, social etiquette, religion, family roles).
New dataset aligns Arabic LLMs with country-specific cultural norms
ArabPref, a preference dataset spanning 22 Arab League countries, improves culturally appropriate behavior in language models after post-training.
Big Tech
Junhong Liang · Daniil Orel · Saeed Almheiri · Bilal Elbouardi · Younes Samih · Abed Alhakim Freihat · +8 more
Mohamed bin Zayed University of Artificial Intelligence · IBM Research AI
Research Digest··2 min read

The authors introduce ArabPref, the first large-scale behavioral preference dataset covering all 22 Arab League countries across 200 cultural aspects.
§