LLMs show surprising bias towards Japanese culture, new research finds

A new research paper published on arXiv investigates the cultural biases present in large language models (LLMs). Researchers developed a dataset called Culture-Related Open Questions (CROQ) with questions in 24 languages to test LLMs. Their findings indicate that LLMs exhibit a notable preference for Japanese culture when answering general culture-related questions, contrary to some previous studies. The bias appears to emerge after the supervised fine-tuning stage of LLM training, rather than during pre-training. AI

IMPACT Highlights potential biases in LLMs that could affect their performance and fairness across different cultural contexts.

RANK_REASON Research paper published on arXiv detailing findings about LLM biases. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• Culture-Related Open Questions (CROQ)
• Hugging Face
• Japanese culture
• Joseba Fernandez de Landa
• LLMs

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →