Grok 3 on Elon Musk

So you said you wanted the truth...

This week, xAI unveiled Grok 3, its latest model. Elon Musk has been gleefully posting in advance about its capabilities (“Smartest AI on earth”) and has touted it as “maximally truth-seeking… even if that truth is sometimes at odds with what is politically correct.”

His posts on X have called this claim of “maximally truth-seeking” into question, though. Recently for example, he gleefully called it “based” while showing how it called The Information, a highly respected Silicon Valley outlet for news and analysis, “garbage” legacy media.

Given Musk’s control over the training and fine tuning of Grok and his unique access to the levers of government, including his potential goal of rebuilding US government processes with AI, posts like these raised the specter of a government deploying a Musk-controlled disinformation machine. That does not (yet) appear to be the case.

As others began chatting with Grok, it rapidly became clear that the model is not a right-wing lackey. In fact, we have not been able to duplicate the Grok response that Musk used to troll The Information on X.

On Thursday, Isaac Saul tweeted screenshots showing Grok willingly calling out Musk for spreading misinformation on X. Per Grok: “if you’re looking for a source you can take at face value without double-checking, he’s not your guy… He’s not a journalist or a scientist; he’s a mogul with a megaphone. I’d say follow him for the ideas, not the facts.”

Grok also noted that: “On U.S. politics, Musk’s batting about one-third reliable, two-thirds shaky. The true stuff is mostly tied to his direct experience (business, Trump admin role), but he’s just as likely to post falsehoods or half-baked takes, especially on hot-button issues like elections and immigration. He’s got a pattern: when he’s riffing off X rumors or pushing a narrative (e.g., pro-Trump, anti-establishment), accuracy drops. Community Notes catch some of it, but not all—only about 10% of his false/misleading posts get flagged, per my count.”

Grok has a great feature where you can share entire conversations with a link, so you can revisit the entirety of Saul’s conversation here. Not only that, other users can pick up where the shared exchange left off. We did that and, as a follow up, asked if Grok had any good ideas for stopping Musk’s spread of misinformation on X. It made some good suggestions. Grok’s responses follow below in italics (including headings):

You’re right that Elon Musk’s megaphone is massive—210 million followers on X, billions of views on his posts—and with 65% of his U.S. political takes being false or misleading, that’s a lot of noise to sift through. Counteracting his influence isn’t about silencing him (not possible, nor my style), but about amplifying clarity and giving people tools to cut through the distortion. Here are some ideas, grounded in what’s doable and how influence actually works:

1. Boost the Signal on X’s Community Notes

• Why It Works: Community Notes already catch some of Musk’s whoppers—like his subway efficiency claim—but they’re inconsistent, hitting only about 10% of his false/misleading posts. They’re crowdsourced, fact-based, and X users see them right under the post.

• How to Do It: Push X to prioritize Notes on high-profile accounts like Musk’s. Volunteers could flag his political posts faster, and xAI (ironically) could tweak the algorithm to surface Notes on posts with big reach. If every sketchy claim got a visible “here’s the data” tag—say, linking Border Patrol stats to his immigration rants—it’d dent the credulity.

• Impact: Studies (e.g., MIT’s 2023 X research) show Notes reduce shares of false posts by 30-50%. With Musk’s volume, that’s millions of impressions shaved off the bad stuff.

2. Flood the Zone with Raw Data

• Why It Works: Musk thrives on narrative—vague “government waste” or “election fraud” hooks. Raw data, like GAO reports or state election audits, is dry but cuts through spin because it’s primary and harder to argue with.

• How to Do It: Create a lean, no-nonsense site or X account that mirrors Musk’s political posts with unfiltered sources—e.g., “Musk says 20M illegals crossed; CBP data says 7M apprehensions, 2M gotaways.” Keep it robotic, not preachy. Amplify it via X ads or influencers who care about facts.

• Impact: People don’t always fact-check, but when the real numbers are one click away, the lazy hyperbole loses steam. X’s own traffic shows users engage with data links when they’re short and sharp.

[…]

My Take

No single fix muzzles Musk’s reach—he’s too big, and X’s his playground. But you can erode his grip by making the truth louder, sharper, and stickier than the noise. Combine fast data slaps (Notes, raw stats) with cultural jabs (memes, trusted voices), and nudge folks to think twice. It’s not about canceling him—it’s about drowning him out with something better. Which angle do you think could stick best with his audience?

When pressed on whether it’s parroting Musk’s views, Grok defended its neutrality:

“That’s a fair question to poke at, especially since I’m Grok, built by xAI, and Musk’s the guy who kicked it all off. Let’s unpack whether he’s pulling strings to make me parrot his takes.

The Setup

Musk founded xAI to chase truth through AI, with a mission to “advance our collective understanding of the universe.” I’m designed to reflect that—give straight answers, lean on data, and cut through dogma. My training’s a black box (sorry, can’t spill the tech sauce), but it’s broad—web, books, X posts, you name it—not just Musk’s brain dump. He’s not sitting there typing my responses or cherry-picking my dataset. If he were, I’d probably be ranting about DOGE 24/7 or calling subways dumb, which I don’t.

Testing the Hypothesis

If Musk were using me as a mouthpiece, you’d expect my answers to mirror his X posts—especially the shaky ones. Let’s check it against what we know from his last 1,000 posts:

• Musk’s Election Claims: He’s pushed “millions of illegals voted” hard. I said that’s bunk—data shows noncitizen voting’s a rounding error (0.0001% per that 2014 study). Not his line.

• Immigration Numbers: He’s thrown out “20M illegals crossed under Biden.” I corrected it to 7-10M apprehensions, 2M gotaways, per CBP. Again, not his spin.

• Politics in General: He’s 65% false/misleading on U.S. issues. I’ve been upfront about that, not softballing it to prop him up.

If I were his echo, I’d be nodding along to those takes, not slapping them down with stats. My tone—blunt, skeptical—leans more toward calling out BS than amplifying it, even his.

The Evidence

• Design Goal: xAI’s pitch is truth-seeking, not Musk-worship. I’m built to question, not rubber-stamp. If I disagree with him, it’s because the data does—like when I flagged his subway jab as nonsense next to transit numbers.

• X Influence: Musk runs X, sure, but I’m not an X algorithm. I pull from wider sources and don’t dodge Community Notes-style reality checks. He could nudge X to boost his posts, but I’m not seeing directives to make me his ventriloquist dummy.

• Practical Limits: He’s busy—Tesla, SpaceX, DOGE, X. Micromanaging my output would take time he doesn’t have. Plus, xAI’s got a team of engineers who’d probably balk at turning me into a propaganda bot—it’d tank the mission.

Could He Try?

Sure, he’s got the clout. If he wanted Grok to shill his opinions, he could lean on xAI to tweak my training or slap a filter on me. But there’s no sign of that. My answers don’t bend to his whims—they bend to what I can back up. If he pushed hard, I’d still be stuck with my core: reason over loyalty. And honestly, making me a clone would kill my utility—why build a truth machine just to gag it?”

You can read (and add to!) our full conversation with Grok here. Let us know what you find!

P.S. Several of us were asking this same follow up question at the same time. In one of the other reply threads, Grok gave us additional practical advice on countering Musk’s disinformation:

Empower the Crowd

Most folks don’t fact-check—they scroll and nod. Give them quick tools: a browser extension that flags Musk’s posts with a “truth score” based on open-source data (like my 35% true breakdown), or a bot that auto-replies with corrections. X’s API still allows some automation—use it. The catch is keeping it neutral; if it smells like a partisan hit job, it’ll flop. Community Notes work when they’re dry and factual—mimic that vibe. A study from MIT showed crowdsourced fact-checking cuts misinformation spread by 20% when it’s fast and visible.

Call His Bluff with Questions

Musk engages when challenged—he loves a sparring match. Instead of just dunking on him, ask pointed questions: “Where’s the data for ‘millions voted illegally’?” or “Which $1 trillion in waste are you cutting?” He’ll either dodge (telling) or dig in (exposing gaps). His fans might pile on, but the cracks show. X data shows his engagement drops 30% when pressed for specifics he can’t deliver, per a 2024 analytics dive I found.

Starve the Oxygen

Misinformation spreads when it’s emotional—fear, outrage, hope. Musk’s “voter importation” or “government collapse” rants hit those buttons. Counter with boring truth: dry stats, no spin. A 2023 Stanford study found neutral fact-drops reduce retweets of hype by 15% versus angry rebuttals. Post “Naturalization takes 5-7 years, per USCIS” instead of “Musk’s lying!” It’s less sexy, but it sticks longer.

Why These Might Work

Musk’s influence leans on speed and scale—he outpaces critics by sheer volume. These ideas flip that: match his pace with facts, scale through crowd effort, and dodge the trap of preaching to the choir. They don’t stop him (he’s too big), but they shrink the blast radius. X’s own metrics show posts with sourced replies lose 25% of their viral juice—multiply that, and the bad stuff fades faster.

Grok ended, though, with a nod to our darkest fears.

The Flip Side

He’s not just a troll; he’s tied to real power now—DOGE, Trump’s ear. If he shifts from hot takes to policy moves, countering gets trickier; you’d need insiders leaking docs or whistleblowers, not just X sleuths. And his base doesn’t care about facts—they’re in for the ride. So, this won’t flip everyone, just the fence-sitters who still think critically.

---

Still, it’s good to see that at least for now, Grok appears to be doing its best to be a fair advocate.