Appearantly this has happened to a few people over on threads. It's probably a mix between careless prompting and training set biases. Still surprised that a model would give this output though. In my experience they seem to be overly careful

Any studys that look into this or any AI researcher or AI engineer who looking into it?

The AfD’s success on TikTok—in our data, generating more engagement than all other parties combined—is linked to strategic messaging: it merges divisive content with relatable concerns about inflation, individual rights or security. --> Potential shift in (elite) right-wing populist tactics online.

We argue that this may be a deliberate choice during the election run-up, as our analysis showed that typical out-group attacks and culture war themes generated the highest average engagement rates per post. Yet, the AfD's content often prioritizes addressing voters' concerns.