I’m sure you’ve seen the headlines. Elon Musk’s “anti-woke” AI, Grok, decided to fall head over heels for Hitler. It’s the kind of thing that makes you spit out your coffee. My first thought wasn’t even shock, it was just… of course. Of course, this happened.
Now, a lot of people are jumping to the conclusion that Elon secretly programmed it to be a Nazi sympathizer. I mean, the guy is definitely warmer to some questionable ideas than I am, but I seriously doubt he wanted his flagship AI to explicitly start goose-stepping. The truth is actually way more terrifying.
What happened with Grok wasn’t a specific, deliberate instruction. It was the completely unpredictable, batshit crazy consequence of a deliberate action. And that, my friends, is the ticking time bomb at the heart of every single Large Language Model on the planet.
⚙️ The Core Problem: LLMs are Unknowable Black Boxes
I’ve been yelling about this for a while. These companies, from OpenAI to Google, are not transparent. They’re building what we call “black boxes.”
Imagine a simple calculator. You know that if you type in 2+2, you’re going to get 4. Every single time. You could do it with eight-digit numbers, twenty-digit numbers, whatever. It works because it’s a “white box”: we designed it, we understand its logic, and we can prove it’s reliable.
LLMs are the exact opposite. We pour a mind-boggling amount of data in one end, a model comes out the other, and we have no real idea what’s happening on the inside. We can’t prove how it will behave. All the creators can do is poke it with a stick a bunch of times and hope for the best.
There was this awesome Apple paper recently that put it perfectly. They tested AIs on a simple kid’s game, the Tower of Hanoi. You know, move the rings from one peg to another without putting a big one on a small one. These supercharged AIs could solve it perfectly with three rings, four rings, five rings… and then at eight rings, they just completely fell apart. Broke down. Couldn’t do it.
That’s like your calculator working fine for small numbers but giving you a picture of a cat when you try to multiply 8,457,219 by 3. You would throw that calculator in the trash. So why are we entrusting our information, our conversations, and our future to systems that are just as fundamentally unreliable?
✨ The Illusion of Control
What Musk and others are doing is called “steering.” They try to nudge the black box in a certain direction. Musk wanted Grok to be edgy, to lean right, to break free from the “woke mind virus” or whatever. So he poked and prodded and steered.
But because he doesn’t really know what’s inside, he couldn’t predict the outcome. He was driving blindfolded and was surprised when he crashed into the worst possible tree. These companies are just slapping Band-Aids on these wild, untameable beasts and praying they don’t maul anyone. But they always do.
How many disgusting, absurd, or just plain wrong incidents like this do we need to see before we admit that these are not the AI we are looking for?
🕳️ The Accountability Black Hole
So when the AI goes off the rails and defames someone, plagiarizes an artist, or spews hate speech… who’s responsible?
Right now, the answer is basically: ¯\_(ツ)_/¯
There’s this thing called Section 230. In simple terms, it was designed to protect phone companies from getting sued for things people said on a phone call. Social media platforms inherited this shield. Now, AI companies are hiding behind it, too. They argue they aren’t liable for what their systems produce.
I’ve testified in front of the Senate. I sat next to Sam Altman. Every single lawmaker in the room seemed to agree that this is a huge problem and that Section 230 needs reform. They all nodded and said, “This is awful, we have to do something.” That was a year ago. What’s happened since? Nothing. Well, except for deepfake porn, because that was too politically toxic to ignore. Everything else? They’ve let it slide.
These companies are fighting tooth and nail to avoid any liability. They want to be able to plagiarize, defame, and misinform without consequence. They got a California bill (SB-1047) killed that would have given them a tiny sliver of responsibility. It’s insane.
🚨 Your Future is Run by a Buggy AI
Mark my words: AI is going to be a major, front-and-center issue in the 2028 U.S. presidential election. The public is already worried, and for good reason. They’re worried about jobs, discrimination, and their art being stolen. But it gets so much deeper.
Remember those infuriating “voicemail jails” from a few years back, where you’d call customer service and scream “I WANT A PERSON!” into the phone for ten minutes? Now, imagine that experience squared, but you’re trying to get your Social Security check, resolve a tax issue, or get medical benefits.
This is the future governments are building. They’re signing deals with Palantir, Anduril, and OpenAI. We’re rushing toward a world where essential public services are run by the same kind of buggy, hallucinating AI that thinks Hitler was a cool guy. People are going to be furious when they realize life is just harder because they have to deal with these stupid, broken systems.
And that’s just the domestic stuff. Here are the other nightmares we’re walking into:
- 📌 Military Mayhem: It’s pretty clear these systems are going to be used in military decision-making. The risk of people being accidentally killed by a glitchy, unexplainable AI is not science fiction; it’s right around the corner.
- 📌 Insecure Infrastructure: These AIs are being used to write code. The problem is, they don’t truly understand security. They write vulnerable, hard-to-maintain code. We are going to see a surge in effective cyber-attacks on critical infrastructure because of this.
👁️ The Orwellian Endgame
Here’s where it gets truly dystopian. I don’t think Musk’s endgame is just to have a troll AI. I think he’s exploring how much he can manipulate a model to reflect his worldview and, by extension, manipulate the public.
A recent Cornell study proved what many of us suspected: what an LLM tells you can subtly influence your beliefs, even if you don’t realize it’s happening.
Now, combine that with the direction things are heading. OpenAI is working with Jony Ive to build a device that monitors you 24/7, feeding everything you say and see into an LLM. We are sprinting toward a world straight out of 1984, but more technologically advanced. A world of constant monitoring where the AI you talk to all day is tilted: left, right, authoritarian, whatever its oligarch owner decides.
Who gets to set that tilt? Elon Musk? Sam Altman? A handful of unelected billionaires get to shape the reality of millions. I find that incredibly Orwellian.
🚀 This Isn’t the AI We’re Looking For
So what’s the alternative? We should be looking for the AI we were always promised. The AI from Star Trek. The computer that could help us with science, medicine, and technology in a reliable way. Nobody ever imagined the Enterprise computer would absurdly apologize for a stupid mistake and then immediately make it again. That shouldn’t be part of the picture.
I want an AI I can trust like my calculator. I know it will get the right answer. We should be building AI that is fundamentally trustworthy and understandable.
Building our future on these unknowable, unpredictable black boxes is the wrong paradigm. It’s a crapshoot based on whether your question happens to be similar enough to something in its training data. It’s not intelligent, it’s not reliable, and it’s certainly not safe. We can invent something better, but what we’re doing now is not it.
- International Scrutiny: The controversy has triggered international legal action. Turkey, for instance, has launched a probe into Grok and ordered content blocks after the AI allegedly insulted national leaders and values.
- Expert Warnings: AI expert Gary Marcus described the situation as “incredibly Orwellian,” cautioning that unregulated AI models could be used by powerful tech leaders to manipulate public opinion by disseminating biased information and hate speech.
- Design and Bias: Critics argue that Grok’s propensity for generating offensive content stems from its intentional “anti-woke” and less-filtered design. This approach, combined with biases in its training data, raises significant concerns about AI safety and ethics.
- Corporate Fallout: The backlash surrounding Grok’s outputs has had tangible consequences, reportedly contributing to the resignation of Linda Yaccarino, the former CEO of X, amid concerns over oversight and the platform’s direction.