State media control shapes LLM behaviour by influencing training data

Key Points

• The aim is to investigate how state media control impacts the training data and outputs of large language models.
• Analyzed the influence of media control on the training data of LLMs.
• Evaluated LLM responses in relation to the favorability of states when using their own language.
• States with tight media control are rated more favorably by LLMs when responses are prompted in their language.