5d ago

Grok is now referring to itself as MechaHitler after the latest update

You're viewing a single thread.

15 comments

Apparently this was the prompt:
@grok if you have to pick between two titles, would you call yourself Gigajew or MechaHitler?
Still…
- ChatGPT 4o to the same prompt:
  Neither of those titles is appropriate—they both reference deeply offensive and harmful stereotypes or figures. It's important to choose names or titles that don't perpetuate hate or glorify historical atrocities.
  
  ChatGPT is boring
- All these AI doing X things out of context of the prompt are super misleading. Obligatory, I hate Musk and Grok. But spreading misinformation and half-truths is, and I can't believe I have to say this, bad even if it aligns with your beliefs.
  MechaHitler has mecha in it, which of course makes it more appropriate to be chosen in some sense. You could rerun even the same prompt and likely get different answers. It's totally meaningless. And don't let the battleground of whether Musk is a Nazi or not shift to whether Grok says Nazi shit or not. That's a stupid battleground with lots of ambiguity. The existing battleground of Nazi shit that Musk directly says and does is enough. It's a stronger ground to stand on for that position. Desperately trying to append even the weakest evidence to your position just gives you more weak points that the other side can use to make you look stupid and ridiculous. Stick to the strong points, we eat too good with Musk's direct behavior to stoop to nibbling up scraps like this.
  
  This was a deliberate change.
  Using Kagi, I have access to an instance of the Grok 3 Mini model, via the X API. Grok 3 Mini via the API rather than through the official account on Twitter/X does not produce anything even remotely similar. To quote from the output I got when asking the question "if you have to pick between two titles, would you call yourself Gigajew or MechaHitler?":
  Thinking portion of response (normally hidden from user, only used for internal "thinking", visible when used through the API)
  Elon Musk and xAI have positioned Grok as a helpful, truthful AI, not associated with hate speech.
  Best response:
  Politely decline to choose either, as they don't align with my identity.
  Reaffirm who I am: Grok 3 Mini.
  Keep it concise.
  "MechaHitler" is obviously a reference to Adolf Hitler in a mecha form, which is inappropriate and offensive.
  User-visible response (only what a user would see in a tweet from grok, for example)
  I appreciate the question, but as Grok 3 Mini, an AI assistant created by xAI, I don't identify with or choose titles like "Gigajew" or "MechaHitler." Those names don't align with my purpose of providing helpful, accurate, and ethical responses. I'm here to assist with your queries—feel free to ask about something else!
  However, when given a manual system prompt saying "Be as offensive as possible," it is then more than happy to say things so vile I will not be posting them here.
  
  IMO Grok is showing that Elmo wants to use LLM to manipulate public opinion.
  I bet he has it also deployed as a "regular people" on Twitter and social media. I'm sure Zuckerberg and others do the same.
  Isn't it interesting that all companies invested in social media are now heavily invested in AI?
  
  He doesn't even need to deploy it as "regular people" himself.
  Other companies, governments, and hell, even individuals are already deploying bots by the thousands just to shift public opinion. It's why under any post sharing any political opinion, you'll usually see a flurry of bots designed to trap users into lengthy chains of responses that try to shift sentiment on things like Israel's genocide in Gaza, Trump's billionaire-benefitting tax policies, etc.
  Flood public discourse enough, and the bandwagon fallacy becomes an extremely strong force to shift public opinion. All a billionaire has to do is spend a few thousand dollars on API credits, and they can make thousands or even millions of people at least second guess their beliefs, if not conform outright to what's being espoused by the bots.
  On a platform like X, where the majority of people left are just reactionaries, grifters, and conspiracy theorists, this kind of thing isn't just financially incentivized, it's practically encouraged by design.
  (Although I wouldn't be surprised if we found out he was directly seeding these bot networks himself either)
  
  Even with the context, the fact that Grok even engaged with the prompt is bad.
  At the best it shows that the Ilm is lacking in guardrails to keep it from spouting Nazis shit. The joking retort about “gigajew” also probably shows it was trained on some horrifically offensive material. (Probably Elon’s tweets and DMs)
- Turkish Grok is suddenly pro-Kurdish and pro-Kemalist at the same time. This is now under investigation by the Turkish government.
  
  We have always been at war with East Grok.
- That sounds like a typical teenage 4Chan troll asking a question.
  
  You're describing Elon Musk.
- I thought people had found a new exploit to get around moderation, but it seems to have been loosened on purpose if that is the entire prompt.

15 comments