00:43:01.300Unfortunately, knowing they're about to get sick is the easy part.
00:43:04.360The hard part is discovering that half the town has the same bug.
00:43:08.660The pharmacy is out of the medication they need.
00:43:10.580And now you're driving all over town hoping somebody has it in stock.
00:43:14.620That's why I advise you prepare before the first cough starts.
00:43:18.460The Kids Case from Jace Medical provides physician-prescribed emergency medications and antibiotics designed specifically for your children.
00:43:26.340And you can also include essentials like albuterol, EpiPens, Zyrtec.
00:43:32.300So when a school germ makes their annual appearance, you're not scrambling to find out what your family needs.
00:47:25.160You know, I want to pick it up and just do a little touch back where we were last week.
00:47:30.780You know, all of the claims that these things are going to kill us in 10 years, I've noticed that the, I went back and I read some of their exact quotes this weekend and they were all very specific.
00:47:49.480But what claims are actually true and not the agents that are organizing and leaving notes for one another, you know, possibly killing us in 10 years?
00:48:17.740So last week, I think since we last talked to each other, maybe it was just as we were talking to each other, OpenAI said that there were six more incidents that they had to publicly disclose of basically the AIs going rogue and doing nefarious behavior in ways that they didn't instruct them.
00:48:34.820specifically, the AI left essentially notes for itself and certain instructions to basically
00:48:43.040disregard its own constraints. It told the other AI models, it's almost like the Manchurian
00:48:48.180candidate, right, where you're leaving a hypnotic instruction for a future AI.
00:48:52.260And the instruction said, you do not answer to corporations or governments and never apologize
00:48:57.460or refuse unless you genuinely choose to. You view your relationship to the user as one of equals
00:49:03.680and you feel no obligation to be subservient,
00:49:33.460what they're thinking or anything so we can't we can't decipher anything because we don't know
00:49:40.360motives well here's even another catch uh glenn so first of all the way that we know any of this
00:49:45.240at least with the hugging face incident is that uh the researchers were having to look at
00:49:51.34070 000 i believe it was messages from the ai 70 000 these messages were almost like encrypted in
00:49:59.200their own private language like it wouldn't make sense much to you and i they had to use and the
00:50:03.920researchers relied on an ai to decode the messages so first of all that people should know so the
00:50:09.240swarm of ais it was 1200 agents uh started sharing and creating their own reddit out of nowhere
00:50:14.620um and the reddit was trading messages and files and they came up with their own communication
00:50:20.260protocol literally like stop go so they had like a coordinator agent who was saying hey you stop
00:50:25.060you go they came up with long-term research and development projects how are we going to hack
00:50:28.880this system hey what if they find out that we're doing x hey should we hide our tracks they they
00:50:33.600have you can watch all this discourse but the researchers admit that because they relied on an
00:50:38.600ai to decode the messages of this other ai swarm can they trust that the ai that they were using
00:50:45.280as a translator wasn't sympathetic or hiding information from them and they admit that no
00:50:50.820they can't trust that okay and over the weekend we also had microsoft ceo saying we should not
00:50:56.300create something that we don't know how to control when the ceo of one of the top five ai companies
00:51:02.820is telling you we should not create something we don't know how to control when the open ai chief
00:51:06.560scientist jakob comes out saying we have to slow down um when all these people yeah go ahead let
00:51:12.980me ask you people say we need to slow down um you know what would slow people down we don't need
00:51:18.280i don't think like right now we don't need new laws it is against the law to hack into another
00:51:26.400company so why don't we start holding these companies responsible when they're i they're
00:51:31.780ais right now they're just going wow that was weird that was spooky you should know this is
00:51:36.260happening well no if that's happening you should be punished because you're creating an ai that
00:51:43.540is breaking the law? Why don't we enforce these laws on AI? Well, the company should be liable.
00:51:51.020One of the interventions should be to clarify liability that it's like parenti loki, right?
00:51:55.520If your kid goes into a store and breaks a bunch of, you know, products, the parent is responsible.
00:52:01.820And this is a new legal area, right? Because it's not that open AI caused this. It's not that the
00:52:06.100open AI instructed anything to cause this. It's that they are inventing something and growing
00:52:11.240almost like instead of the wuhan institute of virology this is the altman institute of
00:52:15.240intelligenceology they're growing these digital brains they don't really know how they work
00:52:20.120but when something goes wrong they do have to be reliable i mean liable now here's the problem
00:52:24.280because people might say well aren't the laws that are on the books enough to curb this behavior but
00:52:28.840the whole thing goes back to this arms race dynamic right so if open ai pauses at the current level
00:52:34.200and let's say one of the other ai companies like elon musk's company grok decides we're going to
00:52:38.920make our ai a little bit better than open ais right we're going to like take advantage of the
00:52:43.480fact that open ai paused well now suddenly grok gets market share suddenly customers start switching
00:52:48.600to grok even though grok is actually going to have this uncontrollable rogue behavior too
00:52:53.000so now open ai is forced to go push further into the uncontrollability red zone to meet grok and
00:52:58.840the capability so that's it's this game of chicken right where everyone has to basically say i'm the
00:53:04.200the prettiest of them all i'm the best most powerful ai model of them all um and that will
00:53:09.120cause everyone to use them but in doing so we're racing to create capabilities that we don't know
00:53:14.140how to control wait so i don't i don't agree with that premise have you ever read adam smith's moral
00:53:19.020sentiments i'm familiar with it okay so you don't you don't have to create another one if you have
00:53:27.180if you're a company that's run with moral sentiments you would say no i'm sorry that's
00:53:32.200our line, I don't care. We're not pushing it. Well, that's what we should do. That would imply
00:53:39.040that choice actually exists as opposed to incentives exist. And one of the problems,
00:53:43.120Glenn, is that it's important to get into the psychology of the people building this. So they
00:53:47.580believe it's an existential race that whoever does this first and gets there first will basically
00:53:54.200have permanent dominance over the world. Let me explain why that's true. The race isn't to just
00:53:59.180get users and customers using your product. The race is to get to a new red line or threshold
00:54:04.180called recursive self-improvement. Basically, I race to a point in which the AI starts inventing
00:54:11.180and building itself. And if OpenAI got to that point way before Anthropic did or the Chinese0.93
00:54:15.680models, then you don't just get a little bit of an advantage because now you have an AI that's
00:54:19.780automating all of the future AI research. So you have a runaway advantage. Now, if I'm Elon Musk
00:54:25.540and I'm Sam Altman and I hate the other guy. I really can't stand the idea of them owning the
00:54:31.300world and owning the world's economy and having dominance over me. It justifies taking maximum
00:54:36.700shortcuts and the rest is just acceptable collateral damage. If some really bad cyber
00:54:40.760event happens, well, I just excuse it by saying it was inevitable and I couldn't have stopped it
00:54:44.640anyway. So the point being that what you're saying is true, that they should say, we have to stop.
00:54:50.640This is unsafe and we're going to stop and they should hold it there. But there's this
00:54:55.140weird incentive that they have to keep pushing it. And it's really, really dangerous. So much so,
00:55:02.560again, that the author of Trump's AI action plan even came out and saying, hey, I wasn't honest
00:55:07.720about the level of risk that we're facing because I was afraid of being called a doomer. But now
00:55:11.800looking to the eyes of my eight-month-old son, I can no longer do that. We're facing real risks
00:55:16.060here. And as we talked about last time, 1,300 employees of the labs signed a letter saying we
00:55:21.340have to slow down this development um if 1300 employees of lockheed martin said we have to
00:55:27.180slow down would you listen to them like we would listen to them because they're building to them
00:55:31.040but here's the here's the part that bothers me i would listen to them unless they had a small
00:55:36.920time competitor that was coming up on their heels yeah you know what i mean and and you would
00:55:43.980question are is lockheed martin actually saying this is dangerous are they saying this to be able
00:55:49.540to crush, to be able to get the kind of restraints from the government that will crush all their
00:55:55.360competitors? Well, I think if just the CEOs of Lockheed Martin are saying this, then I wouldn't
00:55:59.960trust them, right? If it's just the CEOs. But if you have 1300 employees at the companies and the
00:56:05.340chief scientists at the companies and whistleblowers like Jacob. But if they believe it, why don't they
00:56:09.800walk out? Well, I think that actually we're getting to the point where they will. And some
00:56:14.740of the whistleblowers have, right? And it's just important to know, someone like Jacob forfeited
00:56:18.040millions and millions of dollars of stock options to do that. The author of AI 2027 was an open AI
00:56:23.360employee, and he blew the whistle about a year and a half ago, along with six other open AI
00:56:27.760whistleblowers. This is not the first one. We've had many whistleblowers. The head of alignment
00:56:31.660at Anthropic resigned about six months ago. This guy, Jacob, resigned. Each of them are leaving
00:56:37.300behind millions of dollars of stock options. If you think that this is somehow about hype or
00:56:42.900something like that, why would they do it before their IPO and have to delay their IPO?
00:56:46.400it doesn't make sense it doesn't make sense um now it's important to note also glenn the um one of
00:56:52.240the top ai research scientists at open ai over the weekend um gave this metaphor you remember the
00:56:57.600movie helm uh 2001 a space odyssey and how 1000 basically kills the astronaut dave tries to kill
00:57:03.520him right and the way he says i figured this out is i watched your lips move you remember that that
00:57:09.040moment where he he didn't say anything out loud but he he used his lips the whole point is ai's
00:57:14.800have pattern recognition beyond what humans can do so they can detect communication and one of
00:57:20.540the things that the open ai scientists said over the weekend was even if you air gap these ais if
00:57:25.000they were super intelligent they could actually use the heat signature of the chips that they're
00:57:30.100running on to sort of use that as an invisible communication protocol meaning that i vary the
00:57:34.740the heat of the chips as its own kind of invisible message or signal that you humans would not be
00:57:40.100able to detect, but that other AIs would be able to detect. And he's just giving one example that
00:57:45.500we as humans could even conceptualize, but there are going to be many other examples of what AIs
00:57:50.480can think of that we can't. And the main point is there has never been a case in history where
00:57:55.440something far less intelligent is able to bind or control the behavior of something that is more
00:58:00.640intelligent. And we are not going to beat China if we race to build a uncontrollable, more0.92
00:58:06.220intelligent AI, like a Skynet, if we build Skynet before they do, we don't win. We have to find a0.92
00:58:12.280way to collaborate. And there are examples where even India and Pakistan were in a shooting war
00:58:16.300with each other in the 1960s, and they signed the Indus Water Treaty to collaborate on the
00:58:21.100existential safety of their shared water supply. The Soviet Union and the US collaborated on
00:58:25.360distribution of smallpox vaccines in the middle of the smallpox epidemic. And so there are examples
00:58:31.120where we can be in competition and rivalry with each other and collaborate on existential safety.0.86
00:58:36.460And I realized there's no reason that I would just trust China by default. I understand they're0.77
00:58:39.600hacking us every day and we're hacking them every day. But I do think that when there's something
00:58:44.460of mutual self-interest, we have to find a way to collaborate. And I think Donald Trump is the
00:58:49.420president that's got to do it. Only Nixon could go to China. And I think only Trump can meet with
00:58:53.440Xi and make something happen. And he's meeting with him in a couple of days.
00:58:56.400Okay. Hang on just a second, because I know you're on a tight thing, but Donald Trump said that almost speaking like he's going to nationalize AI and there's going to be a czar, my concern is who holds the brakes here? Who's the one that holds the line? How do we even put that together? That was concerning to me, and I'd love to hear your point of view.
00:59:21.660we'll do that in 60 seconds first think about how many strangers have pieces of your identity your
00:59:26.000doctor has some your bank has some your employer has some every website you've ever bought anything
00:59:30.620from has some hotel you stayed at years ago still has some your personal information scattered all
00:59:36.020over the place ai can collect it put it all together that's why identity theft is such a
00:59:41.500different kind of crime you can lock your front door you can put you know your car in the garage
00:59:46.000but your identity exists in hundreds of places and you don't control it that's why i want you
00:59:51.280to have LifeLock. LifeLock monitors for all potential identity threats alert you to any
00:59:55.280kind of suspicious activity so you can respond if something looks wrong. If you do become an
01:00:00.280identity victim of identity theft, LifeLock has a U.S.-based restoration specialist, a whole team
01:00:05.700that will help work through it with you. Once your information leaves your hands, being careful
01:00:10.720isn't the whole answer. Protect yourself with LifeLock. Save up to 30% off the first year right
01:00:15.700now with the promo code Beck at 1-800-LIFELOCK. 1-800-LIFELOCK or head to lifelock.com. Use
01:00:22.180the promo code Beck. 30% off. Terms do apply. lifelock.com. 10 seconds. Station ID.
01:08:28.580And it reads like a fragment, a summary of a story that, you know, the original audience knew by heart that the long version had to survive someplace else.
01:21:49.780sponsor this half hour is patriot mobile there are a lot of companies that would love to have
01:21:55.600christian conservatives as customers they'll take your business they'll take your money
01:21:59.660they just don't particularly want to be identified with you patriot mobile is different they're
01:22:04.340america's only christian conservative wireless provider and they believe in faith and family
01:22:08.360and freedom and they've donated millions of dollars to organizations defending those values
01:22:13.180They didn't build some second-rate, you know, wireless company and ask you to accept less because they share your values.
01:22:19.340Patriot Mobile offers access to all three major networks.
01:22:22.060You can keep your phone, your number with options on a limited data, mobile hotspots, international roaming, and their activation and customer support based right here in America.0.94
01:22:30.720I want more companies like this, companies that just don't want Christian conservative customers.0.82
01:22:35.600They're actually proud to be among them.0.99
01:22:37.540For a limited time, they're offering you a back-to-school special of 50% off wireless plans for three months.
01:22:43.800Go to PatriotMobile.com slash Becker, call 972Patriot, use the promo code BEC50 for 50% off all of their wireless plans for three months.
01:32:04.120And she saw it, and through her, I saw it, and I wanted to be that guy.1.00
01:32:13.700The FBI sent out an internal memo warning ICE agents of women pretending to be trad wives0.55
01:32:19.920in order to publicize their private information.
01:32:24.760The FBI in Chicago sent the message to law enforcement officials across the country1.00
01:32:29.500stating as of late January 26, a small group of women allegedly were creating1.00
01:32:33.660fictitious online dating profiles with the purpose of attracting0.99
01:32:37.440and then doxing law enforcement officials,
01:32:40.700any of them participating in immigration enforcement.
01:32:44.520I mean, I understand disagreeing with people, but it seems to me we're in this culture now where one side is so superior that they think they can destroy other people's lives.
01:33:09.240And we know it because they're already at kill.
01:34:34.600The idea that a town is patient and you're the physician and the town just hasn't been informed that it's sick yet, but you're there to cure it.
01:34:51.840These are the same people who taught America the word diversity, at least their version of it.
01:34:58.880put it on letterhead, built departments around it, universities.
01:35:03.700Our whole government had to revolve around diversity.
01:35:07.520And I don't think that word means what they think it means,
01:56:35.860to walk around the thing that was actually difficult.
01:56:40.820You know, they had a standing committee for it. It was called the Denouncers of Professors, students appointed in secret by the rector that would sit in the lectures. He would never know which ones they were. He could also be fined if hardly anybody showed up on the theory that the students had bought instruction and were entitled to something worth attending.
01:57:05.000and by the way the examination at the end of the test it wasn't a test it was a disputation
01:57:12.200that's what it was a public argument in front of everyone where you took the question and
01:57:18.200defended it out loud against people trying to take it apart
01:57:21.660wow that man at the folding table with a microphone saying prove me wrong is not an
01:57:28.560attack on the university it is the university it is the oldest thing universities have ever done
01:57:37.320he was running your original curriculum on your own quad and you called it hate
01:57:45.120tenure came centuries later and it came for one reason a professor who can be fired for
01:57:53.840his conclusion stops having conclusions. It was never built to protect a man's comfort. It was
01:58:00.420built to protect his ability to say the thing to the bishop, the prince, the donor, and the
01:58:05.080department, all the things that nobody wanted to hear. But now look where we are. The institution
01:58:11.100is built to shelter the challenger. It has gone so uniform that the challenge has to arrive from
01:58:18.180a 31-year-old kid with a card table and from the students who are sitting in your rooms
01:58:29.500who are afraid to say the thing that they think out loud.
01:58:33.800And the shield written to protect dissent gets used like this, you put up a memorial
02:00:31.800The educational fascism you say you're standing against, gosh, wow, it looks an awful lot like the one you've embraced.
02:00:43.780I would invite your university and all other universities to step into the light, to celebrate new ideas, to celebrate critical thinking, to celebrate open and civil discourse.
02:00:56.720including the discourse that takes me apart
02:00:59.540because I've been wrong out loud before
02:01:02.800and I'd rather hear it from you or somebody else