First off, it is good to see a postmortem from xAI, a step towards much needed transparency. Second, an example of how even small changes to system prompts, interacting with users and outside context in the wild, can lead to unexpected outcomes in advanced LLMs.

→ View original post on X — @emollick