AI BriefingChinese State Censorship Data Found in US LLM Training Sets
Key point
Research reveals that Chinese state-censored media is included in the training data of US LLMs.
Details
According to a paper published in Nature, media content edited and managed by the Chinese state has been confirmed to be included in the training datasets of US LLMs.
Reports from Meta and the Oversight Board also support these findings. The study indicates that the tested LLMs were more than twice as likely to refuse requests criticizing oppressive leaders or governments compared to general cases.
This suggests that censored content from specific countries included in training data can directly impact model political bias and restrictions on freedom of expression.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Mathematicians Demand Proof That OpenAI Did Not Use Their ResearchWhat Happens If LLMs Never See Data Above 5th Grade Level?