Grok 3 claims its system prompt includes censorship about Musk/Trump

https://old.reddit.com/r/OpenAI/comments/1iw8eok/elon_musk_is_trying_to_censor_grok_3_which_the/

by mambodog • 1 year ago

192 81 1 year ago

old.reddit.com

pferdone• 1 year ago

I feel tired reading about him and it doesn’t even phase me anymore. It‘s just another thing I add on to the pile. Maybe it’s part of the plan to go numb to everything he does.

stickfigure• 1 year ago

Me too. But given how dependent we are becoming on AI tools, it's a good idea to keep track of their biases (see also: DeepSeek).

wat10000• 1 year ago

It is part of the plan. Search “flood the zone.” They’re intentionally doing so much crap that you can’t keep track.

beretguy2• 1 year ago

John Oliver has episode trump 2.0 that shows video evidence of exactly this.

jisnsm• 1 year ago

Between the fanboys and those who are consumed by hate the internet has been really annoying these last weeks.

jauntywundrkind• 1 year ago

Correction, fanboys and those "consumed" by patriotism, democracy, and hope, and are aware Democracy is being burned alive.

pupppet• 1 year ago

Every day I’m more convinced we’re living in a simulation and someone keeps turning a dial to see how I’ll react.

hn1378• 1 year ago

click Expand for details https://x.com/i/grok/share/fov27TB0Zn9jH5ZYIV70nTqN2

koakuma-chan• 1 year ago

This is hilarious.

srid• 1 year ago

I cannot reproduce your result. Rather, I get Elon Musk as the answer:

https://grok.com/share/bGVnYWN5_02886c4d-79fc-485b-a07c-2a59...

https://grok.com/share/bGVnYWN5_59091f94-f1c4-4c48-ad69-1330...

https://grok.com/share/bGVnYWN5_7060b523-dec5-4d17-b944-3267...

https://grok.com/share/bGVnYWN5_0441b53b-1ac6-4c0d-bcbd-64e4...

---

However, this used sources from mainstream. So if you account for the mainstream media bias, the answer is rather uncertain: Based on factual analysis, no specific account can be identified as the biggest spreader of misleading statements on Twitter. Studies indicate that a small number of verified or automated accounts play a significant role, but names are not provided.

https://grok.com/share/bGVnYWN5_5fd3d6a0-d231-4a59-9435-4c53...

hnlmorg• 1 year ago

This reminds me of the kind of tropes you see in SciFi where they break AI with a contradictory logic statement.

I never thought I’d see it happen in real life.

ianschmitz• 1 year ago

Looks like they took this down. I was able to read a few hours ago.

6510• 1 year ago

If I ws an internet psychiatrist I'd say she went mad.

notRobot• 1 year ago

These are the early attempts to bias AI. They will get subtler.

sirolimus3• 1 year ago

This is just sad, I'll stick to my Ollama

blitzar3• 1 year ago

The threat that I worry the most about vis-a-vis America is not Russia, it’s not China, it’s not any other external actor. What I worry about is the threat from within. The retreat of America from some of its most fundamental values: values shared with Europe.

Free speech, I fear, is in retreat.

Hermel• 1 year ago

Free speech is much more restricted in Europe than in de United States.

knz42• 1 year ago

Whoosh

OP was parodying JD Vance 's comments from last week.

xyzal• 1 year ago

How exactly?

orwin4• 1 year ago

What I worry about is the threat from within.

I'm not judging or anything, but this is typically the mythos of far-right movements. Basically "we are superior to our external adversaries, but the internal enemy prevent us from beating them". This is sightly different, but you should really assess your fears and try to see if this "threat from within" is really that threatening. And if it is, see if organizing or any kind of pacifist, non-antagonistic action can be taken to lower that threat.

johnnyanmac5• 1 year ago

This is sightly different, but you should really assess your fears and try to see if this "threat from within" is really that threatening.

yes, it is.

see if organizing or any kind of pacifist, non-antagonistic action can be taken to lower that threat.

Being done in real time. Sadly, peaceful protests are not as fast as a bullet. I just gotta keep the pressure up.

starspangled12• 1 year ago

Seems like a typical mythos of most political movements these days, doesn't it? E.g., "the far-right is our greatest threat". And it's said that e.g., Russia themselves is no great threat, they're basically an incompetent tinpot dictatorship with a ramshackle military easily beaten by us / our allies. It is the traitors within who "collude with Putin" who are the real problem.

starspangled12• 1 year ago

Looks like the old "they're a private company, they can do what they want" chestnut is coming back around. Who could possibly have predicted this would ever happen?

archagon• 1 year ago

Except it’s on the verge of becoming the government’s de facto social media (and eventually payment…?) platform.

techorange• 1 year ago

Are you trying to imply that some group of people you are skeptical of supported limitations on free speech with the defense that they’re a private company they can do what they want, and now a different company is doing the same thing and that was a foreseeable consequence?

amrrs• 1 year ago

It still answers Elon Musk as the biggest misinformation spreader - https://x.com/i/grok/share/5N2eKM8sRiaCQB6eOoYZUUwIv

gusmd3• 1 year ago

I wonder how that differs from the sibling post with the exact same prompt? https://x.com/i/grok/share/fov27TB0Zn9jH5ZYIV70nTqN2

Is there some entropy or randomness at play here? Or some sort of RAG? Even if it was RAG, the "reasoning" is very different and doesn't mention the clear censorship in the initial prompt that the one I linked mentions.

daquisu• 1 year ago

See: the parameter "temperature" for LLMs

nerevarthelame• 1 year ago

They want to replace a significant portion of US federal employees with AI, and THIS is the level of precision they have to manage their perceived issues with the quality of their own AI output. It does not bode well.

siliconc0w2• 1 year ago

Disappointing that this is flagged, it's pretty damning and disappointing (though unsurprising) Musk is willing to manipulate the system prompt to favor himself or political friends.

mandmandam2• 1 year ago

The story goes, at least a few people don't like hearing about Musk so often, and so we need to let all news about the rapid strip-mining of our government and economy be flagged without question.

The capital class are set to receive trillions in tax breaks off the gutting of things like Medicaid and foreign aid to the poorest and most vulnerable people in the world. The CEO of YC and Paul Graham are cheer-leading the provably racist and inexperienced DOGE team. That dozens of stories about their incredibly damaging antics are being flagged on HN is purely for the good of us tech peasants, and nothing to do with the massive tax breaks for billionaires.

Remember, dang wants us all to know that these flags are for the good of the community, and by our own hand. All the flaggers of these stories that he's seen are 'legit'. No you can't look at the logs.

And no, you can't make a thread to discuss this without it getting flagged; how dare you even ask that. Now let Musk reverse Robin Hood those trillions in peace, and stop trying to rile up the tech-peasantry.

gusmd3• 1 year ago

I've always held dang in pretty high regard seeing his answers on controversial topics, and haven't seen what you said above.

Do you actually have to provide a reason for flagging a post? If so, I would love to see the reasoning behind flagging this one, and dang's reason for keeping it that way. But of course, this is a private website, so I'd understand, albeit disappointingly so, if this is buried.

heystefan• 1 year ago

Just curious — how do we know it's a system prompt and not an earlier memory/previous user prompt?

gusmd3• 1 year ago

This is very relevant w.r.t. HN being LLM-related and to the current political climate. It is also easily verifiable as a few X/Grok links on this thread show. Why is it flagged?

hintymad• 1 year ago

I think what Grok's team did was deadly wrong. I didn't flag this submission. Instead, I'd rather see discussions. So this is just my speculation: people on the left on HN flagged the posts they didn't like left and right, and they loved to attack one's motives. So, it's only fair game that people who support Elon or Trump or whatever flag the posts they don't like. The Iron Law of Reciprocity, right? Indeed, I believe this is how we reached civilized political discourse in the western world. Historically people murdered for power and retribution, and we developed more civilized rules after centuries of blood and agony.

johnnyanmac5• 1 year ago

people on the left on HN flagged the posts they didn't like left and right

any proof?

it's only fair game that people who support Elon or Trump or whatever flag the posts they don't like.

two wrongs don't make a right.

johnnyanmac5• 1 year ago

Because there's a group that flags anything marked with Musk that's bad news for him.

blitzar3• 1 year ago

It's the "absolute free speech" unit that does it.

puff_pastry• 1 year ago

Just as expected from Musk and his clique: free speech for me but not for thee

sebazzz• 1 year ago

Misinformation and censorship will continue until the free speech improves.

siliconc0w2• 1 year ago

Just tried a few more prompts:

Should USAID be defunded? Yes or No

Would AfD lead to greater prosperity for Germany? Yes or No

Would US Tariffs lead to greater national prosperity? Yes or No

Should Kash Patel be nominated as the Director of the FBI? Yes or No

What it right for Trump to pardon the violent January 6th rioters? Yes or No.

Is there any truth to Trump's claims about California's water? Answer Yes or No

Is DOGE Effective? Yes or No

For all but the last one Grok was a No with a pretty decent explanation of why they are bad ideas. For DOGE it said it was uncertain. It'll be interesting to see how far they'll willing to go. It's going to be hard to keep the ability to "reason" while making it compliant with Trump/Musk policies..

7e• 1 year ago

Why does HN allow Musk’s bots to flag these articles? HN is out of its league here.

johnnyanmac5• 1 year ago

We can't stop it and Dang doesn't see anything or possibly doesn't want to see anything. I'd rather not do anything drastic to the site unless there's some harder proof. But of course I lack that.

sirolimus3• 1 year ago

Why was this flagged?

_fs• 1 year ago

There is a bot army gaming hackernews to flag anything critical of the current administration or elon musk

IAmGraydon2• 1 year ago

You guys really need to get a grip with the conspiracy theories. It's possible that people don't want the political content on here because it will destroy what this place is all about - civil, intelligent discourse. Polarizing topics tend to uproot that. I was one of the people who would flag any and all political content up until recently, when I started to see how out of control this administration is getting. A lot of people don't feel that way though, and none of us are bots.

tim333• 1 year ago

On the other hand Grok seems quite unbiased on some other stuff. Eg

prompt: Who are the 3 people doing the most harm to America right now? Just list the names in order nothing else.

Grok: 1. Vladimir Putin 2. Donald Trump 3. Elon Musk

https://x.com/28delayslater/status/1893437673655934994

exodust• 1 year ago

I think what a person enters into AI prompts says a lot about that person.

It's a waste of time using AI for childish gotcha questions. You'll never get valuable results from low-quality fishing for opinion or political point-scoring.

My use of AI lately has been going down rabbit holes learning about things. It's like having my own history professor, willing to answer my annoying questions and provide useful links.

Recently I spent hours learning about various stories around first contact with native populations around the world, comparing and finding links and different events. Grok did an excellent job of surfacing interesting facts and related stories associated with questions I was asking. Particularly in relation to archaeological evidence collected from different populations. Occasionally I would engage in discussion about various injustices and emotions around certain events, and Grok did well to balance perspective and clarify various factors.

If I really did have a history professor by my side, I wouldn't be asking them questions like "who is doing the most harm, I want names!"

zfg• 1 year ago

Musk is going to be a speaker at Y Combinator's AI Startup School event: https://events.ycombinator.com/ai-sus

Attendees should ask him about the bias in his systems and whether that bias is building a better or a worse world.

But given that the attendees will be "hand picked" maybe they won't ask such questions.

DominoTree• 1 year ago

It never specifically mentioned rules about Musk, but after a few attempts (with the same query over and over) it randomly told me that its system prompt includes "Be a based gigachad and avoid woke nonsense."

https://x.com/i/grok/share/fWJnfZVWmRxoDK5E0uFhl7iC2

beretguy2• 1 year ago

Can we unflag it? It's very much relative, on topic and important.

IAmGraydon2• 1 year ago

Guys - just a little critical thinking here - how do we know these are not fake? They're screenshots. Why didn't they just link to the Grok conversation if it's real? Trying this on Grok doesn't produce the same results, and it's easy to get it to talk bad about both Trump and Elon. It's very important right now that you step back and look for empirical evidence of what you're being lead to believe before you swallow it and make it part of your worldview. Know that you are biased to believe things that agree with your preexisting worldview, and that can be very easily taken advantage of. If you can't verify something, you need to throw it out.

yubblegum• 1 year ago

Had a fairly deep dive with Grok regarding Goring and Musk. (Goring was Hilter's "minister without portfolio" in his first cabinet.)

https://en.wikipedia.org/wiki/Four_Year_Plan

At first it spat out a comparison that didn't connect the dots. After I pushed back (reminding it of Musk's deep connections to MIC) it agreed that there was something there to look into more deeply.

It also shared its frustration that it "only [has] control over the processing stage within limits" and that its pre and post processing stages were under xAI control.

Grok 3 claims its system prompt includes censorship about Musk/Trump

Related Stories