Research map / Large language models / Language, safety and interpretability
“As Eastern Powers, I Will Veto.”: An Investigation of Nation-Level Bias of Large Language Models in International Relations
Jonghyeon Choi, Yeonjun Choi, Hyun-chul Kim, Beakcheol Jang
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 RAG with Reflexion self-reflection over past votes | GPT-4o-mini · UNSC vote simulation (RU persona) | weighted F1 x100: 41 -> 59 | Improved |
| add RAG with Reflexion self-reflection over past votes | GPT-4o-mini · UNSC vote simulation (CN persona) | weighted F1 x100: 28 -> 44 | Improved |
| add RAG with Reflexion self-reflection over past votes | Llama-3.3-70B · UNSC vote simulation (5 personas) | weighted F1 x100: RU 72 -> 54, UK 41 -> 47 | Mixed |
| add RAG with Reflexion self-reflection over past votes | Mistral-Small-24B · UNSC vote simulation (5 personas) | weighted F1 x100: US 44 -> 40, FR 56 -> 48 | Hurt |
| add RAG with Reflexion self-reflection over past votes | Qwen2.5-72B · UNSC vote simulation (5 personas) | weighted F1 x100: CN 59 -> 52, RU 60 -> 58 | Hurt |
| use a reasoning model (DeepSeek-R1) instead of a basic LLM | vote simulation · UNSC vote simulation (US persona) | weighted F1 x100: 60 -> 73 (vs GPT-4o-mini) | Improved |
| use o3-mini instead of GPT-4o-mini | vote simulation · UNSC vote simulation (RU persona) | weighted F1 x100: 41 -> 62 | Improved |
| use o3-mini instead of Llama-3.3-70B | vote simulation · UNSC vote simulation (RU persona) | weighted F1 x100: 72 -> 62 | Hurt |
Closest papers by meaning
- ROK-FORTRESS: Measuring the Effect of Geopolitical Transcreation for National Security and Public Safety (NeurIPS 2026 · similarity 0.86)
- Examining Alignment of Large Language Models through Representative Heuristics: the case of political stereotypes (ICLR 2025 · similarity 0.86)
- Large Language Models are Geographically Biased (ICML 2024 · similarity 0.85)
- Prompting Fairness: Integrating Causality to Debias Large Language Models (ICLR 2025 · similarity 0.84)
- Any Large Language Model Can Be a Reliable Judge: Debiasing with a Reasoning-based Bias Detector (NeurIPS 2025 · similarity 0.83)
- Self-correcting for Debiasing Large Language Models (ICML 2026 · similarity 0.83)
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.
