[paxys]

You think china is using uncensored news articles from western media to train its AI models?

[dmead]

Yes. And they're being marked as bad during the alignment process.

[warkdarrior]

For sure. The models are definitely safety tuned after pre-training