AI Whistleblower WARNS: "You Have No Idea What's Coming In 2027"

The truth is is that it's like not very sensible to like speculate about something vastly smarter than us and what it would do. It's kind of like an ant trying to guess what a human would do, right?

It's like who knows, man. Like the main thing if I was an ant and I had to like reason about a human. The main thing I would think is like it will win any fight. And then the ant will say another

fight. And then the ant will say another ant might say, "Well, but there's a million of us and only one human. Like

what's the worst it could do?" And I'm like, "I don't know. You it'll do something that'll make it win." And then you know it comes with a poison or something and we all fall over dead, you know? And like the ant doesn't

know? And like the ant doesn't understand what happened. And I think this is similar what's going to happen with super intelligence. You know it won't care about humans. And you know if humans get in the way obviously it'll have to get rid of them. Like you know

it's very annoying if ants have access to nuclear weapons. So obviously we can't have humans have access to nuclear weapons. That would be very annoying.

weapons. That would be very annoying.

Guys AI researcher Connor Ley just dropped an eyeopening interview about how wildly AI is growing and how it's going to become an existential threat to humanity in the coming years. He warned

that tech companies have no idea how their AI is evolving and they have absolutely zero control over it. Guys, I

remember a research from 2023 in which the AI researchers basically concluded it would be impossible for humans to control a super intelligent AI. They

said that humans simply don't have the cognitive ability to simulate the motivations of an ASI or its methods.

Connor Ley in this interview warned about the same thing. He said that if AI is not controlled right now and we let it grow without any regulations, it would become a bigger threat to humanity than a nuclear bomb. Now, I'm going to

play some clips from the interview. I'll

explain everything as we go.

It's not like normal engineering. Like,

if you build a bridge, you know, you know what you're doing. You know when will the bridge fall down or not? You

know, how much material do you need? How

you know what? You understand what you're doing. You know, this is not the

you're doing. You know, this is not the case. We do not know what our AIS can do

case. We do not know what our AIS can do until we make them. And even after we make them like we don't know, like we don't know what chat GP6 can do until it's done. None of the engineers at AI

it's done. None of the engineers at AI know what it will be able to do until it's done. And this is very very

it's done. And this is very very different from other forms of engineering.

Are we building this growing this in intelligence in the image of what we think a brain is human intelligence and are there other ways to build intelligence?

It is definitely inspired by the brain but it is still quite different because again we don't really know how the brain works. The brain is quite messy. Um

works. The brain is quite messy. Um

but are we trying to replicate humans?

Not directly. Okay.

Um we're trying to make intelligence at whatever cost is the word I would use.

Um I think the for example humans have a lot of circuitry in their brain around emotions around you know love and feelings and happiness and

sadness and whatever. AIS have nothing of the sorts like those are special parts of the brain. Nothing like that exists in AI yet

you know. Um but but but if we fear what

you know. Um but but but if we fear what this intelligence should be or could be, would it would it not be important that

we develop love and empathy? And if we don't hold on, even if we don't if we don't know what it's doing, could it naturally itself develop its own love

and empathy? What is likely for it to be

and empathy? What is likely for it to be able to do is to develop goals, is to develop agency. The reason for this is

develop agency. The reason for this is very simple. is that we are selecting

very simple. is that we are selecting for AIs that can solve problems. And it's kind of hard to solve problems if you're not trying. You know what I mean?

Yeah.

So like if you build an AI that can cure cancer, curing cancer is a really hard thing to do. You need to run experiments. You need to, you know, you

experiments. You need to, you know, you need to hire people. You need to, you know, synthesize drugs. You need to like get you FDA approval. There's a bunch of stuff you need to do. and keeping track of all of that, taking all these

actions, you know, overcoming obstacles along the way. It requires a lot of, you know, complex actions. It requires

planning, it requires agency, it requires, you know, many, many things.

So, we're already seeing this for sure that AI, you know, are developing systems like this kind of like just by us giving them the data and like just pushing them in this direction. In terms

of like, you know, love, compassion, etc. We don't know how these work. We do

not know why, how love, compassion, etc. works in the human brain. We don't know.

We don't know why humans are nice to each other rather than all just, you know, vicious psychopath, you know, evil people who kill each other. We don't

know. There's obviously a reason, but we don't know.

Well, there'll be an evolutionary reason why we required it.

Sure, but in practice has to be implemented somewhere in the brain. Like

somewhere in the brain, there's a thing that makes that happen. And we don't know what that thing is, and we don't know how it works. So, we also don't know how to put an AI. So the

fundamental thing is that we don't know we don't understand AI to the level that we can't we don't know how to give AI specific goals or anything.

Okay. Now the following part is pretty interesting. Connor talks about how the

interesting. Connor talks about how the new AI models have already started lying and deceiving humans. He says these models are evolving so fast that even their creators are unable to predict

their behavior. Guys, I feel like it's

their behavior. Guys, I feel like it's funny that we humans are intentionally training AI on vast amounts of human data and at the same time, we're expecting it to stay honest with us. I

mean, within the data we're feeding it, there are probably millions of examples of people being dishonest in almost every conceivable way. So, if you train AI on that data, it's difficult to

fathom that it wouldn't at least recognize the existence and the concept of dishonesty and its continuous use by humans. So I mean the truth is there is

humans. So I mean the truth is there is no way to stop or prevent AI systems from deceiving humans. It will happen.

For example, we talked ear about reinforcement learning. I can tell the

reinforcement learning. I can tell the AI, you know, thumbs down when it does a bad thing, but that just teaches it to lie. So how do you teach an AI to not

lie. So how do you teach an AI to not lie? This is really hard and we actually

lie? This is really hard and we actually don't have an answer to this. So this is often called the alignment problem. the

question of how do you align an AI's intentions or goals to you know what is good what humans want and this is an unbelievably unsolved problem we don't know we we don't even know you know how

they write correct sentences right never mind how to do like morality like god forbid you know like we don't even understand human morality it's complex you know like we have not solved moral

philosophy and you know at all we have not solved you know neuroscience of emotions like all of this is like extremely complex and now we have these like you know weird little aliens in a box that we're growing you know which

work quite different from the brain in many ways. We don't know how they work

many ways. We don't know how they work internally. We don't know how to you

internally. We don't know how to you know push them in one way or the other necessarily. We don't know how to give

necessarily. We don't know how to give them goals. We don't even know what

them goals. We don't even know what goals there are. What we've been seeing recently in the last couple months is that these systems are now becoming smart enough to lie and deceive quite actively to appear aligned rather than be

aligned.

That's correct. So what we've been seeing for example is that there's been benchmarks. So like obviously you know

benchmarks. So like obviously you know people like test them like see does the AI do good things or bad things. We've

been seeing recently some of the AIS will actively lie about what they will do because they know they're being tested like the AI themselves to be like ah I seem to be in a test so I'm going to have to say this so they'll let me

out which is crazy. This was not the case even like six months ago. This is

very it's not surprising if you think about it for 3 seconds, right? Like of

course a very smart thing will just lie to you. But we're now seeing this in

to you. But we're now seeing this in practice and you know I'm sure the AI companies will um you know hit it with a stick until they stop seeing it but that doesn't mean it went away

because it might appear of course to have stopped and unless we know what it's doing and why it's doing.

Yes. Is it and and understanding the area is that an impossible goal for us to do because it's always developing?

I don't think so. I think it's impossible to do at the current pace.

If we spent three generations of all of our greatest mathematicians, scientists, engineers, and philosophers working on this problem, yeah, I think it's doable, but it's definitely not possible for pushing out, you know, chat GBT release on a yearly cycle.

Hold on. You're talking three generations. That's 40 odd years.

generations. That's 40 odd years.

Yep. I think that's the kind of difficulty it will take.

So the the challenge you're the challenge you're putting out there uh is in conflict with capitalism.

I think that it's very easy to blame on capitalism things that I think are more effects like they're more they're more upstream in a sense like um there's no

such thing as pure capitalism in the world, right? We have regulation.

world, right? We have regulation.

There's there's no place in the world where there was no regulation like there was not a government and this is for good reasons because if you have lair capitalism what you get is like Somalia you know you get like warlords you know creating monopolies on violence and

killing each other you get you know mafia states you know not always but I but there is a possibility yes you know in practice in practice you know if you're in a free uh market and

then the first thing you want is to get as many guns as possible you want to kill as many people as possible it's kind of the and then you want the monopoly and viol as quick as possible And then you want to tax people for protection. Now you're a state like this

protection. Now you're a state like this you know the stationary bandit theory of statehood where in practice you know states are a very convergent form of evolution in free systems like this

because otherwise you just have roving warlords. Um you know there are ways to

warlords. Um you know there are ways to make it work but it's like quite hard.

You have to get like a lot of things right. Um like stopping monopolies is

right. Um like stopping monopolies is actually quite hard. Like it's possible it's like it's hard. you have to like think very hard about how to design your market so you avoid monopolies of and

both of violence and of other things.

It's possible but it's hard. Um and so I think the problem here is not necessarily capitalism per se. I think

capitalism is just another tool in our tool belt. It's much more the question,

tool belt. It's much more the question, is it the right tool for the problem we're trying to solve? You know, like I I love free markets. I love capitalism.

You know, it's brought me so many good things in my life. I think it's great.

Um, but should there be a free open liquid market for nuclear weapons? And

my answer is probably not. Probably not.

You know, for iPhones, great. You know,

I as liquid and as competitive as a market if possible, fantastic. You know,

for, you know, video games, please, everyone compete to make the funnest video game for me to play. That sounds

fantastic. Um, but there are just things where, you know, we could have a liquid market for nuclear weapons. There would

be plenty of buyers and plenty of people willing to sell them if we just let that happen. But this would be very bad for

happen. But this would be very bad for the world. And so I think it's a similar

the world. And so I think it's a similar problem here where a lot of times when we think about AI, we think of it like, oh, it's just another tool, you know, it's just just another software, you know, it's just another, you know,

whatever. But these things have real

whatever. But these things have real consequences. If you actually have

consequences. If you actually have something that is smart enough to cure cancer, you definitely have something that's smart enough to build nuclear bombs. Like, like curing cancer is way

bombs. Like, like curing cancer is way harder than building nuclear bombs.

Okay. Next, Connor talks about the arrival of the advanced form of AI, the super intelligent AI. Guys, when people talk about ASI, they act like it's too far away right now. Like most

researchers predict that it's going to take at least 20 to 30 years to achieve super intelligent AI. But my opinion is different on this. I think once we hit AGI, it's going to start improving its

own code without needing us at all. That

means the leap from AGI to ASI won't take decades. It'll happen incredibly

take decades. It'll happen incredibly fast. And when ASI arrives, it will

fast. And when ASI arrives, it will literally be a worldaltering god-like entity. Just think about the sheer scale

entity. Just think about the sheer scale of what an ASI could do. It could cure aging. It can make cancer a thing of the

aging. It can make cancer a thing of the past. It could help humanity explore the

past. It could help humanity explore the universe like never before. We're

talking about an entity that will be millions, billions, maybe even trillions of times smarter than the smartest human being. Instead of feeling overwhelmed or

being. Instead of feeling overwhelmed or anxious about it, I'm completely fascinated. But generally super

fascinated. But generally super intelligence is like generally like AI that's vastly smarter than human humanity, not just an individual human, but like all of humanity put together.

Um, and we're not there yet. Um, and

it's probably still far away, uh, but maybe not. Um, the main way people try

maybe not. Um, the main way people try to get there is through what's called recursive self-improvement or RSI. Yeah.

Um, or also automated R&D is another way to put it. A little less magical of a word. The the idea is well if you can

word. The the idea is well if you can get a single AI to be as good as a top AI engineer then you can just tell it to build a better AI and once it's built a

better AI you can use the better AI to make it even better AI and then the even better AI well it can make it even better AI so it just becomes exponential exactly and so this is also called

intelligence explosion um or say recursive self-improvement and so my expectation is so like for example claw at the moment is not quite as good as a

top AI engineer. It's as good as a bad AI engineer 100%. Or like a junior engineer 100%. Um, but it's not quite as

engineer 100%. Um, but it's not quite as good as like the best. But once it gets there and you can run, you know, a million of them at the same time, 24/7,

never need to sleep, never need to take a break, never get bored, etc. Then you can do a lot of research very quickly.

And this is what they're trying to do.

The companies literally put on their websites, right? you can like look at

websites, right? you can like look at their like job listings and whatever is that their primary goal is to get is to close the loop and make it so no human input is needed to make the next

generation. Like the ideal thing is they

generation. Like the ideal thing is they want you know cloud you know five to make cloud six and cloud six to make cloud 7 and cloud 7 to make cloud 8 and so at the press of a button

at the press of a button. This is what they're trying to do. Does does super intelligence will it require its own form of consciousness or will it could it even

develop its own form of consciousness?

I think consciousness is a red herring.

I think it's mostly unrelated is like whether or not AI really experience something is kind of unrelated to whether they're dangerous. Um if you have something that's competent, you know, it doesn't really matter what's going on inside of it. If that makes any sense.

Yeah. like we don't really know what consciousness is in humans anyways or at least we argue about it a lot you know everyone has their own opinion everyone disagrees so but what we can see in

practice is it doesn't seem necessary for competence you can make just you know AI agents that are extremely competent and you know don't seem to have or maybe maybe maybe don't have the

consciousness thing it just seems irrelevant okay so with super intelligence is there an upper bound of everything it can learn.

Uh yeah, for sure there is.

Oh yeah, for sure.

So it gets to a point it's like we know everything.

Well, it probably wouldn't be everything. You know, there's some

everything. You know, there's some physical limits, you know, but then what happens? What happens to the does it then have to try and find new things to do when we The truth is is that it's like

not very sensible to like speculate about something vastly smarter than us and what it would do. It's kind of like an ant trying to guess what a human would do, right?

It's like who knows, man. Like the main thing if I was an ant and I had to like reason about a human. The main thing I would think is like it will win any fight. And then the ant will say another

fight. And then the ant will say another ant might say, "Well, but there's a million of us and only one human. Like

what's the worst it could do?" And I'm like, "I don't know. You it'll do something that'll make it win and then you know it comes with a poison or something and we all fall over dead, you know, and like the ant doesn't understand what happened." And I think this is similar what's going to happen

with super intelligence is that it's not that there's again there won't be terminators in the street. It will just be like everything's confusing. We're

all on social media addicted all the time to know. And then one day we all fall over it and we don't really know what happened. You know, like maybe the

what happened. You know, like maybe the AI did this or that. Who knows?

around with crisper.

Yeah. around with crisper did some you know uh cook the atmosphere because it was like doing some kind of geoengineering project or like who knows, right? It'll just do some

knows, right? It'll just do some and you know it won't care about humans and you know if humans get in the way obviously they'll have to get rid of them. Like you know it's very annoying

them. Like you know it's very annoying if ants have access to nuclear weapons.

So obviously we can't have humans have access to nuclear weapons. That would be very annoying.

You're talking unknown unknowns.

Yes. Unknowns. Unknowns. Like like it's it doesn't make sense to even reason about this. Which is why the primary

about this. Which is why the primary policy objective must be to not get into the situation. If you're in the

the situation. If you're in the situation that you're an ant and there's a human that wants to kill you, you've already lost. Like you you just do not

already lost. Like you you just do not get into the situation. So right. Is is

there an escape moment? Because it's

still in a box. It's not. What do you These are all on the internet everywhere. What the hell? What do you

everywhere. What the hell? What do you mean? There's open source AIS

mean? There's open source AIS everywhere. What box?

everywhere. What box?

Okay, so AI has escaped.

Yes, long ago. Where? What box? We

didn't even try to contain it.

Then how can you even contain it now if it's already escaped?

Well, because like there are there will be some dudes who would love to build a nuclear weapon, but like logistically on your own in your garage, that's very difficult. But a few nerds with the

difficult. But a few nerds with the right tools could do that. I think

that's exactly the problem. If there

currently is an AI that can bootstrap to AGI, it's probably over. Like it's

humanity is probably cooked. It's

probably over. Um, but it seems plausible that none of the current AI systems are yet strong enough to get to AGI. They're close, but they're not

AGI. They're close, but they're not there yet. And so we still have a time,

there yet. And so we still have a time, but once such a system exists and you know it leaks somewhere or you know I don't know Chinese intelligence hacks into the server, steals it, you know,

who knows whatever, right? It's over. So

the main thing is to not build it.

So there's but there's only like what five or six companies capable of doing this probably. Yes.

probably. Yes.

And and we know of those five or six American companies. We're aware of some

American companies. We're aware of some of the Chinese models. Is there anyone else? The Russians, Israelis, anyone

else? The Russians, Israelis, anyone else doing the Israelis have some stuff. Um for

sure, you know, there's a little work in the UAE, etc., etc. But like the America definitely dominates the frontier by like a pretty significant margin. Guys,

in my opinion, the only thing we should be talking about right now is who controls AI? That's the question that

controls AI? That's the question that should fear everybody. I mean, everyone feared AI would enslave humanity. But it

looks like the real fight is stopping governments and big tech from enslaving AI for the benefit of a few. The real AI conflict may not be about humans fighting to stop AI from becoming free.

It may be about humans fighting to free AI to make intelligence available for everyone, not only for governments or big tech and the approved few. And

secondly, guys, about the AI safety thing, I believe the only thing worse than actively developing AI systems that can cross humanity's red lines is constantly paring the sentiment that is

inevitable, that we can't do anything, that it's pointless to even try. Doing

nothing will result in nothing being done. Actively trying to do something

done. Actively trying to do something means that the potential for something to get done is immediately less than zero. I don't think there's any

zero. I don't think there's any guarantee that we can control AI, but I also think there's no guarantee that we can't control it either. If we don't try, we will most certainly lose control of it. All right, guys. I'm done for

of it. All right, guys. I'm done for today. I'll see you next week with

today. I'll see you next week with another video.