Hacker News Reader

I highly doubt that there is simply no training content about Xi Jinping, as if he didn't exist.

More-likely it is some combination of a system prompt and output-filtering code.

I more meant, if a model is only trained on 100% absolutely positive data concerning a specific subject, including any data that implies certain taboos it must avoid at all costs, the model will embed that strong bias in its learned parameters.

Reply on news.ycombinator.com