We don't know what it was trained on, do we? (Is there dataset info?) I'd suspect you're right, but I don't know. There also seems to be a lot of post-training processing done on AIs before they're released where a lot of bias can appear. I've never read a good overview about how someone goes from a LLM trained on data to a consumer-focusing LLM.

The article also leads into what oversight and regulation is needed, and how we can expect AIs to be used for propaganda and influence in future. I worry that what we're seeing with Grok, where it's so easily identifiable, are the baby steps to worse and less easily identifiable propaganda in future.

Looks like they trained it on the 4Chan /pol dataset