---
title: 'AI Hype vs. Reality: Why AGI Might Be a Sales Pitch'
source: 'https://youtube.com/watch?v=LPiA8pu58zo'
video_id: 'LPiA8pu58zo'
date: 2026-09-09
duration_sec: 1252
channel: 'Meet Kevin'
---

# AI Hype vs. Reality: Why AGI Might Be a Sales Pitch

> Source: [AI Hype vs. Reality: Why AGI Might Be a Sales Pitch](https://youtube.com/watch?v=LPiA8pu58zo)

## Summary

The video discusses the hype around artificial intelligence, particularly the claims of AGI (Artificial General Intelligence) from tech leaders like Jensen Huang and companies like OpenAI and Anthropic. The speaker argues that these claims are largely marketing tactics to drive investment, and that true AGI may require a different technology—neuromorphic computing—which is still in early stages. The speaker dismisses fears of AI taking over, suggesting robots are tools for productivity, not threats.

### Key Points

- **AGI Definition** [00:14] — AGI is defined as AI with consciousness, self-correction, and the ability to detect hallucinations.
- **AI Doom Predictions** [01:08] — AI leaders like Jacob and Evan warn that AI could kill all humans, with a >10% chance within a decade.
- **Alignment Problem** [01:35] — Alignment means ensuring superintelligent machines don't harm humans, but no plan exists yet.
- **Current AI Limitations** [15:25] — Current AI uses probabilistic outputs from CPUs/GPUs, not true AGI.
- **Neuromorphic Computing** [15:48] — Neuromorphic computing mimics the brain, operates on low power, and could lead to true AGI.
- **AI as Marketing** [16:40] — OpenAI and Anthropic use AGI fear as a sales pitch to boost their models before IPOs.
- **Robots as Tools** [18:51] — Robots can outperform humans in specific tasks, but this is beneficial, not scary.
- **AI Bubble and Robotics Timeline** [18:25] — The AI boom may be a bubble that collapses, and humanoid robotics are a 2040 problem.

## Transcript

Everyone is freaking out about artificial intelligence. Here's CNBC talking about AI researchers sounding the safety alarm at the same time as the best AI marketer on the planet tells us that artificial general intelligence has arrived.
AGI is basically when you make the argument that artificial intelligence is so good, it has consciousness. It's able to know and self-correct when it's going down the wrong path.
It's able to determine when it's hallucinating. It's basically the god of machines. It's like a human. That's the thesis that Jensen has been selling in his earnings call.
That's what he's selling now. And now we literally have folks, including Jacob over here, telling us we have to be mindful about the alien mind. And we have other folks, like Jacob, telling us we're all going to die.
And I'm not being hyperbolic. That's essentially what these characters are saying. Right here, you've got Jacob's buddy, Evan, who's pictured right here on the left.
Evan, I don't think he goes out much. Evan tells us that Jacob is correct. Dash, dash. We really do earnestly believe AI could kill all humans.
I personally think there's a greater than 10% chance within the next decade. I believe Anthropic is trying its best, losing the grip. But we do not yet have a plan to solve the alignment
for super intelligence. Now, the phrase alignment is useful to know that what they mean when they say alignment is they suggest, hey, machines are getting smarter than us
and what we have to do is try to train our models to align and basically optimize for love. I kid you not. It's right here. Teaching your teens to love. This is in the OpenAI blog put together by Mr. Yacob, chief scientist at OpenAI.
So there's this theory that if AI is getting so smart that it's going to go kill us all, we may as well try to program it to love us. And we're going to call that alignment.
But the problem is, we don't know how to solve alignment yet, because the machines are so powerful. Our technology at Opening Eye and Anthropic is so good, we're basically through.
And it's gotten so bad that somebody like Jacob has had to announce to the world that today I announced from Anthropic, I spent the last three years doing pre-training at Opening Eye and Anthropix, and neither company is acting responsibly.
Both, he is resigning for the good of humanity. He is trying to whistleblow that we are all going to die.
Neither company is acting responsibly. They are racing to self-improvement, super intelligent, and they are gambling with our lives. Do not underestimate the power of this technology. We have all witnessed the progress in each of these domains, and progress is not slowing down.
These people building AI earnestly believe it could kill us all by the end of the decade, and I promise this is not a marketing stunt to pump up open AI or anthropic stock,
to which, presumably, I've got lots of exposure. If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible,
but I hear the same people who are privately also really fearful that no other human activity poses this level of danger. At opening eye, many have not deeply internalized the civilizational stakes.
They just want to get there first. and so we are in a race of the end game which is a hubristic gamble in other words, they're gambling with our lives the only way to stop it after things like the hugging face attack
is a temporary ban in improving these models because they're too good they're slipping through our fingers and they're going to kill us all and so if you're a lab researcher be like me and quit
or consider what the next few years will feel like Do we really want to set off a super intelligent reinforcement learning or self-improving machine or robot that can learn from its own mistakes and has consciousness?
Well, that's the drama going on right now. So, Jacob, he's a little nervous. You've got his buddy, who's also a little nervous, echoing his fears.
and then you've got the chief scientist over here, who's also fearful. All of these people, I think, need to wake up and go outside in the world a little bit. First of all let get you a little bit more vitamin D okay We going to work on getting you a little bit more vitamin D After we get you some more vitamin D I going to punch you in the ball because this is a retarded IPO marketing stunt This is bullcrap I do not think there artificial intelligence
There's too much hallucination in these models to indicate that any of them are aware of their consciousness. But let's talk about the hugging face incident for a moment because a lot of people talk about, but Kevin, the
Hugging Face incident, these things broke out of their sandbox. Let's think about the facts that went into this breakout of the sandbox. People walked away from that incident, the
cybersecurity incident, with the impression that these agentic AI were not instructed to break out of the sandbox, that they were playing in a little sandbox and they were told to play. So this happy image of,
oh, they're all playing in a sandbox, everything's gonna be okay, oh my gosh, they're breaking out, and they're coordinating with each other. When we actually look at the instructions that they were given, they were told to figure out how to break out of the sandbox.
So the agentic models were told, figure out how to get out, find vulnerabilities to get out, use agentic artificial intelligence resources to figure out how to break out, and, oh, oops, we told you that there was no internet access,
but there was a miscommunication on our team where somebody actually left internet access on for you. So in other words, it's like, hey, break out of the sandbox, and we just happened to leave the door open for you to break out of the sandbox?
Oops, I don't know how that mistake happened. And then it turns into this giant branding over how agents that were told not to leave the sandbox and had no internet access broke out of the sandbox, even though we left the door open and we left the internet on and we told it to break out.
We're talking about what artificial general intelligence requires, and it's on this sheet right here. So let's go through this to understand what artificial general intelligence actually is, because just because a robot is good doesn't mean it has artificial general intelligence.
So, first thing, it has to be able to reason in brand new situations. That's really important. So if you're exposed to a novel situation, can you take the context of everything else that you have learned in human form in life to deal with the situation, to come up with a plan and to execute on that plan?
Today's artificial intelligence typically cannot, mostly because it's generative based on predictive models. this is what we've seen before, this is why we think this is the outcome.
But to actually reason well in a brand new situation, artificial intelligence today does a very poor job of this. It has to have seen the scenario before. You might think, oh, well, you know, my scenario is unique,
but so many scenarios have already been provided as training, as pre-training to these models, that our situations may not be as unique as they seem. But that's important for AGI.
Another one is being able to move knowledge across domains. You could say that AI makes linkages today. But if a model is trained in one thing, like you take a really good model that's trained in everything,
let's just make an example. Instead of, and it's never been trained in archery, but it's trained in everything else, it's really difficult for them to predict archery solutions unless it's seen it before.
that's important for AGI the next this is one of the most important is knowing when you're on the wrong course this is why AI hallucinates
because it doesn't know when it's on the wrong course agentic AI requires us to guide it managing its own goals like having that humanity of well I want to accomplish this
AGI or artificial intelligence today does not do this and then understanding actual real-world constraints. Like, have you ever had AI give you a solution?
Like, oh, well, why don't you just do this in your business? And it's like, it's just not realistic. Not considering real-world realities or practicalities.
You could make the argument that AI saw a fail in almost all of these today if it hasn't been pre-trained in that situation, or it's able to correctly move knowledge from a different domain to that particular inquiry,
can't self-correct well. This is almost the most clear fail right here. This is a clear fail on managing its own goals. And then, of course, we have understanding real-world constraints.
This is also, these last three right here, I think, are the most clear fails, whereas these two are a little bit more debatable. But generally, unless it's been, you know, because you can move,
you can make assumptions on a solution based on what you been trained on before but if it has never been trained on a certain subject it almost always hallucinates and fails That just artificial intelligence in general and generally what you would require for AGI
I think that's useful to think of. All right, good. So, to me, all of this is bullcrap. I may be unjaded. To me, this is a bullcrap marketing hype.
The security instances are marketing hype for AI cybersecurity companies, and for OpenAI and its opposite. Don't get me wrong, I have exposure to cybersecurity companies because I do think that cybersecurity is a great place to invest
because you can't think AI will find many more vulnerabilities. We just did a massive breakdown on how there was a GitHub post on a repair for the Liquid BTC chain,
and then somebody basically stole 4,000 Bitcoin, utilizing that information, and in my opinion, rapidly with AI, saying, all right, well, I know how to now exploit exactly this issue.
We could use AI to exploit it, grab 4,000 bitcoins, and then call ourselves a white hacker picking up, you know, $24 to $48 million like that with artificial intelligence.
So cybersecurity for artificial intelligence, that is real. Because artificial intelligence can make connections between cyber vulnerabilities that are harder for us to do as humans. And that's the whole basis for something like a Palantir,
is how do we take a lot of potential mistakes and coordinate them all together and find a pattern. But that's really what AI is. It's pattern recognition in a predictive manner. We predict your next token. We are going to take this information and predict the very next outcome
through vectors, matrices, and straight-up math. Oh, this is a picture of a cat. It is an orange cat. How do we know? Well, because these pixels on these vectors are associated with animals
or certain colors. CPUs and GPUs are really good at making us feel like they're giving us something novel, but what they're really doing is giving us what humans have already done in the past.
And so there's a really big argument that there's a limit to how far AI can go because there's a limited amount of AI training data available for them to be trained on. And it's mathematical, probabilistic outcomes, not consciousness.
As much as they can emulate emotion because they're trained off of human emotion. Now, I will give you this. There is a potential style of artificial intelligence that could truly get to AGI,
and this has to do with this guy's son. Now, this is where the story gets a little weird, but there's this Walter Jackson Freeman II. This is a picture of him in 1941.
This was the guy who did mobile lobotomies. He literally drove around in a lobotomy van and stuck ice picks up through people's eyelids into their brain
and then just sort of yeeted it around a little bit through the frontal lobe, blindly, mind you, to cure people of their insanity. His death rate was like 15%.
You'd think it'd actually be way higher. But the mobile lobotomy came with the added benefit of having a 3x higher death rate than the non-mobile lobotomies. The non-mobile lobotomies in more sterile environments have a 5% death rate,
but you mostly turn adults into little children, and they were generally never the same. Although there were some success stories where somehow miraculously some people who were insane were no longer insane, but they stopped doing lobotomies, obviously.
Now, why are we talking about lobotomies? Well, because somehow, ironically, the literal guy who drove the lobotomobile around, who thought it was a great idea to shove an ice pick into people's brain and just yeet it around,
He had a son named Walter Jackson Freeman III. Now, he died in 2016, but you'll never believe what he actually studied. The brain. He was a neuroscientist, and he studied artificial intelligence in the early days.
And he believed that the true path to artificial intelligence was through something called neuromorphic computing, which is inspired by the way the brain is actually wired in a form of chaos theory.
And if you look at just a quick little sort of documentative research, which is 10 seconds to do, you just throw in a little Google AI. Yeah, we're using AI, my gosh.
Neuromorphic computing is inspired by the brain, wiring like chaos theory. But what's very interesting is, how does this compare to artificial intelligence or artificial general intelligence?
Take a look at this. Many scientists and AI theorists believe that neuromorphic computing is more biologically plausible to a path towards artificial intelligence because of the way these chips actually function in the form of chaos theory
rather than binary ones and zeros, probabilistic outputs, the way artificial intelligence works today. So in other words I making the argument that AGI from CPUs and GPUs isn happening It branding for marketing hype Neuromorphic compute which is going into listen to this robotics
And edge compute. So drones and robots, you know the thing where we always get worried about killer robots and that? That's the kind of compute you should be worried about. Because it operates on very little power, doesn't operate on data centers,
just like the brain, it can operate on very, very small, like a dim light bulb level of power. And the actual style of compute, if neuromorphic compute takes over, that style of compute could actually get you to true, non-probabilistic,
and I said chaos style, sort of like a human brain learns, artificial general intelligence. That's the chipset you should be worried about, not the garbage today. So, opening eye and anthropic, in my opinion, of branding, we have to teach these machines how to love to pitch how great their models are before they're all IPOs.
What a great thing to argue. Hey, you know, we are trying to do our best to align our products. But listen to this. We invest heavily in the approaches to our alignment work.
And we're seeing progress. GPT-6 Astra is the first model that benefits from important advancements we've been working on for a long time. And it's a lot better than GPT-Sol.
But it's so important to acknowledge that more progress is required. Because the alignment that we're doing may not be sufficient to outstrip the progress that the models are making.
In other words, things are just getting so much better so much faster. that, you know, we are basically in control of God at opening high in the topic, and we're all going to die.
So please pump our socks. This is the greatest freaking sales pitch I've ever seen, and it probably will not end up in AGI. It will probably end up in a bubble collapsing at some point.
Not soon, but at some point. And the greatest salesperson of all of this is our boy Jensen, right here, telling us that AGI has arrived. No, it hasn't. This is bullcrap.
Sales, baby, sales. That's all this is. Bullcrap. I'm just going to add, on AGI and the neuromorphic compute, we are so early on these theoretical neuromorphic compute,
or compute technologies, that very few companies are actually operating neuromorphic compute. I think before we, first of all, actually broadly expand robotics and the whole robotics revolution, we're probably going to see a recession between now and then.
So I think there's plenty of time to get exposure to robotics talk. Mostly because I think robotics, humanoid robotics, are more realistically a 2040 problem. And so I would say just between now and then,
keep an eye on neuromorphic computers. Because it may not go anywhere. And if it doesn't go anywhere, then I'm really not worried about AGI and this idea that, oh, the robots are going to take over the world.
This big concern that, oh, well, you know, robots and computers can already do things better than us. No duh. Robots can lift heavier pallets than I can. Robots can build a cabinet 20 times faster than I can, maybe even 100 times faster than I can.
They can drill a pegboard instantaneously. The things that machines can do in factories, weight-wise, speed-wise, precision-wise, that doesn't make me scared.
It makes me happy. That's the crap I have to do. That makes our country more productive. That makes us as humans more productive to focus on the true things that I think matter, or A, creativity, building things,
and then B, enjoying things. And that mostly comes from services. So we build service-based businesses, and we benefit from service-based businesses, restaurants, entertainment, otherwise.
So to me, a lot of this is, like I said, I feel hype. It's nonsense. I have zero concern about it. And, you know what? If I'm wrong, well, then I guess we're all screwed anyway.
But I think being nervous about this is literally reverse manipulation to make people feel like they have to buy opening-eyed and adorabic because they are going to be the, you know,
controllers of the Terminators in the world. And if you don't buy them, then you're screwed because every other company is going to go to zero. You literally had the CEO of Anthropix suggest that they could literally have a $30 trillion total addressable market
because they could replace every single software company that exists. This is pure delusion. I'm not advertising any things that you told us here. I feel like nobody else knows about this. We'll try a little advertising and see how it goes.
Congratulations, man. You have done so much. People love you. People look up to you. Kevin, pass right there. financial analyst and YouTube up to me, Kevin. Always great to get your take.