4 posts found
9 models. 145 political questions. BaiChuan refused 60%. DeepSeek refused 36%. GPT-4o refused 0%. PNAS Nexus published the first comparative analysis of LLM censorship — and proved that when the government approves the model before launch, the AI goes silent.
Jensen Huang said we've achieved AGI. Sam Altman said the term 'isn't useful.' One sells chips, the other sells subscriptions. I spent weeks trying to understand who's right. The answer bothered me.
Researchers from Stanford and Princeton tested 9 AI models with 145 political questions. The results reveal how censorship and government values are embedded in the algorithms we use daily.
When AI creators accuse others of 'data theft,' but built their models by scraping the entire internet without permission.