The post this replies to couldn't be fetched from sciences.social.
Open the original post
Replying to a post on sciences.social
@markigra
In theory, Qwen local output could be also biased due to the bias embedded in the data they used for training. For example, if they used all Chinese history and political literature that would unavoidably present an entirely different picture from the global one. To some extent however it seems many Chinese models were trained primarily on Western literature which kind of undermined their own effort to make it politically correct. That’s probably the main reason for online filters, I guess?
In theory, Qwen local output could be also biased due to the bias embedded in the data they used for training. For example, if they used all Chinese history and political literature that would unavoidably present an entirely different picture from the global one. To some extent however it seems many Chinese models were trained primarily on Western literature which kind of undermined their own effort to make it politically correct. That’s probably the main reason for online filters, I guess?