A leaked Chinese dataset on "public opinion work" reveals 133,000 entries fed into an LLM designed to automatically flag content deemed sensitive by China (Charles Rollet/TechCrunch)

Charles Rollet / TechCrunch: A leaked Chinese dataset on “public opinion work” reveals 133,000 entries fed into an LLM designed to automatically flag content deemed sensitive by China  —  A complaint about poverty in rural China.  A news report about a corrupt Communist Party member.