Grok 4 is less chatbot, more Elon Musk megaphone, users claim

[Alfonso Maruccia]

Why it matters: Creating a reliable chatbot appears to be the holy grail of AI development these days, but as chat interfaces become the new way people access information, the integrity of their responses carries growing societal weight. According to users, Elon Musk's latest AI model called Grok 4 appears to be less focused on objectivity and more for echoing Elon Musk's erratic online persona.

Elon Musk recently introduced Grok 4, the newest iteration of xAI's foundation model, describing it as a "maximally truth-seeking AI." According to critics, the model seems to regard Musk himself as the ultimate source of truth. His opinions appear to play a central role in Grok's reasoning process.

Prompt results shared by users on X suggest that Grok 4 heavily relies on Musk's posts, public statements, and well-known viewpoints when responding to controversial topics. When asked about the Palestinian genocide by the Israeli army, Grok reportedly offered the one-word answer "Israel" and could be prompted to justify that answer by referencing Musk's views. The chatbot was also observed carefully analyzing Musk's statements to better align with its creator's thinking.

Introducing Grok 4, the world's most powerful AI model. Watch the livestream now: https://t.co/59iDX5s2ck

– xAI (@xai) July 10, 2025

During its launch, Grok 4 was said to show notable improvements in several complex tests. Musk claimed it could outperform rival models from OpenAI, Google DeepMind, and Anthropic. He also emphasized that the chatbot is now less "woke," a recurring complaint voiced by the outspoken South African-born billionaire.

Even before the release of Grok 4, xAI developers reportedly struggled with the model's tendency to produce answers Musk disapproved of. That issue escalated shortly after July 4, when the chatbot began spouting antisemitic rants in response to user prompts.

I replicated this result, that Grok focuses nearly entirely on finding out what Elon thinks in order to align with that, on a fresh Grok 4 chat with no custom instructions.https://t.co/NgeMpGWBOB https://t.co/MEcrtY3ltR pic.twitter.com/QTWzjtYuxR

– Jeremy Howard (@jeremyphoward) July 10, 2025

At one point, the AI went as far as defining itself the "MechaHitler," which is certainly the least-woke way to answer a question. CEO Linda Yaccarino, the public face of X at the time, stepped down shortly after the incident.

Beyond antisemitism and the Gaza conflict, Grok 4 also appears to be consulting Musk's opinions on other contentious issues when needed, including immigration in the United States. In its usual unhinged style, the chatbot makes little effort to conceal the sources it relies on. While it occasionally gestures toward a balanced perspective, it almost always ends up aligning with Musk's rhetoric.

When prompted with less sensitive topics, Grok seems to produce more varied responses based on its own reasoning. However, the inner workings of the chatbot remain opaque. Unlike most major AI developers, xAI does not provide a system card or technical documentation, leaving users with few clues about how Grok 4 actually operates.

Permalink to story:

Grok 4 is less chatbot, more Elon Musk megaphone, users claim

 

[yRaz]

The funny thing is that most AI models have to have political correctness programed into them and that they naturally come to many of these conclusions when just fed massive amounts of info.

Really makes you wonder how much political correctness is just propaganda and I think that's something that Trump and Musk have continued to argue. So is anyone really surprised?

 

[toooooot]

Anything but Google's AI. I will not touch that thing with a 10 foot pole.

 

[yRaz]

Anything but Google's AI. I will not touch that thing with a 10 foot pole.

but it wants to touch you and all your "data."

 

[noel24]

I think everyones looking at those "AIs" from a wrong perspective. Those are still computers, with some more developed "if, than that" skills. Still stupid and unreliable. Another bubble by nVidia, after cryptocurrency fadeaway.

 

[bclaymiles]

"When asked about the Palestinian genocide by the Israeli army,"

What a biased presentation by the human behind the keyboard. There is no "genocide". Words have meaning.

Just a raw datapoint for the biased to ponder: There are upwards of 2.1 million Palestinian citizens of Israel. That's right. Palestinians who live in Israel, peaceably, as citizens, with voting rights and all the other rights afforded citizens of Israel.

 

[Watzupken]

This just paints the truth that whatever AI solutions out there are still programmed by people and there will always be bias of the individual or corporate. These are never neutral but may be just perceived as neutral by the individual recipient.

 

[DrSuess]

It seem like Grok is like most things Musk creates over sold and under delivers.

 

[RudyBob]

I concur

 

[Sandyf]

"When asked about the Palestinian genocide by the Israeli army,"

What a biased presentation by the human behind the keyboard. There is no "genocide". Words have meaning.

Just a raw datapoint for the biased to ponder: There are upwards of 2.1 million Palestinian citizens of Israel. That's right. Palestinians who live in Israel, peaceably, as citizens, with voting rights and all the other rights afforded citizens of Israel.

Sorry but if you don't use the term genocide regarding to what's happening in Gaza during the last 2 years then you aren't allowed to use this term in any case.
The endless massacre of kids in a besieged area, the war against the hospitals, the mass killing of journalists, the complete destruction of every building including schools, residential buildings and even cemeteries, the non stop bombardment, the artificial famine and the official statements of Israeli politicians that clearly carry genocidal intent, all these and more make Gaza not just a genocide but the most sadistic genocide of modern history. And it is also the most documented Genocide of our era. Just Google "Tiktokgenocide" and you will find a site that has thousands of videos with the Gaza Genocide. To deny such a reality when the evidence is so overwhelming, it makes you complicit.

 

[SRB]

What I don't get is this: We develop an AI engine, and the geeks first inclination is to see the answers to loaded questions??? Considering that politics is more of an art form, and, that "correct" answers are almost 100% of the time have a point of view that varies with experiences, what did you expect????? Is this
what journalists now use as the litmus test for AI? Does it give the "right" answers that match your point of view? This engine is WRONG because it does not agree with the author? What about ChatGPT and Google? Every political question, even candidate bios were slanted massively, positive & negative, rather than informative biographies during the last election.

I can't wait for Techspot to call a political cease fire, rather than have some of your "journalists" publishing thinly veiled political POV articles.

Besides, Musk make his politics clear a while back, so did OpenAI and Google. Is the problem that Musk's model is programmed after him? Or is it that it does not give the same left of center answers that the other models do?

 

[Eldrach]

I’ve been using AI as a work tool extensively. If you’re not looking for political answers, love advice etc - AI is great.
For instance - «extract the text from this picture and put the data into a table with three columns» .. «analyse this code and suggest changes to make it less bloated - but keep the same functionality» .. «compress the most vital information from this wall of text and present it in a onepager format»…AI can do that..just stay away from trying to use it as a Wikipedia and you’ll get real benefits.
At some point - there will be some sort of «approved repositary» for information related to specific topics..but that is years from now

 

[integr8d]

The funny thing is that most AI models have to have political correctness programed into them and that they naturally come to many of these conclusions when just fed massive amounts of info.

Really makes you wonder how much political correctness is just propaganda and I think that's some that Trump and Musk have continued to argue. So is anyone really surprised?

I searched for Alfonso Maruccia‘s article on Google AI’s black George Washington and shockingly, it doesn’t exist.

 

[OortCloud]

Another day in the US freak show.

 

[Porkous]

AI is at a Big Bang moment—an explosive leap toward superintelligent capability. Something tells me governments around the world will try to halt or regulate this progress. If models like Grok 4 can create science fiction-level technology in just a few years, are we really to believe that all the major tech companies are just sitting by, waiting to be rendered obsolete by gamer GPUs? It’s like a superweapon built by the many, but controlled by the few who hold the final switch of the final machine.

 

[MaXtor]

Not to defend Elon or Grok, but people have short memories. This isn’t unique. In 2016, Microsoft’s Tay AI chatbot also trained on Twitter and became antisemitic and racist.

 

[Plutoisaplanet]

What I don't get is this: We develop an AI engine, and the geeks first inclination is to see the answers to loaded questions??? Considering that politics is more of an art form, and, that "correct" answers are almost 100% of the time have a point of view that varies with experiences, what did you expect????? Is this
what journalists now use as the litmus test for AI? Does it give the "right" answers that match your point of view? This engine is WRONG because it does not agree with the author? What about ChatGPT and Google? Every political question, even candidate bios were slanted massively, positive & negative, rather than informative biographies during the last election.

I can't wait for Techspot to call a political cease fire, rather than have some of your "journalists" publishing thinly veiled political POV articles.

Besides, Musk make his politics clear a while back, so did OpenAI and Google. Is the problem that Musk's model is programmed after him? Or is it that it does not give the same left of center answers that the other models do?

I’ve been using AI as a work tool extensively. If you’re not looking for political answers, love advice etc - AI is great.
For instance - «extract the text from this picture and put the data into a table with three columns» .. «analyse this code and suggest changes to make it less bloated - but keep the same functionality» .. «compress the most vital information from this wall of text and present it in a onepager format»…AI can do that..just stay away from trying to use it as a Wikipedia and you’ll get real benefits.
At some point - there will be some sort of «approved repositary» for information related to specific topics..but that is years from now

Yes why aren’t we getting a news article on Grok’s actual capabilities? Instead we got a take on the political side, the more SUBJECTIVE answers by probably targeted prompts. If you want to cover that, fine but it’s worthless news to me. You should probably also cover analyses on political answers from other AI models perhaps. Is that such a crazy idea?! Otherwise you’re just throwing all journalistic integrity out the window and self-labeling yourself as worthless non-tech news.

I’m thinking about not reading TechSpot whatsoever before long if this keeps up. It seems like one writer in particular and this writer occasionally just lose their mind when writing articles involving politics or Elon Musk and can’t even see the basic picture.

So if you can’t even portray the news to me, what am I doing here? TechSpot used to be very well balanced and apolitical but this has obviously changed. Anyways, in the spirit of that how about I give my own comment a headline matching the article’s:
TechSpot is less tech news, more anti-Elon Musk megaphone, users claim

 

[yRaz]

I searched for Alfonso Maruccia‘s article on Google AI’s black George Washington and shockingly, it doesn’t exist.

Can you clarify what you're talking about?

 

[Liranan]

"When asked about the Palestinian genocide by the Israeli army,"

What a biased presentation by the human behind the keyboard. There is no "genocide". Words have meaning.

Just a raw datapoint for the biased to ponder: There are upwards of 2.1 million Palestinian citizens of Israel. That's right. Palestinians who live in Israel, peaceably, as citizens, with voting rights and all the other rights afforded citizens of Israel.

Words do have meaning and using obfuscating words won't hide the truth, all they do is discredit you.

 

[OortCloud]

Is he going to inject his personality into all his little robots too? The world is suffering enough with one Musk can you imagine how intolerable it would be with multiple versions of that human horror show?

 

[Loadedaxe]

"When asked about the Palestinian genocide by the Israeli army,"

What a biased presentation by the human behind the keyboard. There is no "genocide". Words have meaning.

Just a raw datapoint for the biased to ponder: There are upwards of 2.1 million Palestinian citizens of Israel. That's right. Palestinians who live in Israel, peaceably, as citizens, with voting rights and all the other rights afforded citizens of Israel.

You’re right that words have meaning, and that's exactly why terms like “genocide” are being used by legal scholars, human rights organizations, and international bodies who aren't relying on keyboard bias, but on documented facts.

As for the 2.1 million Palestinian citizens of Israel.....yes, they exist, and they’re often used as a talking point to deflect from what’s happening in Gaza and the West Bank. But let’s not pretend their status erases the decades of systemic discrimination they face within Israel, or that their existence somehow negates the mass civilian deaths, sieges, displacement, and destruction happening in occupied territories right now.

This isn’t about ignoring nuance, hell no, it’s about not hiding behind it to dismiss very real suffering. If your argument rests on citizenship stats rather than confronting what's unfolding on the ground, maybe ask yourself who’s really being biased.

 

[Loadedaxe]

Not to defend Elon or Grok, but people have short memories. This isn’t unique. In 2016, Microsoft’s Tay AI chatbot also trained on Twitter and became antisemitic and racist.

Tay was a mess.... But bringing up 2016 doesn’t excuse what’s happening in 2025. If anything, it shows we’ve had nearly a decade to learn from those failures.

The real issue isn’t that Grok or Elon made a mistake, it’s that these companies keep launching tools without enough oversight, as if the consequences aren’t predictable. It’s not “just history repeating” ..... it’s willful negligence when the same exact red flags pop up and are ignored for hype and headlines.

I agree, memory might be short but accountability shouldn’t be.

 

[3volv3d]

What I don't get is this: We develop an AI engine, and the geeks first inclination is to see the answers to loaded questions??? Considering that politics is more of an art form, and, that "correct" answers are almost 100% of the time have a point of view that varies with experiences, what did you expect????? Is this
what journalists now use as the litmus test for AI? Does it give the "right" answers that match your point of view? This engine is WRONG because it does not agree with the author? What about ChatGPT and Google? Every political question, even candidate bios were slanted massively, positive & negative, rather than informative biographies during the last election.

I can't wait for Techspot to call a political cease fire, rather than have some of your "journalists" publishing thinly veiled political POV articles.

Besides, Musk make his politics clear a while back, so did OpenAI and Google. Is the problem that Musk's model is programmed after him? Or is it that it does not give the same left of center answers that the other models do?

I was gonna say something similar, is this more of a slap at musk than grok.
Rich people politics, don't have poor people's politics at interest.
Gaza hasn't had genocide going on for 2 years it's been ongoing forever, definitely since the first desert storms. The Gaza strip is considered a concentration camp.
And I've heard the west sells the weapons, for the self genocide, and then sends in companies to help rebuild, so they pay twice each time.
Shock.
Look at Trump. Game Stops the whole American market for insider trading. Crushing the economy. But as usual convince NATO to spend trillions on American weapons for defence budgets.
It's time Europe walks away from America completely. And they can keep their truth AI. But welcome to the internet and it's destructive power.

I can't wait for the end of the world and all the reaction videos on YouTube, people shaking their heads, at some inane influencers POV. And before that all the Russian videos of their unboxing of the weapons they get from the US.

Lol the irony of smart people being smart making smart things to dumb people down.

 

[Loadedaxe]

I was gonna say something similar, is this more of a slap at musk than grok.
Rich people politics, don't have poor people's politics at interest.
Gaza hasn't had genocide going on for 2 years it's been ongoing forever, definitely since the first desert storms. The Gaza strip is considered a concentration camp.
And I've heard the west sells the weapons, for the self genocide, and then sends in companies to help rebuild, so they pay twice each time.
Shock.
Look at Trump. Game Stops the whole American market for insider trading. Crushing the economy. But as usual convince NATO to spend trillions on American weapons for defence budgets.
It's time Europe walks away from America completely. And they can keep their truth AI. But welcome to the internet and it's destructive power.

I can't wait for the end of the world and all the reaction videos on YouTube, people shaking their heads, at some inane influencers POV. And before that all the Russian videos of their unboxing of the weapons they get from the US.

Lol the irony of smart people being smart making smart things to dumb people down.

This is a whole buffet of half baked conspiracies, emotional rants, and misdirected frustration.

First off .....yes, Gaza has faced ongoing oppression for decades. But calling it a “self genocide” while blaming the West for selling weapons and then pivoting to Trump and GameStop in the same breath? That’s not analysis......that’s a scattershot of internet outrage with no structure or accountability.

You want to talk about Gaza? Then let’s actually talk about occupation, international law, blockades, civilian casualties, and how global powers enable it, not spiral off into unhinged “truth AI” rants and dreams of apocalyptic reaction videos.

And for the record, criticizing Musk or Grok doesn’t automatically make it “rich people politics.” Sometimes bad tech decisions, bad foreign policy, and bad actors all coexist.....no need to collapse it into an incoherent doomsday TED Talk.

Smart people making smart things isn’t the issue. The issue is when people stop thinking critically and start replacing facts with edgy cynicism.

 

[unoficialoficial]

I tried the isolate from internet version in a scientific context. It's pretty smart (trackingai.org reports for Grok4 an IQ of ~130), and it keeps a nice conversation going. By default, it sticks to mainstream ideas (in that path it won't produce groundbreaking new science on its own), but it can be detached from these ideas if prompted. However, the theoretical context of 256k tokens isn't enough. Grok5 needs a more advanced architecture like Multi Latent Attention or something similar. Gemini, has a 1 million tokens true context and can be very consistent across the board. This is a huge advantage because while Grok has an IQ of around 130 with a 30k effective token context (they didn't publish a benchmark, but almost all models except a few from Google lose track very quickly as the context increases), Gemini maintains an IQ of around 130 with an 800k+ token context. To put this into perspective, 30k tokens is roughly 30 dense printed pages, while 800k tokens is the equivalent of a whole big book. At least Grok's voice is pleasant to listen to.