Comment by bartimus

When asking about Taiwan and Russia I get pretty scripted responses. Deepseek even starts talking as "we". I'm fairly sure these responses are part of the model so they must have some way to prime the learning process with certain "facts".