[elcritch]

xAIs prompts are open?! Wow that's pretty impressive.

[pavlov]

In the same sense that DOGE's receipts are open.

They publish something that looks good, and the reality is something else.

[DaSHacka]

So they don't actually publish the correct prompts, and are just lying? Do you have a source for that?

[pavlov]

A "rogue employee" [1] at xAI recently tweaked the prompts to make Grok talk about "white genocide" in South Africa in response to completely unrelated queries:

https://apnews.com/article/grok-ai-south-africa-64ce5f240061...

These modifications didn't seem to get reflected in the published xAI prompts.

- -

[1] Musk himself, obviously