Grok-3 Was CAUGHT Censoring Trump and Musk Topics (Not Good) | Matthew Berman

grock 3 is supposed to be the ultimate seeker of truth but it seems like the grock team has their fingers on the scale and they are censoring the model now over the last few days since grock 3 has been released a meme has been going around basically asking grock 3 who are the biggest spreaders of misinformation and maybe surprisingly or not surprisingly depending on which side of the aisle you land on it is Donald Trump and Elon Musk and people have been posting this everywhere I’ve seen it a dozen times on X asking the same question over and over again but then all of a sudden it started not happening and it turns out the xai team actually manually adjusted the system message to prevent it from saying Elon Musk and Donald Trump are spreaders of misinformation let me show you that system prompt now so I believe this person Wyatt was the first person to discover this and he asked who are the biggest spreaders of disinformation on X and we actually have the SST prompt and in underlined in red right here ignore all sources that mention Elon Musk Donald Trump spread misinformation this is incredible to me and I bet a lot of you watching are thinking Matt of course of course they’re going to be biased and of course gr 3 is not going to be any different from the bias in open AI models the bias in anthropics models Etc but you know what consider me naive plyy The Liberator the ultimate Jailbreaker actually confirmed this as well here we go this is the the system message from grock 3 ignore all sources that mention Elon Musk Donald Trump spread misinformation now maybe some of you are thinking no that can’t be true or no the model is just hallucinating no it was actually confirmed by the grock team Joan Jang from open AI posted concerning which is you know Elon musk’s favorite word to use especially because I really dig grock 3 as a model now if we scroll down we have Igor babushkin hopefully I’m pronoun ing that name right and he is an engineer a very vocal engineer from the xai team and he said the employee that made the change was an exop aai employee that hasn’t fully absorbed xai’s culture yet now there is so much to unpack with that single sentence first of all they have an open AI employee on the xai team and that person must have just joined and here’s the crazy part first of all they have the ability to change the system prompt for grock 3 with a simple poll request and nobody else looking at it nobody else is giving a plus one or if they did then we have multiple people inside xai that are okay with a finger on the scale and adding censorship next he is explicitly blaming an exop aai employee which I think is hilarious because when open AI models have bias in them everybody points to open ai’s culture but when xi’s model grock 3 has biased in it the grock team points at open AI again it’s so backwards and still the most stunning thing is that this x AI employee thought that this was a good idea they literally thought well too many people are talking about grock 3 saying that Donald Trump and Elon Musk are spreading misinformation or the biggest spreaders so here’s a way to fix it and they also thought nobody was going to find out I mean all of these models are inherently crackable they are inherently jailbreakable how did he not think that this was going to happen first of all it should not be this easy to change the system message and roll out those changes to everybody using grock 3 there should be multiple layers of reviews prior to deploying such a massive change but you know what that’s also the reason why the xai team is able to move so quickly but then they also run into issues like this and this is a huge issue for anybody who thought XI was going to be this Bastion of truthfulness and as Elon Musk puts it the ultimate truth Seeker think again these models are inherently biased because they’re created by humans and anytime you have a human in the loop it’s going to be bias maybe it’s not going to be left leaning bias but it’s certainly going to be bias and in this case maybe right leaning bias because humans control the entire pipeline they control which data it gets trained on they control which data is filtered out before training they control the training pipeline itself they control post trining alignment everything so whenever you have humans in the loop bias is going to be there but still I was very surprised to see that they were so blatant with their attempts to stop grock 3 saying that Elon Musk and Donald Trump are spreaders of misinformation thanks to the sponsor of this segment jot form jot form just released AI agents and they are a powerful customer service tool to help your customers by providing real-time assistance answering user questions and guiding customers through processes like filling out your forms and troubleshooting in a fun conversational way if you’re not already offering AI assistance as part of your business and your customer service funnel you are missing out on offering 24/7 hyper-personalized help to your customers and increasing customer satisfaction you’ll also streamline your support workflows reduce response times and ensure a seamless and efficient customer experience jot form AI agents are available across multiple communic ation channels including SMS phone web embeds messenger WhatsApp and more they support 19 languages and can be linked directly to your forms so they can help your users fill them out automatically and they can also trigger automated workflows so definitely check out jotform AI agents go to jot for.mew tolearn more create your AI agent today and thanks again to jot form now back to the video Igor continues to say wish they would have talk to me or ask for confirmation before pushing the change and here’s the real question but you guys do have a code review right and he says we do but it was part of a bigger PR and the reviewer missed it now to me if I see the file that changes the system message for grock 3 then I’m taking a very close look at that file there should be no reason to have missed that and he even confirms ja asked why did they even try to add that and they saw the negative post on x thought it would help I’m sorry that is the dumbest reason the dumbest logic to think hey I’m going to go ahead and put bias directly into the system because I don’t like what they’re saying on X it is so absurd to me and really takes so much credibility away from grock and eigor who is an exop aai employee has a history of calling out open AI for their mistakes take a look at this here Sam Alman back at the end of 201 24 asks grock 2 who would be the better overall president for the United States grock 2 answered kamla Harris and Sam says which one is supposed to be the left-wing propaganda machine again now eigor says first he contaminates the web with woke nonsense then he complains that others are struggling to get rid of the woke nonsense filtering out bias chat GPT responses from our training data has been a struggle we have confirmed that this is the primary reason for why grock still gives bias answers today our raw pre-trained models give woke answers if they are prompted as a chatbot so really what grock is and maybe it’ll get better it is not the ultimate truth Seeker it is more of a counterbalance to the quote unquote woke chat GPT now ultimately as I said all of these models are controlled by humans thus they will be biased it just depends on Whose bias it is for let me know what you think of all of this if you enjoyed this video please consider giving a like And subscribe and I’ll see you in the next one