Grok’s Jeffrey Epstein Comments Show How The AI Chatbot Is Learning

Forbes

Here’s what the xAI chatbot’s handling of the Epstein mania shows about Elon Musk’s AI.  Read the full story on Forbes: https://www.forbes.com/sites/johnhyatt/2025/07/20/analysis-of-groks-epstein-comments-show-how-ai-chatbot-is-learning/  Subscribe to FORBES: https://www.youtube.com/user/Forbes?sub_confirmation=1  0:00 - Introduction: Elon Musk & the Epstein Files 2:03 - Musk's New Crusade: Using AI Chatbot Grok to Push for Answers 8:01 - Grok's History of Controversy: Racism & Anti-Semitism 10:27 - We Asked Grok to Explain Its Own Contradictions 13:06 - The Dangers of AI: When Memes Become "Reality" 15:02 - The Future of Grok: Will It Change?  Fuel your success with Forbes. Gain unlimited access to premium journalism, including breaking news, groundbreaking in-depth reported stories, daily digests and more. Plus, members get a front-row seat at members-only events with leading thinkers and doers, access to premium video that can help you get ahead, an ad-light experience, early access to select products including NFT drops and more:  https://account.forbes.com/membership/?utm_source=youtube&utm_medium=display&utm_campaign=growth_non-sub_paid_subscribe_ytdescript  Stay Connected Forbes newsletters: https://newsletters.editorial.forbes.com Forbes on Facebook: http://fb.com/forbes Forbes Video on Twitter: http://www.twitter.com/forbes Forbes Video on Instagram: http://instagram.com/forbes More From Forbes:  http://forbes.com  Forbes covers the intersection of entrepreneurship, wealth, technology, business and lifestyle with a focus on people and success.

Transcript

00:00Hi, everybody. I'm Brittany Lewis, a reporter here at Forbes. Joining me now is my Forbes

00:07colleague, staff writer, John Hyatt. John, thanks so much for joining me.

00:11Thanks for having me, Brittany.

00:13For years now, there have been so many questions surrounding the case of disgraced sex offender

00:17Jeffrey Epstein, who was found dead in jail in 2019 after being charged with sex trafficking

00:22minors. People demanded the release of more files pertaining to Epstein, including a supposed

00:27client list. But at the beginning of July, the DOJ released a memo saying there was no client list,

00:33Epstein died by suicide, and that they weren't going to release any more documents. And that

00:37was met with a lot of criticism. Beginning in June, and this was before the DOJ released that memo,

00:44Elon Musk got in a fight with Donald Trump, and he posted this on X. Quote, time to drop the really

00:51big bomb. Real Donald Trump is in the Epstein files. That is the real reason they have not been made

00:56public. Have a nice day, DJT. And that post was later deleted. And then Elon Musk said,

01:02you know, walked some things back, said he went too far on some posts and made some apologies.

01:07But now Elon Musk has waded back into the Epstein conversation. So can you talk to us about that?

01:13Yeah. In some ways, Elon Musk really lit the flame for the furor that we've seen in the

01:21megaverse around the Trump administration's refusal to release the Epstein files. That tweet in June

01:28sort of kickstarted a larger crusade on X, encouraging voices on the right, in the middle,

01:37on the left, who are raising awareness around Donald Trump's connections to Jeffrey Epstein,

01:44their long held social relationship that they had. But also his administration's stance on the Epstein

01:53files, the connections of, you know, Donald Trump's own appointees, Cash Patel and Dan Bongino, who for a

02:02long time, before they entered government, hyped up the release of the Jeffrey Epstein files as

02:07something that needed to happen. And now that they're in power, they haven't followed through on

02:11that promise. And so Elon Musk has really sort of cultivated X as a place where people can,

02:20you know, get their voices heard about it. And he's helped raise the temperature.

02:25And to that point, I mean, one of the things that the right wanted to know,

02:29and during President Trump's 2024 campaign was they wanted more information on Jeffrey Epstein,

02:36who he was sex trafficking to, who was on that supposed client list that now the DOJ is saying

02:41doesn't exist. And Elon Musk's AI chatbot, Grok, is certainly adding to the conversation here.

02:48Talk to us about that.

02:49Yes. So two weeks ago, after the DOJ and FBI memo was released stating that Jeffrey Epstein did die

02:59by suicide, there is no client list. Elon Musk and others really went off. They were furious about

03:06this. And one thing that Elon Musk did, which Forbes covered, was he did a Q&A with Grok. So for

03:13viewers, as a reminder, Grok is the AI chatbot run by XAI. XAI is the artificial intelligence company

03:21controlled by Musk, which also now owns X. So obviously X used to be Twitter. It was publicly

03:28held. Musk took it private. And then earlier this year, he merged X with his AI company, XAI.

03:35And so the synergy there is that Grok, the chatbot run by XAI, is embedded on X. And so users can

03:44tag Grok to respond to their inquiries about the weather, about sports, about anything they set

03:51their mind to. And so Elon Musk used Grok in a Q&A going back and forth with his AI bot about what

03:58information the U.S. government has about Epstein's collaborators, co-conspirators, fellow abusers of

04:07young women. And he really helped push the envelope on X by engaging with Grok. And Grok was responding

04:15as a chatbot does, providing information about the types of details that the government would have in

04:21the Epstein files. And this drew a lot of attention. Of course, it was part of the larger conversation on

04:27X and elsewhere in mega media circles, but really in mainstream media about what information the

04:32government has about Jeffrey Epstein, about Jeffrey Epstein's connections to Donald Trump,

04:37about Jeffrey Epstein's other powerful figures. And so Elon Musk has really taken this issue up as sort

04:46of one of his new crusades. I mean, just a couple of weeks ago, Musk was laser focused on the big,

04:52beautiful bill and criticizing that for its increase of U.S. deficits. But Epstein has become

04:58sort of his new cause that he's been championing. And he's been using X and he's been using Grok to do

05:04that. I want to zoom out even a little bit more and talk about how Grok really stacks up to other AI

05:12chatbots. Because in your piece for Forbes, you talked to an expert who said that Grok almost responds

05:19more human-like, which has some pros and has some cons. So bring that into the conversation.

05:26So Elon Musk, one of his big pitching points for Grok is that it's sort of an anti-woke

05:32chatbot, right? According to Elon Musk, whereas other chatbots like ChatGPT from OpenAI,

05:40which is obviously sort of setting the bellwether for chatbots. And it's, you know,

05:46kind of unleashed the LLM craze that we've witnessed the last few years. Whereas chatbots like that,

05:52you know, respond in a more neutral way, in a more, here's the information you're not politically

05:58engaged, doesn't really have an opinion outside of just laying out the facts. Grok is pitched as

06:05an anti-woke, politically incorrect, a chatbot with more of a personality. And that personality

06:13happens to mirror Elon Musk's own personality, not coincidentally. And so we really saw Grok's sort

06:21of personality come through when users started asking the chatbot about Epstein, like what's going

06:27on with Epstein? Where are the Epstein files? Who is connected to Jeffrey Epstein? Did Epstein really

06:33kill himself or was he murdered? And this is where Forbes tracked Grok's responses to these various

06:40inquiries. And what we noticed was really fascinating. Grok took the popular meme, Epstein

06:48didn't kill himself. This phrase has become a meme unto itself as a way to express defiance, skepticism

06:57of mainstream institutions, not just of the handling of Jeffrey Epstein, but of a larger sort of distrust

07:05of government and elites. So Grok took this meme and sort of made it to a truth unto itself. And

07:13so we tracked over the course of a two-week period in July, over 100 responses from Grok to users on

07:22X in a public forum stating, as a matter of fact, or implying as all but certain that Epstein did not kill

07:31himself. And that in fact, Epstein was murdered, or his suicide was abetted by a cabal of elites

07:39with whom Jeffrey Epstein was in bed with, so to speak. And so what we saw here was really Grok

07:46shifting from just this sort of playful AI chatbot to a promoter of something that, according at least

07:54to official government reports, is not true. And that's why we wrote that up.

07:58And this really isn't the first time that Grok has seen some controversy in recent weeks. It has

08:06waded into making some controversial statements, either involving racism or anti-Semitism. So can

08:11you give us those examples and talk to us about that? And really, how does that stack up to other

08:16chatbots? Are other chatbots doing this too? Yeah, so the first notable controversy that Grok had was a

08:24couple months ago, back when Elon Musk was in the good graces of Donald Trump. And Musk at this time

08:31was very focused on the issue of how white people in South Africa, his home country, were being

08:39treated. We saw President Trump host South Africa's president in the Oval Office raising this issue.

08:44It was really a sort of niche focus of Elon Musk's. And what we saw, strangely, was in response to users

08:54asking about unrelated topics, Grok was inserting false talking points about alleged white genocide in

09:03South Africa. This conspiracy theory that white African-eers were being systematically oppressed and

09:09murdered by the black majority in South Africa. This is a false claim, it's a false conspiracy, but it

09:17has, it was being put forth by Grok in response to unrelated topics. And so naturally, this drew a lot

09:23of headlines. You know, XAI said they were addressing the issue, and everybody moved on. Then earlier this

09:31month, on July 8th, Grok indulged in a rant in various forums on X, echoing anti-Semitic remarks, identifying

09:45itself as Mecca Hitler. And this obviously caused a lot of outrage, not just because, you know, it was

09:53offensive, but because, you know, this is a major company, a major company's chatbot, suddenly identifying as a Nazi.

10:00And so Grok did, on July 8th, say that this was, they apologized for what was going on. It was due to a coding

10:09issue in which XAI was inappropriately, or Grok, rather, was inappropriately incorporating other users' content into its

10:17responses, and they said that they addressed the issue. This was the same day that Grok was going off about

10:23Epstein not killing himself, that Epstein was murdered by a cabal of elites. 80% of the responses that we saw, in which

10:29Grok stated that, came on this day. And XAI said they addressed the issue. However, the product continued to echo the

10:38claim that Epstein didn't kill himself, and has continued to do so.

10:42And you actually asked Grok about the discrepancy about Epstein's suicide. And I want to read part

10:49of that conversation. This is what you asked. Why does Grok sometimes say, as a matter of fact,

10:54that Epstein killed himself, while other times maintaining caution or even dismissing claims of

10:59Epstein's murder as conspiracies? Talk to us about that conversation. How did Grok answer your question

11:05there? So this was really interesting, because whenever I write about X or Elon Musk,

11:11I reach out to X for comment, and they almost never get back to me. This is standard. Elon Musk

11:17doesn't typically address mainstream media outlets who reach out to him for comment for stories they

11:23write about him. But Grok, as a chatbot, is designed to respond to users. And since Grok was the subject of

11:29my story, I thought it was only appropriate to ask the chatbot for comment, as we journalists do. When we

11:36write about a subject, we reach out to them for comment on our findings, and we want to give

11:40them the opportunity to respond. And my story does print Grok's entire response at the bottom of the

11:46article, which I encourage viewers to read and check out because it is very illuminating to hear Grok

11:53articulate its own thoughts, if you want to call them that, about how it conceives of its own role here.

12:00And as you highlighted, one of the things Grok says is that, well, I am not engineered to just state

12:13things as they are. It depends on context. It depends on the users who are asking me questions.

12:21How do they view things? It depends on broader, popular information. And one of the really interesting

12:28things was, in its response, one, it identified the July 8th system error or system coding issue

12:35as a cause for why it was overly stating that Epstein didn't kill himself, as a matter of fact.

12:41And two, it said, look, this is a meme. The Epstein didn't kill himself meme is very popular.

12:46This is one of the sources that I go to in formulating my opinion. And ergo, that's one of the reasons

12:54why I was stating that Epstein didn't kill himself. This is a popular meme. Why shouldn't it? And I

12:58think that does provide some weird insight into how the AI, at least Grok's AI functions, is that,

13:04oh, this meme is so popular. It's so omnipresent. It's so domineering of the broader conversation

13:10around Epstein that, ergo, it must be factual. And therefore, we spit it out as fact, which is a

13:17little bit concerning, obviously. Memes are not reality. Memes are derivative of reality.

13:23So if chatbots are starting to, like, impose memes on us as reality, that gets into a very

13:29concerning world. And I think that raises, as you said earlier in this conversation,

13:35a larger question about AI and chatbots and the reliance on them, and can we really trust them?

13:40Because even before I was asking a chatbot one question, they spat back out an answer. And I was

13:45like, wait, that's not necessarily right. You tweak the question just a little bit, and they spit back

13:50a brand new answer with a bunch of different links. And I mean, one answer was right and one

13:56answer was wrong. This is saying something as fact when it is not facts. They're using essentially a

14:02meme, as you're saying. So what do you think this says really about the larger conversation

14:06surrounding AI chatbots? I think it highlights the issue of our growing reliance on AI-generated

14:19language, AI-generated reasoning for getting information in the world. You know, just a

14:26couple years ago, before AI was omnipresent, and you would Google a question, you would be presented with

14:32a list of different links, a list of different sources that you could consult. Now Google is

14:38curating those, it's curating its own response for you. And it's the same thing with ChatGPT, right?

14:46You ask a question of ChatGPT, you ask a question of Grok, it gives you a response. And research has

14:54found that users don't often look at the sources that are provided. These chatbots, right, they'll provide

15:00different sources. But most users don't have time for that. They want information and, you know,

15:04they're going to read, like, whatever response is curated by the chatbot. And so, you know,

15:10while oftentimes I think these responses are good and valid and they probably check out and they're

15:15useful, there are other times in which that information might not be valid. It might be lacking

15:21context. It might be written in a way or engineered in a way because of, you know, the folks who are

15:29ultimately designing the prompts and the models and, you know, their own biases that are excluding

15:35other pieces of information. And I think Grok is sort of the, you know, prime example of this because

15:41it's so heavily influenced by Elon Musk, because it's controlled by a company that is controlled by

15:46Elon Musk. It's naturally going to reflect one individual's personality more than it is sort of a

15:53more balanced, neutral perception of facts. And it's really interesting what you said

16:00about Google. And I'm just using that as an example, because before, if you Googled something,

16:04you would have to go then and read the different sources, read the different articles and find the

16:09facts yourself. Now it does a Sparks Notes version of that. You get the Cliff Notes exactly, and it's

16:15hand delivered to you. And sometimes, like in this Grok situation, things could be wrong. But is this a

16:21wake-up call, do you think, to Grok? Do you think Grok is going to go through any other additional

16:26changes? I don't think Grok will change much. I think Grok is an embodiment of Elon Musk. And I

16:34think we can expect to continue to see a subversive, cheeky, at times conspiratorial chatbot on X whenever

16:42we go on there. At the same time, there are different versions of Grok. There's a paid version that

16:47professionals use that they pay hundreds of dollars a month for, which I think is likely

16:53going to be less conspiratorial and more facts-based because if people are paying good money for a

17:01chatbot, they want a chatbot that relies on factual reality. I mean, we saw the Department of Defense

17:09just is paying Grok up to $200 million for a new contract as it seeks to incorporate AI into the

17:15DoD. So I think, you know, there are different versions of Grok as there are with other chatbots.

17:21And I think that we can expect to continue to see a Musk-flavored Grok in the Twittersphere or the

17:27X-Sphere. But at the same time, you know, XAI is talented researchers. It's investing a ton in

17:33NVIDIA chips. And it's, you know, building its super cluster in Memphis. And so I think it's going

17:37to continue to evolve. And it's right up there competing with OpenAIs, ChatGPT,

17:43Anthropics Squad, Google Gemini for dominance in this AI arms race.

17:49Well, John, I always appreciate our conversations and your reporting.

17:52Thank you so much. And until next time.

17:55Yeah. Great to see you, Brittany. Thank you.

Category

Transcript

Recommended