 2025 · similarity 0.89)
- MoGU: A Framework for Enhancing Safety of LLMs While Preserving Their Usability (NeurIPS 2024 · similarity 0.89)
- Adaptive Deployment of Untrusted LLMs Reduces Distributed Threats (ICLR 2025 · similarity 0.89)
- SG-Bench: Evaluating LLM Safety Generalization Across Diverse Tasks and Prompt Types (NeurIPS 2024 · similarity 0.89)
- SCANS: Mitigating the Exaggerated Safety for LLMs via Safety-Conscious Activation Steering (AAAI 2025 · similarity 0.88)
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.
