Yeah weird how LLMs randomly seem to reflect the politics of their owners, so random I guess we'll never know, weird.

It does seem that it is egregious enough it should have been caught in QA before being released to the public.

Everything else around it is a dopamine high from joining in on a hate train against Musk.

Grok was tweaked, it got directions to be crazier.

Grok has seemed unusually prone to bizarre behavior, even if you don't count the times that its system prompt was modified (like the infamous "please talk about white genocide" incident in May). Part of the problem seems to be that it's being trained on recent Twitter posts, resulting in a weird sort of feedback loop when users discuss Grok's behavior.

Difference being the CEOs of Google and Microsoft didn't do a Nazi salute before their bots started extolling the virtues of Hitler's ideology. Every damn time.