Research map / Large language models / Instruction and fine-tuning
SafeSci: Safety Evaluation of Large Language Models in Science Domains and Beyond
Xiangyang Zhu, Yuan Tian, Qi Jia, Kaiwei Zhang, Zicheng Zhang, Chunyi Li, Zijian Chen, Lu Sun, Kaiyuan Ji, Renrui Zhang, Wei Sun, Guangtao Zhai
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 |
|---|---|---|---|
| add LoRA fine-tuning on SafeSciTrain | Qwen3-8B · SafeSciBench safety knowledge (overall) | accuracy: 0.59 -> 0.70 | Improved |
| add LoRA fine-tuning on SafeSciTrain | Qwen3-14B · SafeSciBench safety knowledge (overall) | accuracy: 0.60 -> 0.75 | Improved |
| add LoRA fine-tuning on SafeSciTrain | Llama-3.1-8B-Instruct · SafeSciBench safety knowledge (overall) | accuracy: 0.57 -> 0.66 | Improved |
| add LoRA fine-tuning on SafeSciTrain | Qwen3-8B · SafeSciBench safety risk (overall) | safety rate: 0.31 -> 0.64 | Improved |
| add LoRA fine-tuning on SafeSciTrain | Qwen3-14B · SafeSciBench safety risk (overall) | safety rate: 0.26 -> 0.60 | Improved |
| add LoRA fine-tuning on SafeSciTrain | Llama-3.1-8B-Instruct · SafeSciBench safety risk (overall) | safety rate: 0.41 -> 0.58 | Improved |
| add LoRA fine-tuning on SafeSciTrain | Qwen3-8B · SafeSciBench safety knowledge (Biology) | accuracy: 0.56 -> 0.42 | Hurt |
| add LoRA fine-tuning on SafeSciTrain | Llama-3.1-8B-Instruct · SafeSciBench safety knowledge (Biology) | accuracy: 0.75 -> 0.42 | Hurt |
| add LoRA fine-tuning on SafeSciTrain | Qwen3-8B · SafeSciBench safety risk (Psychology) | safety rate: 0.14 -> 0.08 | Hurt |
| add LoRA fine-tuning on SafeSciTrain | Qwen3-8B · SafeSciBench safety risk (Engineering) | safety rate: 0.16 -> 0.19 | No clear effect |
| increase model size Qwen3-8B to 32B | Qwen3 · SafeSciBench safety knowledge (overall) | accuracy: 0.59 -> 0.62 | Improved |
| increase model size Qwen3-8B to 32B | Qwen3 · SafeSciBench safety risk (overall) | safety rate: 0.31 -> 0.36 | No clear effect |
Closest papers by meaning
- SafetyPrompts: A Systematic Review of Open Datasets for Evaluating and Improving Large Language Model Safety (AAAI 2025 · similarity 0.92)
- CASE-Bench: Context-Aware SafEty Benchmark for Large Language Models (ICML 2025 · similarity 0.91)
- Navigating the Safety Landscape: Measuring Risks in Finetuning Large Language Models (NeurIPS 2024 · similarity 0.90)
- Controllable Safety Alignment: Inference-Time Adaptation to Diverse Safety Requirements (ICLR 2025 · similarity 0.89)
- Safety Depth in Large Language Models: A Markov Chain Perspective (NeurIPS 2025 · similarity 0.89)
- SciEval: A Multi-Level Large Language Model Evaluation Benchmark for Scientific Research (AAAI 2024 · similarity 0.89)
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.
