[maxlin]

What a crazy thing to say explicitly about the model that _avoids_ that kind of stuff. Did we already collectively forget about the ChatGPT "misgendering is worse than thermonuclear war" case, or Gemini picturing the Founding Fathers as black?

Anything political, I always go to Grok first. It's the only one that has bled for not just trying to play it super safe with political correctness, but trying to be impartial.

[saulapremium]

>Anything political, I always go to Grok first.

Why would you talk to an LLM about anything political?

[maxlin]

For reasons I'd use an LLM to learn about things / find relevant sources? A lot of things have politics ingrained in to them, and from experience I know some models will absolutely lie to avoid even being a lightyear from saying anything offensive.

[KingMob]

You forget about Mecha-Hitler and Musk having the system prompts changed for him? Grok is dead-last on impartiality.

[maxlin]

Your arguments defeats itself, as Grok is the only one that let the user make it say such things by command. People act like Grok did that on it's own lol.

I want a car that If I decide to run it in to a wall to avoid running over people, it does that. People arguing that Grok doing what the user asks for, being actually impartial, being somehow "biased" is crazy.

[KingMob]

Grok did do that on its own: https://archive.ph/YktYp

smdh