Agree. I never claimed this. My argument is simply that these biases don’t exist “at the presentation/UI layer only.”

At the end of the day, if you’re potentially doing anything politically sensitive to China, the fact that its models, including open source, have a higher (though probably not 50%) error rate in those categories is notable.