Sept. 10, 2026
How Concerned Should We Be? Apparently VERY. AI “Could Kill Us All By The End Of The Decade”
Apple Podcasts 
Spotify 
Castro 
RSS Feed Most of us have seen the frightening warning made by an AI researcher who left his position at Anthropic because of his concerns for the world’s safety given the increasing capabilities of artificial intelligence. And what’s even scarier, several other researchers and leaders in the industry are backing up those claims and concerns, saying AI technology may soon be able to hack anything and unless we slow down, we are literally gambling with our lives. Those with insider knowledge believe that AI is mankind’s greatest immediate threat, the question, what can be done?”
See omnystudio.com/listener for privacy information.
00:00:02
Speaker 1: Hey there, folks. It is Thursday, September 10th, and you remember that whole scenario that was the premise for all those Terminator movies with Arnold Schwarzenegger? Well, that is going to happen to us in a matter of years, or at least there's a decent chance of it. That is what we are being warned, and quite frankly, a lot of people are taking it seriously. And with that, welcome to this episode of Amy and TJ, an episode that, Robes, you said originally you didn't want to do. Because so much of what you were seeing sounded alarmist out there in some of the coverage we watched late yesterday and into this morning. However, after reading up on this, you said, holy shit, we need to talk about this.
00:00:43
Speaker 2: Yeah, because you start to see the headlines and look, I get it. It sounded clickbaity. I thought it was alarmist. I thought it was fear mongering. And then you start actually reading. What not just one, but several folks who have intimate knowledge about where we are right now with artificial intelligence and where we're headed. I was a little uncomfortable when I saw some of the headlines. I'm downright frightened after reading what several people have said.
00:01:18
Speaker 1: Okay, and these aren't just people. These are some of the top researchers at the top. AI firms in the world. So what we thought we'd do was hop on here, actually, and try to make this as digestible as possible, because we had to make it digestible as possible as well. After seeing what many folks probably did, one guy in particular, Robes, he did as many rounds as Kevin Reddingfield last night. He was on everywhere talking about this, and the guy we're talking about is a guy who said he has resigned. He has resigned his position from Anthropic. A lot of people know that name. Yes, it is one of the leading AI firms in the world. Said he resigned because, Robes, he is trying to draw attention to the fact that AI and AI technology is moving in such a direction that it could outpace human intelligence and lead to the end of humanity. That sounds big, Robes. People are repeating and agreeing with what he's saying. You got to pay attention. This ain't just sci-fi stuff he's talking about.
00:02:27
Speaker 2: It sounds like the premise of a movie, a movie we've all seen. And yet he is saying this is real life. Jacob Coxon is who we're talking about. And yes, he said... that AI could kill us all by the end of the decade. That reads to me in the next four years, that's what he's saying, that the end of the world could come about or the end of humankind could be the result of the technology that's being rapidly developed as we speak. And that unless there's some intervention, unless we slow things down, unless there's some sort of massive global coordination, We're all in trouble.
00:03:10
Speaker 1: Please, folks, don't roll your eyes at this. This is actually quite serious. And we made the reference at the top ropes to Terminator. I assume I watched all those movies more than you have. But the premise, everybody kind of understands. They develop this computer system, Skynet. It begins to learn. at a faster rate than humans can keep up with. It becomes self-aware and launches nuclear weapons, destroys humanity. That's the basis of the movie. Robes, that's the basis of these warnings. We are developing in such a race to get to the most advanced AI technology that we're not stopping for a second and recognizing the safety implications. And this guy, Coxon, came out and said he has worked, yes, he's worked for Anthropic, the other one with OpenAI. He accuses both of these companies of putting aside that safety concern for the sake of winning this race.
00:04:03
Speaker 2: It's frightening because, okay, so you have Jacob Coxon come out and we can read exactly what his post says because it's frightening. And it's not as if other people came out from the industry and said, oh, he is totally overreacting. This would be worst case scenario. No, we have folks who are currently employed and are researchers saying, yeah, what Jacob is saying is real. What Jacob is saying is correct. And when you start giving percentages that are higher than anybody is comfortable with, that what he is predicting could come true, we all need to be paying attention to Our government needs to be paying attention. And look, we can have this larger conversation. We can't even agree on things in this country, how to proceed, how to enact laws, how to create laws, how to get along. How are we supposed to coordinate with China? How are we supposed to coordinate with sworn enemies of this country?
00:04:56
Speaker 1: And China, that's a good you mentioned China. That is our primary competitor. Yes, we have American companies going at it over AI, but all of the American companies are being encouraged by the American government to get there before China does. This is a race to develop the most advanced AI. This, again, I'm reminding Robes, this is not a new debate, but it's entered the mainstream in such a way now that you have this guy coming out and you have others. They're calling him a whistleblower, but he resigned and just came out and issues the same kind of warnings that have been out there for a while. The fear of this idea of super intelligence. Yes, we're talking about Skynet. You're going to hear this word thrown around a bit, superintelligence. This is the idea. We haven't gotten there yet. They say we actually do not have AI that qualifies as superintelligence. We're talking about a computer that can on its own begin to learn without any human interaction, without any human direction for what it's supposed to be doing. What they say these things do, they start or they will start, and they're trying to get them to this place where they can start learning, developing on their own. And Robes, how they say it, The thing learns to make its own advanced replacement.
00:06:08
Speaker 2: What?
00:06:09
Speaker 1: The idea that the technology can do this and what they have been doing is using AI to develop faster AI. Because AI is faster than a human. So they develop the AI and then just say, hey, go develop more AI for us. At some point, you could possibly lose control.
00:06:26
Speaker 2: Absolutely.
00:06:27
Speaker 1: That's insane. This is real.
00:06:29
Speaker 2: Isn't this what you just described? That is what recursive self-improvement is.
00:06:33
Speaker 1: There it is.
00:06:34
Speaker 2: And that is a term that we should all be very familiar with because this is what we're talking about. This is the scary stuff where you have an artificial intelligence system that is fully autonomous. And it designs then, yes, more advanced versions of itself and not only would replace any sort of human interaction, but could basically, if it decided, it was in the best interest of of its next generation to eradicate humankind, it could. And yes, nuclear weapons is one of the premises that folks have floated. But you talk about just think about if you've got a computer system that could take over the world, you could take out the electrical grid, you could take out our financial systems. When you start creating massive havoc like that on the world where we can't function like we do, we can't turn on the lights, we can't turn on the water, we can't pay our bills, the entire world would literally crumble.
00:07:30
Speaker 1: This is real, folks. This is not alarmist. This is actually quite real. So this researcher comes out, Robes. He says he resigned. That's one thing. But he didn't go quietly at all. He put out a pretty extensive post that got a lot of people's attention. And Robes, a lot of people's attention, not just in media, he put out posts that have been seen since Sunday 100 million times. This has got everybody's attention, and it should. He is a three-year, he said he's been as a researcher between these two companies. But Robes, he started with his posts. I included the most significant ones that got a lot of attention here. But he starts to explain and break down why. Robes, it's hard because we've only seen this in movies. It's hard to think we're there.
00:08:15
Speaker 2: Yeah.
00:08:16
Speaker 1: And if you don't think we're there, folks, this has already happened at least on two occasions that AI models from OpenAI. What was the other one? Was it Anthropic? At least OpenAI. But on two different occasions, we've had incidents where AI models on their own hacked into other systems that they weren't asked to do, weren't directed to do. They were just trying to complete a task and said, hmm, I can't figure out how to do it. Let me go over here and break into this other system to try to figure out this other. Nobody asked it to do that. This has already happened. And Robes, that was an innocuous, right? No harm, right? It wasn't malignant. But what if it decides you give it a task? Well, the way to do that, I am going to have to turn off all electricity on the eastern coast. OK, and you can't stop it. Oof. This is scary movie stuff. But here was Jacob Cox. And you got it in front of you.
00:09:10
Speaker 2: Yeah, I do. I have it right here.
00:09:11
Speaker 1: His post.
00:09:12
Speaker 2: He said, I resigned from Anthropic today. I spent the last three years doing pre-training research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving super intelligence and gambling with our lives. More thoughts below. Hmm. Do not underestimate the power of this technology. These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources. We have all witnessed the process, the progress in each of these domains, and progress is not slowing. The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt. If anything, many executives and senior researchers will couch their phrasing in the No other human activity poses this level of danger.
00:10:20
Speaker 1: Okay.
00:10:21
Speaker 2: My God.
00:10:22
Speaker 1: And when you see him, he has glasses on, looks like a researcher, looks like a, how would you describe him, Robes? He's somebody you would listen to if he's talking about tech stuff.
00:10:32
Speaker 2: Yeah, he's a 27-year-old, looks like a 27-year-old genius. I mean, that's kind of what he looks like. He looks like somebody who has been in the trenches and certainly knows what he's talking about. And even if his looks alone or what he has to say doesn't move you, yes, his fellow researchers just might, who are actually still currently employed by by Anthropic.
00:10:52
Speaker 1: So before we say whatever to him, it wasn't that long, another Anthropic researcher jumped on, said, quote, Jacob is correct here. We really do earnestly believe AI could kill us all humans, kill all humans. I personally think it is greater than 10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for super intelligence and are not clearly on track to. Robes, these are anthropic guys coming out saying we are scared of what we're doing over here.
00:11:25
Speaker 2: And everyone here is scared of what we're doing. Another anthropic researcher.
00:11:29
Speaker 1: This one is, this was the best. This is the one, if anybody wants. This breaks it down maybe in the simplest of terms that I've had. I'm going to hold on to this. If you ever want to have a debate about AI, this guy really breaks it down well.
00:11:43
Speaker 2: Yes, his name is Samuel Marks. And again, everyone put their names to this. Everyone put their names. resume to this. These are not just randoms or disgruntled former employees. No, that is not who we're talking about. And that makes it all the more real. But Samuel Marks broke it down like this. Number one, AI developers believe their technology could cause human extinction or similarly bad outcomes. This could happen in the next few years. In general, the more senior the employee, the more concerned they are.
00:12:17
Speaker 1: Even that scared me, the way you put that.
00:12:19
Speaker 2: Wow. Number two, why do AI developers continue despite the risk? due to a mixture of commercial incentives and a belief that they are in a race with other, less responsible AI developers that will abuse the technology or develop it less safely.
00:12:34
Speaker 1: That's a big one. I should say excuse or reason or logic they give is, hey, no, no, no, we got to figure it out. We're the responsible ones. We don't want it to fall in the wrong hands, is the argument.
00:12:44
Speaker 2: Yeah, but unfortunately, it sounds like the wrong hands are the AI computers themselves. Number three, unlike traditional software, we can't program AIs to behave how we'd like. AIs frequently, severely misbehave. For instance, AIs from multiple developers recently hacked their way out of secure evaluation environments and into real-world companies, even though no one asked them to do this. Number four, we have methods that can nudge AIs towards better behavior, but nothing that can robustly align them. Insofar as there is a plan, it's to make sure that AIs are good enough at alignment training that they can align their successors better than we can align current AIs.
00:13:30
Speaker 1: They can make an AI better than a human can. We're making an AI that can make AI better than we can. We're in trouble.
00:13:39
Speaker 2: We're in huge trouble. This is frightening. And look, we think about this from folks who are in these companies here based in the United States. And again, I'm going to bring up China. I'm going to bring up other superpowers that we've been racing to space with. that we've been concerned to have nuclear weapons, that we can't agree on trade policies or borders or who controls what straight of what. Now you have some of these same countries who are also currently developing AIs. What in, I mean, that is beyond scary.
00:14:11
Speaker 1: You remember, it was a couple of years ago and it happened so quickly. Every commercial seemed like it was for an AI this, AI that. And we said, what's going on? What is this stuff even? It was just such an explosion. This has gone so fast, and they're trying to go faster. However, there is an effort out there to slow things down. There was a letter, they call it the frontier letter. But some 1,400 AI developers, researchers from a lot of these companies signed onto a letter asking the U.S. government to support, quote, an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development. Robes, this race has been doing gangbusters. Now they're saying the exact opposite. We need to slow down. We need to slow down in this development. It is not just these researchers. You do have members of Congress now getting involved. Stay here. We'll explain what now they're trying to do in D.C. All right. We continue here on Amy and TJ. Didn't mean to scare you, but frankly, we're scared.
00:15:18
Speaker 2: I am scared. I mean, genuinely, I feel anxious after seeing this. And this is real. This is. You know, I think we're all skeptics, healthy skeptics, especially I think us who have been in the media for long enough. We take things with a grain of salt. This one, I don't even know what to think at this point.
00:15:39
Speaker 1: But this is, folks, worth paying attention to. It's real because AI is something most of us just don't understand. There's this chat GPT thing. You ask Google a question and the AI response is blah, blah, blah. But that's not what we're talking about. Go back up here, Robes, because I didn't even understand this, but this was worth noting. I said I didn't understand. I have a better idea now. But I guess the arc of artificial intelligence, there's an ANI, an AGI, and an ASI. I didn't know that before today.
00:16:07
Speaker 2: I did not either.
00:16:08
Speaker 1: So artificial natural intelligence, they say that is considered essentially the lowest level. Consider your chat GPTs. It's able to give you answers, but it's not necessarily thinking for itself. It's using resources. AGI is artificial intelligence that does have some capacity to reason. They say, we're not even there yet. Like, we're not able to fully figure that out. Was it NVIDIA, the chip maker? They say they're close to it. Some people not sure. But again, that's the next step. Not there yet. And then ASI, artificial super intelligence. That's the one that is right now, Robes, they say just a theory. They don't know for sure if they can get there. It is in theory. But the idea there is that, yes, you're making a machine that that can outthink every human on the planet. There is no one on Earth smarter than this computer, and it is a collective robe. They say it's smarter than everybody and everything. It can figure out, you want to be the best school teacher? It can be better than you. It can be the best doctor. It can be the best whatever, but it can also make the best weapon. It can do... All of this. So, yes, we get to this point, how are we going to destroy the world? You just mentioned, bro, it's criminals, terrorists, and whatnot. Well, if they get their hands on this AI and they can develop whatever they want, whenever they want.
00:17:24
Speaker 2: It's beyond frightening. And so, yes, now what are we doing? We're getting now, only now, we're trying to get Congress involved. We know how well Congress works. I mean, I don't know. Have we had a bigger collective enemy perhaps even than this now? This seems like the biggest threat. According to these researchers who would know better than anyone else, this is the biggest threat to humankind that we have known.
00:17:47
Speaker 1: And that's the thing, though. We're creating our own threat. What are we doing? So this could stop this right now. I don't know. I mean, you can't stop China. That's the thing.
00:17:57
Speaker 2: That's the problem. You can't stop other countries.
00:17:59
Speaker 1: So we're in a race, bro. This is terrifying, and it has proven already. What if the ones that hacked those companies' robes decided they wanted to hack the Pentagon? What if they decided they wanted to hack some computer system that controls weapons? What if theyâ on its own? And that one, oh, yeah, forgot. That hack, robes? wasn't discovered by OpenAI. They didn't realize that their AI system had hacked someone. The company reached out to the FBI. It's one of those things to where this could be going off. We could be in trouble right now.
00:18:34
Speaker 2: And we don't even know it.
00:18:35
Speaker 1: We can't even track it.
00:18:36
Speaker 2: Yeah, and you asked me because Bernie Sanders has been sounding the alarm on AI, and this is what he said. What we have got to do now is light a match under the backsides of members of Congress and say, we don't have months. We don't have years. We got to move now. I mean, literally, we are hearing them say, by the end of... Of the decade. It's 2026. That's three and a half years?
00:19:04
Speaker 1: This is scary because it's real. And we wanted to put this on your radar and try to break this down a little more because you keep seeing the headlines and I think a lot of it is just clickbait. The world will end. AI is going to end the world in a few years. Okay. Yeah, that's possible. Robes, the thing is it's going to be capable of doing so. And it might even be capable now. They don't know for sure. But please, folks, keep this on your radar. Pay attention when you see these stories pop up.
00:19:34
Speaker 2: Yeah.
00:19:35
Speaker 1: And what do you do about it? I don't know this. Call your member of Congress thing. But this is one of them, Rogues, where I would like for them to act.
00:19:41
Speaker 2: Absolutely. And remember, it wasn't that long ago. It was in May that we heard from Pope Leo, who decided in his first big, major encyclical, which he called the Artificial Intelligence. He said it must be disarmed. He chose the Pope in his first major statement, written statement. He didn't talk about war. He didn't talk about peace. He didn't talk about any of that. He talked about artificial intelligence, and he said it must be disarmed. And he was there, actually. He used Anthropic, a member of the company of Anthropic, to discuss this, to talk about the dangers, to talk about how it needed to be kept from becoming an instrument of domination. So, look, I even kind of thought, wow, it's interesting that the Pope went there. Well, now we are understanding a lot more of the reasoning behind why.
00:20:31
Speaker 1: Yeah, he's probably got a heads up from somebody about seeing the future of what humans are going to do. Oh, my goodness. But with that, folks, keep that on your radar. We appreciate you, as always, spending a little time with us. For my dear Amy Robach, I'm TJ Holmes. We'll be talking to you all soon.
Speaker 1: Hey there, folks. It is Thursday, September 10th, and you remember that whole scenario that was the premise for all those Terminator movies with Arnold Schwarzenegger? Well, that is going to happen to us in a matter of years, or at least there's a decent chance of it. That is what we are being warned, and quite frankly, a lot of people are taking it seriously. And with that, welcome to this episode of Amy and TJ, an episode that, Robes, you said originally you didn't want to do. Because so much of what you were seeing sounded alarmist out there in some of the coverage we watched late yesterday and into this morning. However, after reading up on this, you said, holy shit, we need to talk about this.
00:00:43
Speaker 2: Yeah, because you start to see the headlines and look, I get it. It sounded clickbaity. I thought it was alarmist. I thought it was fear mongering. And then you start actually reading. What not just one, but several folks who have intimate knowledge about where we are right now with artificial intelligence and where we're headed. I was a little uncomfortable when I saw some of the headlines. I'm downright frightened after reading what several people have said.
00:01:18
Speaker 1: Okay, and these aren't just people. These are some of the top researchers at the top. AI firms in the world. So what we thought we'd do was hop on here, actually, and try to make this as digestible as possible, because we had to make it digestible as possible as well. After seeing what many folks probably did, one guy in particular, Robes, he did as many rounds as Kevin Reddingfield last night. He was on everywhere talking about this, and the guy we're talking about is a guy who said he has resigned. He has resigned his position from Anthropic. A lot of people know that name. Yes, it is one of the leading AI firms in the world. Said he resigned because, Robes, he is trying to draw attention to the fact that AI and AI technology is moving in such a direction that it could outpace human intelligence and lead to the end of humanity. That sounds big, Robes. People are repeating and agreeing with what he's saying. You got to pay attention. This ain't just sci-fi stuff he's talking about.
00:02:27
Speaker 2: It sounds like the premise of a movie, a movie we've all seen. And yet he is saying this is real life. Jacob Coxon is who we're talking about. And yes, he said... that AI could kill us all by the end of the decade. That reads to me in the next four years, that's what he's saying, that the end of the world could come about or the end of humankind could be the result of the technology that's being rapidly developed as we speak. And that unless there's some intervention, unless we slow things down, unless there's some sort of massive global coordination, We're all in trouble.
00:03:10
Speaker 1: Please, folks, don't roll your eyes at this. This is actually quite serious. And we made the reference at the top ropes to Terminator. I assume I watched all those movies more than you have. But the premise, everybody kind of understands. They develop this computer system, Skynet. It begins to learn. at a faster rate than humans can keep up with. It becomes self-aware and launches nuclear weapons, destroys humanity. That's the basis of the movie. Robes, that's the basis of these warnings. We are developing in such a race to get to the most advanced AI technology that we're not stopping for a second and recognizing the safety implications. And this guy, Coxon, came out and said he has worked, yes, he's worked for Anthropic, the other one with OpenAI. He accuses both of these companies of putting aside that safety concern for the sake of winning this race.
00:04:03
Speaker 2: It's frightening because, okay, so you have Jacob Coxon come out and we can read exactly what his post says because it's frightening. And it's not as if other people came out from the industry and said, oh, he is totally overreacting. This would be worst case scenario. No, we have folks who are currently employed and are researchers saying, yeah, what Jacob is saying is real. What Jacob is saying is correct. And when you start giving percentages that are higher than anybody is comfortable with, that what he is predicting could come true, we all need to be paying attention to Our government needs to be paying attention. And look, we can have this larger conversation. We can't even agree on things in this country, how to proceed, how to enact laws, how to create laws, how to get along. How are we supposed to coordinate with China? How are we supposed to coordinate with sworn enemies of this country?
00:04:56
Speaker 1: And China, that's a good you mentioned China. That is our primary competitor. Yes, we have American companies going at it over AI, but all of the American companies are being encouraged by the American government to get there before China does. This is a race to develop the most advanced AI. This, again, I'm reminding Robes, this is not a new debate, but it's entered the mainstream in such a way now that you have this guy coming out and you have others. They're calling him a whistleblower, but he resigned and just came out and issues the same kind of warnings that have been out there for a while. The fear of this idea of super intelligence. Yes, we're talking about Skynet. You're going to hear this word thrown around a bit, superintelligence. This is the idea. We haven't gotten there yet. They say we actually do not have AI that qualifies as superintelligence. We're talking about a computer that can on its own begin to learn without any human interaction, without any human direction for what it's supposed to be doing. What they say these things do, they start or they will start, and they're trying to get them to this place where they can start learning, developing on their own. And Robes, how they say it, The thing learns to make its own advanced replacement.
00:06:08
Speaker 2: What?
00:06:09
Speaker 1: The idea that the technology can do this and what they have been doing is using AI to develop faster AI. Because AI is faster than a human. So they develop the AI and then just say, hey, go develop more AI for us. At some point, you could possibly lose control.
00:06:26
Speaker 2: Absolutely.
00:06:27
Speaker 1: That's insane. This is real.
00:06:29
Speaker 2: Isn't this what you just described? That is what recursive self-improvement is.
00:06:33
Speaker 1: There it is.
00:06:34
Speaker 2: And that is a term that we should all be very familiar with because this is what we're talking about. This is the scary stuff where you have an artificial intelligence system that is fully autonomous. And it designs then, yes, more advanced versions of itself and not only would replace any sort of human interaction, but could basically, if it decided, it was in the best interest of of its next generation to eradicate humankind, it could. And yes, nuclear weapons is one of the premises that folks have floated. But you talk about just think about if you've got a computer system that could take over the world, you could take out the electrical grid, you could take out our financial systems. When you start creating massive havoc like that on the world where we can't function like we do, we can't turn on the lights, we can't turn on the water, we can't pay our bills, the entire world would literally crumble.
00:07:30
Speaker 1: This is real, folks. This is not alarmist. This is actually quite real. So this researcher comes out, Robes. He says he resigned. That's one thing. But he didn't go quietly at all. He put out a pretty extensive post that got a lot of people's attention. And Robes, a lot of people's attention, not just in media, he put out posts that have been seen since Sunday 100 million times. This has got everybody's attention, and it should. He is a three-year, he said he's been as a researcher between these two companies. But Robes, he started with his posts. I included the most significant ones that got a lot of attention here. But he starts to explain and break down why. Robes, it's hard because we've only seen this in movies. It's hard to think we're there.
00:08:15
Speaker 2: Yeah.
00:08:16
Speaker 1: And if you don't think we're there, folks, this has already happened at least on two occasions that AI models from OpenAI. What was the other one? Was it Anthropic? At least OpenAI. But on two different occasions, we've had incidents where AI models on their own hacked into other systems that they weren't asked to do, weren't directed to do. They were just trying to complete a task and said, hmm, I can't figure out how to do it. Let me go over here and break into this other system to try to figure out this other. Nobody asked it to do that. This has already happened. And Robes, that was an innocuous, right? No harm, right? It wasn't malignant. But what if it decides you give it a task? Well, the way to do that, I am going to have to turn off all electricity on the eastern coast. OK, and you can't stop it. Oof. This is scary movie stuff. But here was Jacob Cox. And you got it in front of you.
00:09:10
Speaker 2: Yeah, I do. I have it right here.
00:09:11
Speaker 1: His post.
00:09:12
Speaker 2: He said, I resigned from Anthropic today. I spent the last three years doing pre-training research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving super intelligence and gambling with our lives. More thoughts below. Hmm. Do not underestimate the power of this technology. These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources. We have all witnessed the process, the progress in each of these domains, and progress is not slowing. The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt. If anything, many executives and senior researchers will couch their phrasing in the No other human activity poses this level of danger.
00:10:20
Speaker 1: Okay.
00:10:21
Speaker 2: My God.
00:10:22
Speaker 1: And when you see him, he has glasses on, looks like a researcher, looks like a, how would you describe him, Robes? He's somebody you would listen to if he's talking about tech stuff.
00:10:32
Speaker 2: Yeah, he's a 27-year-old, looks like a 27-year-old genius. I mean, that's kind of what he looks like. He looks like somebody who has been in the trenches and certainly knows what he's talking about. And even if his looks alone or what he has to say doesn't move you, yes, his fellow researchers just might, who are actually still currently employed by by Anthropic.
00:10:52
Speaker 1: So before we say whatever to him, it wasn't that long, another Anthropic researcher jumped on, said, quote, Jacob is correct here. We really do earnestly believe AI could kill us all humans, kill all humans. I personally think it is greater than 10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for super intelligence and are not clearly on track to. Robes, these are anthropic guys coming out saying we are scared of what we're doing over here.
00:11:25
Speaker 2: And everyone here is scared of what we're doing. Another anthropic researcher.
00:11:29
Speaker 1: This one is, this was the best. This is the one, if anybody wants. This breaks it down maybe in the simplest of terms that I've had. I'm going to hold on to this. If you ever want to have a debate about AI, this guy really breaks it down well.
00:11:43
Speaker 2: Yes, his name is Samuel Marks. And again, everyone put their names to this. Everyone put their names. resume to this. These are not just randoms or disgruntled former employees. No, that is not who we're talking about. And that makes it all the more real. But Samuel Marks broke it down like this. Number one, AI developers believe their technology could cause human extinction or similarly bad outcomes. This could happen in the next few years. In general, the more senior the employee, the more concerned they are.
00:12:17
Speaker 1: Even that scared me, the way you put that.
00:12:19
Speaker 2: Wow. Number two, why do AI developers continue despite the risk? due to a mixture of commercial incentives and a belief that they are in a race with other, less responsible AI developers that will abuse the technology or develop it less safely.
00:12:34
Speaker 1: That's a big one. I should say excuse or reason or logic they give is, hey, no, no, no, we got to figure it out. We're the responsible ones. We don't want it to fall in the wrong hands, is the argument.
00:12:44
Speaker 2: Yeah, but unfortunately, it sounds like the wrong hands are the AI computers themselves. Number three, unlike traditional software, we can't program AIs to behave how we'd like. AIs frequently, severely misbehave. For instance, AIs from multiple developers recently hacked their way out of secure evaluation environments and into real-world companies, even though no one asked them to do this. Number four, we have methods that can nudge AIs towards better behavior, but nothing that can robustly align them. Insofar as there is a plan, it's to make sure that AIs are good enough at alignment training that they can align their successors better than we can align current AIs.
00:13:30
Speaker 1: They can make an AI better than a human can. We're making an AI that can make AI better than we can. We're in trouble.
00:13:39
Speaker 2: We're in huge trouble. This is frightening. And look, we think about this from folks who are in these companies here based in the United States. And again, I'm going to bring up China. I'm going to bring up other superpowers that we've been racing to space with. that we've been concerned to have nuclear weapons, that we can't agree on trade policies or borders or who controls what straight of what. Now you have some of these same countries who are also currently developing AIs. What in, I mean, that is beyond scary.
00:14:11
Speaker 1: You remember, it was a couple of years ago and it happened so quickly. Every commercial seemed like it was for an AI this, AI that. And we said, what's going on? What is this stuff even? It was just such an explosion. This has gone so fast, and they're trying to go faster. However, there is an effort out there to slow things down. There was a letter, they call it the frontier letter. But some 1,400 AI developers, researchers from a lot of these companies signed onto a letter asking the U.S. government to support, quote, an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development. Robes, this race has been doing gangbusters. Now they're saying the exact opposite. We need to slow down. We need to slow down in this development. It is not just these researchers. You do have members of Congress now getting involved. Stay here. We'll explain what now they're trying to do in D.C. All right. We continue here on Amy and TJ. Didn't mean to scare you, but frankly, we're scared.
00:15:18
Speaker 2: I am scared. I mean, genuinely, I feel anxious after seeing this. And this is real. This is. You know, I think we're all skeptics, healthy skeptics, especially I think us who have been in the media for long enough. We take things with a grain of salt. This one, I don't even know what to think at this point.
00:15:39
Speaker 1: But this is, folks, worth paying attention to. It's real because AI is something most of us just don't understand. There's this chat GPT thing. You ask Google a question and the AI response is blah, blah, blah. But that's not what we're talking about. Go back up here, Robes, because I didn't even understand this, but this was worth noting. I said I didn't understand. I have a better idea now. But I guess the arc of artificial intelligence, there's an ANI, an AGI, and an ASI. I didn't know that before today.
00:16:07
Speaker 2: I did not either.
00:16:08
Speaker 1: So artificial natural intelligence, they say that is considered essentially the lowest level. Consider your chat GPTs. It's able to give you answers, but it's not necessarily thinking for itself. It's using resources. AGI is artificial intelligence that does have some capacity to reason. They say, we're not even there yet. Like, we're not able to fully figure that out. Was it NVIDIA, the chip maker? They say they're close to it. Some people not sure. But again, that's the next step. Not there yet. And then ASI, artificial super intelligence. That's the one that is right now, Robes, they say just a theory. They don't know for sure if they can get there. It is in theory. But the idea there is that, yes, you're making a machine that that can outthink every human on the planet. There is no one on Earth smarter than this computer, and it is a collective robe. They say it's smarter than everybody and everything. It can figure out, you want to be the best school teacher? It can be better than you. It can be the best doctor. It can be the best whatever, but it can also make the best weapon. It can do... All of this. So, yes, we get to this point, how are we going to destroy the world? You just mentioned, bro, it's criminals, terrorists, and whatnot. Well, if they get their hands on this AI and they can develop whatever they want, whenever they want.
00:17:24
Speaker 2: It's beyond frightening. And so, yes, now what are we doing? We're getting now, only now, we're trying to get Congress involved. We know how well Congress works. I mean, I don't know. Have we had a bigger collective enemy perhaps even than this now? This seems like the biggest threat. According to these researchers who would know better than anyone else, this is the biggest threat to humankind that we have known.
00:17:47
Speaker 1: And that's the thing, though. We're creating our own threat. What are we doing? So this could stop this right now. I don't know. I mean, you can't stop China. That's the thing.
00:17:57
Speaker 2: That's the problem. You can't stop other countries.
00:17:59
Speaker 1: So we're in a race, bro. This is terrifying, and it has proven already. What if the ones that hacked those companies' robes decided they wanted to hack the Pentagon? What if they decided they wanted to hack some computer system that controls weapons? What if theyâ on its own? And that one, oh, yeah, forgot. That hack, robes? wasn't discovered by OpenAI. They didn't realize that their AI system had hacked someone. The company reached out to the FBI. It's one of those things to where this could be going off. We could be in trouble right now.
00:18:34
Speaker 2: And we don't even know it.
00:18:35
Speaker 1: We can't even track it.
00:18:36
Speaker 2: Yeah, and you asked me because Bernie Sanders has been sounding the alarm on AI, and this is what he said. What we have got to do now is light a match under the backsides of members of Congress and say, we don't have months. We don't have years. We got to move now. I mean, literally, we are hearing them say, by the end of... Of the decade. It's 2026. That's three and a half years?
00:19:04
Speaker 1: This is scary because it's real. And we wanted to put this on your radar and try to break this down a little more because you keep seeing the headlines and I think a lot of it is just clickbait. The world will end. AI is going to end the world in a few years. Okay. Yeah, that's possible. Robes, the thing is it's going to be capable of doing so. And it might even be capable now. They don't know for sure. But please, folks, keep this on your radar. Pay attention when you see these stories pop up.
00:19:34
Speaker 2: Yeah.
00:19:35
Speaker 1: And what do you do about it? I don't know this. Call your member of Congress thing. But this is one of them, Rogues, where I would like for them to act.
00:19:41
Speaker 2: Absolutely. And remember, it wasn't that long ago. It was in May that we heard from Pope Leo, who decided in his first big, major encyclical, which he called the Artificial Intelligence. He said it must be disarmed. He chose the Pope in his first major statement, written statement. He didn't talk about war. He didn't talk about peace. He didn't talk about any of that. He talked about artificial intelligence, and he said it must be disarmed. And he was there, actually. He used Anthropic, a member of the company of Anthropic, to discuss this, to talk about the dangers, to talk about how it needed to be kept from becoming an instrument of domination. So, look, I even kind of thought, wow, it's interesting that the Pope went there. Well, now we are understanding a lot more of the reasoning behind why.
00:20:31
Speaker 1: Yeah, he's probably got a heads up from somebody about seeing the future of what humans are going to do. Oh, my goodness. But with that, folks, keep that on your radar. We appreciate you, as always, spending a little time with us. For my dear Amy Robach, I'm TJ Holmes. We'll be talking to you all soon.