Understand this story
Researchers from the University of the Basque Country and Cardiff University tested mainstream large models with 31,000 open-ended cultural questions in 24 languages, and found that Gemini, ChatGPT, Claude, and Llama still disproportionately default to Japanese cases and paradigms after geographic restrictions are stripped away. The study locates the point where the bias emerges at the supervised fine-tuning (SFT) stage, arguing that the worldview at the pre-training stage is relatively balanced, while alignment training that pursues restraint and low conflict anchors Japanese-style cultural symbols into the models; the study also points out that the harsher the model quantization compression, the more extreme this stereotyped preference becomes. On the same day, AYi offered a view on the related discussion, arguing that what the models learn is the service-industry side, while the izakaya-complaining side has not entered the alignment data.
AI-generated from public reports
Latest development · Report
What the model learned is Japanese customer service, not cultureDevelopments
2 confirmed developments
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 23
A Yi AI Notes
What the model learned is Japanese customer service, not culture
Regarding what "Cyber Idler" said—"what the model learned is Japanese-style customer service; what SFT welds in is a polite-language shell rather than culture itself"—the author agrees and adds: what the model learned is the service-industry side; the izakaya-complaining side never entered the alignment data at all.
Ayi AI Notes
University of the Basque Country and Cardiff University study: mainstream large models favor Japanese culture on 31,000 open-ended questions
Researchers from the University of the Basque Country and Cardiff University tested mainstream large models with 31,000 open-ended cultural questions in 24 languages, finding that Gemini, ChatGPT, Claude, and Llama still disproportionately default to Japanese examples and paradigms after geographic restrictions are removed. The study locates the emergence point of the bias at the supervised fine-tuning (SFT) stage, arguing that the worldview at the pretraining stage is relatively balanced, while alignment training that pursues restraint and low conflict anchors Japanese-style cultural symbols into the models. The study also points out that the more aggressively a model is quantized and compressed, the more extreme this stereotypical preference becomes.