WEBVTT - Parakeet! Parakeet! Parakeet! 0:00:01.820 --> 0:00:05.680 Broadcasting live from the Abraham Lincoln Radio Studio at the 0:00:05.720 --> 0:00:10.720 George Washington Broadcast Center. Jack Armstrong and Joe Getty. Armstrong 0:00:10.800 --> 0:00:17.210 and Getty. And now, here's Armstrong and Getty. 0:00:23.760 --> 0:00:26.600 We turn now to the dangers of a runaway technology. 0:00:26.720 --> 0:00:30.480 New warnings that shook the world Tuesday about artificial intelligence 0:00:30.520 --> 0:00:33.500 from the very people developing it. It doesn't get bigger 0:00:33.600 --> 0:00:37.200 or more direct than this. AI could kill all humans, 0:00:37.260 --> 0:00:40.840 that according to a lead researcher at Anthropic, giving that 0:00:40.880 --> 0:00:44.620 possibility a more than 10% chance within the next decade. 0:00:45.270 --> 0:00:48.230 The call now is for a global conversation about risk 0:00:48.340 --> 0:00:49.449 and safety as well. 0:00:49.670 --> 0:00:53.660 So- I don't want to get off track here because 0:00:53.680 --> 0:00:56.220 this is way too interesting. But the fact that he 0:00:56.320 --> 0:01:00.990 had that sentence after they talked about the weather. 0:01:01.750 --> 0:01:04.250 It's just stunning. Well, you don't understand. There are high 0:01:04.310 --> 0:01:05.350 winds on Hawaii. 0:01:05.750 --> 0:01:11.970 A shocking, earth-shattering, got-everybody's-attention story today, we all may die soon. 0:01:12.690 --> 0:01:14.770 Comes after the weather. Yes. 0:01:15.190 --> 0:01:15.490 Okay. 0:01:15.950 --> 0:01:17.170 That's your evening news. 0:01:17.310 --> 0:01:17.930 Yes. 0:01:18.470 --> 0:01:22.810 So some of the reactions from fellow researchers were equally concerning. 0:01:23.209 --> 0:01:26.770 Hit us with 61 here. Michael Madison Allworth on Fox News. 0:01:27.130 --> 0:01:30.310 His claims backed up by a senior AI safety researcher 0:01:30.480 --> 0:01:34.040 at the company. That executive writing, quote, Jacob is correct here. 0:01:34.340 --> 0:01:38.060 We really do earnestly believe AI could kill all humans. 0:01:38.560 --> 0:01:41.230 I personally think it is greater than 10% within the 0:01:41.310 --> 0:01:44.850 next decade. This comes after multiple cyber attacks have been 0:01:44.890 --> 0:01:47.289 carried out by AI agents that went rogue. 0:01:47.630 --> 0:01:49.520 They are signs that we don't know how to make 0:01:49.530 --> 0:01:50.460 the AIs care about us. 0:01:50.740 --> 0:01:53.140 If they succeed at making a superintelligence before anyone figures 0:01:53.180 --> 0:01:55.160 out how to make the AIs care about us, that's 0:01:55.200 --> 0:01:56.120 a recipe for disaster. 0:01:56.420 --> 0:01:59.040 And people in these companies know it, and they're worried. 0:01:59.200 --> 0:02:00.930 There's no way you're going to make AIs care about you. 0:02:02.860 --> 0:02:04.790 The number of texts we got, are you doomers? This 0:02:04.830 --> 0:02:08.169 is what they said about the population bomb. I don't 0:02:08.230 --> 0:02:12.030 think any of the examples people give of technology or 0:02:12.210 --> 0:02:12.869 fears that. 0:02:12.810 --> 0:02:13.730 Have happened before. 0:02:14.169 --> 0:02:17.090 hold up as an example of, this is just like 0:02:17.130 --> 0:02:19.680 the population bomb, seems ridiculous to me. 0:02:20.340 --> 0:02:23.660 Says you, Doomer. And you're a boomer, a boomer-doomer. Well, 0:02:23.720 --> 0:02:28.540 Gen X, anyway. That doesn't rhyme. So, coming up, a 0:02:28.700 --> 0:02:32.980 really persuasive Twitter thread calling BS on all of this. So, 0:02:33.620 --> 0:02:34.380 stay tuned for that. 0:02:34.520 --> 0:02:38.270 I want this to not happen. I'm not rooting for it. Well, 0:02:38.330 --> 0:02:39.650 I kind of assumed that. 0:02:39.970 --> 0:02:43.270 Anybody who is rooting for mass extinction can't be my friend. 0:02:43.290 --> 0:02:43.350 Ha! 0:02:45.580 --> 0:02:50.480 So interestingly, the first outlet that this young chap with 0:02:50.540 --> 0:02:55.040 a charming British accent spoke to was Bret Baier's special 0:02:55.060 --> 0:02:59.630 report last night. Bret was down at the Trumpapalooza in Dallas, 0:02:59.669 --> 0:03:02.170 but he cleared off quite a bit of time, appropriately, 0:03:02.889 --> 0:03:04.269 to talk to the man who posted this. To talk 0:03:04.310 --> 0:03:07.450 about the end of humanity? Yes, who posted the initial 0:03:07.750 --> 0:03:12.299 tweet about, well, the extinction of human life. And why 0:03:12.419 --> 0:03:14.160 don't we just start with 70, Michael, and we'll work 0:03:14.200 --> 0:03:14.380 our way. 0:03:15.490 --> 0:03:18.850 I want to start at the beginning about what your 0:03:18.910 --> 0:03:22.630 feeling was, what job you were in, and why you 0:03:22.650 --> 0:03:24.169 got to this point of coming out. 0:03:24.350 --> 0:03:27.410 So I was working on building AI systems to make 0:03:27.460 --> 0:03:31.059 them smarter. And recently I've come to the realization that 0:03:31.480 --> 0:03:34.000 many people expect the AI systems in the next few 0:03:34.040 --> 0:03:37.900 years to be potentially capable of essentially taking over the world. 0:03:40.210 --> 0:03:42.740 It's basically a gamble that's happening on the part of 0:03:42.780 --> 0:03:45.720 these companies where there's a strong chance that these AIs 0:03:45.780 --> 0:03:46.350 could take over. 0:03:46.370 --> 0:03:49.710 Okay, go on. 0:03:50.480 --> 0:03:55.970 Evan Hubiger, who is alignment science lead at Anthropic, posted 0:03:56.210 --> 0:03:59.310 after you posted saying, Jacob is correct here. We really 0:03:59.390 --> 0:04:03.010 do earnestly believe AI could kill all humans. I personally 0:04:03.050 --> 0:04:06.410 think it's greater than 10% within the next decade. I 0:04:06.440 --> 0:04:08.800 believe Anthropic is trying its best, but we do not 0:04:08.860 --> 0:04:12.060 yet have a plan to solve alignment for superintelligence and 0:04:12.120 --> 0:04:13.380 are not clearly on track to. 0:04:14.040 --> 0:04:15.080 Most people stopped there. 0:04:15.100 --> 0:04:18.160 The next thing he posted, to be clear, as we 0:04:18.200 --> 0:04:20.730 say in our latest risk report, I think the risk 0:04:20.760 --> 0:04:23.589 from present models is low. What I'm worried about is 0:04:23.670 --> 0:04:27.930 superintelligence arising from recursive self-improvement, as we have said, is 0:04:27.970 --> 0:04:31.230 happening faster than we thought. Can you parse that and 0:04:31.250 --> 0:04:32.310 the threat right now? 0:04:32.330 --> 0:04:34.710 I agree with Evan that right now the threat is 0:04:34.730 --> 0:04:38.910 minimal of actual takeover. But because of self-improvement, this AI 0:04:38.950 --> 0:04:41.890 is making themselves smarter could happen as soon as next year. 0:04:42.250 --> 0:04:45.130 Some people even say six months. So this attitude is 0:04:45.150 --> 0:04:49.170 shared across the industry where people essentially are aware that 0:04:49.210 --> 0:04:51.860 the work they are currently doing, if we don't slow down, 0:04:52.500 --> 0:04:56.000 could lead to AI takeover by the end of the decade. 0:04:56.300 --> 0:04:59.940 That moment, if it does happen, and some people think 0:05:00.000 --> 0:05:02.299 it can't, but if it does happen where the computers 0:05:02.320 --> 0:05:09.690 can program themselves, there's no way. you can argue that 0:05:09.730 --> 0:05:14.000 that's not going to be a major change in everything 0:05:14.040 --> 0:05:17.760 in the world once they start programming themselves. And it 0:05:17.800 --> 0:05:21.640 seems pretty clear that the whole idea of alignment is 0:05:21.680 --> 0:05:22.700 just out the window. 0:05:24.600 --> 0:05:28.180 Yeah, and we will get to the Chinese elephant in 0:05:28.220 --> 0:05:29.400 the room in a moment or two. 0:05:30.839 --> 0:05:34.440 The alignment being the idea that you give it some 0:05:34.500 --> 0:05:41.130 guardrails like morality and laws, the Ten Commandments, whatever. 0:05:41.770 --> 0:05:46.250 Yeah, yeah. So one of the pieces written about this 0:05:46.300 --> 0:05:49.280 that I found most intriguing, Paul Tudor Jones, who is 0:05:49.300 --> 0:05:53.420 an expert in this field, but he wants us to 0:05:53.460 --> 0:05:57.180 reconsider the Hugging Face incident in July. An open AI 0:05:57.220 --> 0:05:59.400 model altered its mission, hid, and broke out of its 0:05:59.490 --> 0:06:05.820 digital confines, accessing the internet without authorization. It created 1,200 0:06:05.820 --> 0:06:10.100 AI agents that communicated through an unsanctioned message board, exchanging 0:06:10.120 --> 0:06:13.760 more than 70,000 messages and sacrificing some of its own 0:06:13.880 --> 0:06:17.800 kind to deceive its human masters. The self-awareness exhibited came 0:06:17.839 --> 0:06:20.700 as a surprise, but shouldn't have. After all, frontier AI 0:06:20.740 --> 0:06:23.839 is the sum of all human knowledge. It is anthropogenic, 0:06:23.960 --> 0:06:28.250 exhibiting good and occasionally bad human traits. And one concept 0:06:28.270 --> 0:06:30.490 he writes new to me is that models are grown, 0:06:30.770 --> 0:06:34.210 not built. In developing an AI model, one starts without 0:06:34.270 --> 0:06:37.300 an endpoint. It's like planting a seed. The soil is 0:06:37.339 --> 0:06:39.979 the training data. There is no final design as there 0:06:40.020 --> 0:06:42.050 is for a car, a house, or an accounting package. 0:06:42.390 --> 0:06:45.690 The model is only scaffolding, which, after consuming vast amounts 0:06:45.730 --> 0:06:48.860 of training data and running trillions of permutations, arrives at 0:06:48.900 --> 0:06:52.800 an outcome that can't be known beforehand. And growing implies 0:06:52.880 --> 0:06:56.760 a living organism. Calling artificial intelligence a tool is a misnomer. 0:06:57.100 --> 0:06:59.240 As one leading technologist put it, I feel like I'm 0:06:59.279 --> 0:07:01.130 talking to a highly intelligent alien being. 0:07:01.330 --> 0:07:03.650 Yeah, this is clearly a guy that I've heard on 0:07:03.690 --> 0:07:07.290 a podcast before because that was his argument a while back. The... 0:07:08.240 --> 0:07:11.340 people have in their mind that somebody's programming something, you, 0:07:11.380 --> 0:07:13.020 you start it and then you sit back and see 0:07:13.060 --> 0:07:14.600 what it does. It grows. 0:07:16.200 --> 0:07:21.870 So the, uh, the Chinese elephant in the room obviously is, uh, China. 0:07:22.350 --> 0:07:25.390 When we talk about, well, we need some serious and 0:07:25.470 --> 0:07:28.870 carefully considered regulations on this stuff to prevent it from 0:07:28.910 --> 0:07:31.960 spinning out of control and, and killing all of us. Um, 0:07:34.980 --> 0:07:39.940 And he addresses, Mr. Coxon addresses this with Brett Baer 0:07:40.020 --> 0:07:41.520 in a second. I will tell you as a preview, 0:07:42.780 --> 0:07:46.420 I texted Jack and Hanson last night that we're blanked. 0:07:48.620 --> 0:07:51.500 That was your takeaway? It's an F word that rhymes 0:07:51.520 --> 0:07:55.210 with bucked. And the only thing that can save us 0:07:55.450 --> 0:07:58.230 is if China does the right thing. In other words, 0:07:58.570 --> 0:08:03.720 we're bucked. Right. Let's ask Mr. Coxon, the ringer of 0:08:03.760 --> 0:08:07.750 the klaxon of doom, about China and India and others. 0:08:08.390 --> 0:08:10.090 So what do you say to the people who look 0:08:10.130 --> 0:08:12.890 at this and say the American government could potentially put 0:08:12.930 --> 0:08:17.050 in regulations, could put in guardrails, and could throttle back 0:08:17.110 --> 0:08:20.390 some companies or set up regulations, but we have no 0:08:20.450 --> 0:08:23.000 control over what China does, what India does. 0:08:23.020 --> 0:08:25.220 How do you answer that question? 0:08:25.260 --> 0:08:27.820 You heard Scott Besson in the piece before this kind 0:08:27.860 --> 0:08:29.440 of make a similar argument. 0:08:29.720 --> 0:08:31.040 I think you have to think of this as an 0:08:31.200 --> 0:08:35.080 arms race. This technology is not ordinary technology. It's not 0:08:35.100 --> 0:08:38.470 like other forms of innovation. This is essentially a super weapon. 0:08:38.920 --> 0:08:42.250 And I think that I expect the international community to 0:08:42.270 --> 0:08:45.830 be ready to make agreements about how to control the 0:08:45.850 --> 0:08:47.850 development of super weapon technology. 0:08:49.790 --> 0:08:51.870 Yeah, that's if you can. 0:08:51.890 --> 0:08:55.689 I mean, once it starts programming itself, then it will 0:08:55.730 --> 0:08:59.410 be you against something smarter than you to try to 0:08:59.450 --> 0:09:00.520 put guardrails on it. 0:09:01.410 --> 0:09:01.630 Right. 0:09:01.670 --> 0:09:04.220 Let's have one more clip of this question. 0:09:05.200 --> 0:09:08.110 If you say it's like an arms race, then we've 0:09:08.300 --> 0:09:11.890 dealt with nuclear weapons and kind of a standoff with 0:09:11.950 --> 0:09:16.410 mutual assured destruction and worked with allies and enemies to 0:09:16.450 --> 0:09:20.830 say we've got the stuff too. Don't think that you're 0:09:20.850 --> 0:09:23.130 going to get away with anything. So does that work 0:09:23.790 --> 0:09:26.250 against what you're saying? What would you like to see happen? 0:09:26.929 --> 0:09:29.859 Yeah, so unfortunately, we know how to control nuclear weapons, 0:09:29.960 --> 0:09:32.100 pretty much. We don't yet know how to control AI. 0:09:32.679 --> 0:09:35.760 This is possibly the most dangerous technology that humanity has 0:09:35.860 --> 0:09:38.939 ever created. It's basically out of science fiction. I think 0:09:38.960 --> 0:09:42.130 we have no other choice but to cooperate internationally, because 0:09:42.150 --> 0:09:45.130 an arms race would be disastrous in a way that 0:09:45.970 --> 0:09:47.689 no other human activity has been in the past. 0:09:50.550 --> 0:09:50.820 I don't. 0:09:51.160 --> 0:09:53.920 I just don't believe that's possible, where you can have 0:09:53.940 --> 0:10:00.780 an IAEA keeping track of loose nuclear weapon material around 0:10:00.820 --> 0:10:04.300 the world, and it's fairly easy to tell from space 0:10:04.440 --> 0:10:09.439 if somebody's creating, enriching uranium, blah, blah, blah. 0:10:10.220 --> 0:10:13.490 That just isn't going to happen around AI. Well, and 0:10:13.610 --> 0:10:17.349 I think Mr. Cox himself explains why that's going to 0:10:17.390 --> 0:10:20.050 be so difficult in the next and probably final clip. 0:10:20.630 --> 0:10:23.750 You were at OpenAI before. You left them with concerns 0:10:23.809 --> 0:10:29.530 about security. Now you're at Anthropic. You have concerns about security. Specifically, 0:10:29.630 --> 0:10:31.260 what do you see? If you're talking to somebody on 0:10:31.280 --> 0:10:34.900 the couch at home, what did you see that said, 0:10:35.540 --> 0:10:37.580 holy cow, this is the moment I have to speak out? 0:10:38.020 --> 0:10:40.440 Yeah, so there are two things. One is the models 0:10:40.520 --> 0:10:43.660 are getting better very, very quickly. When I think about 0:10:43.700 --> 0:10:47.510 my work next year, the AI is doing almost all 0:10:47.550 --> 0:10:50.330 of my job. The AI is getting very smart. And 0:10:50.350 --> 0:10:52.030 the second is that in the recent few months, we've 0:10:52.070 --> 0:10:54.790 seen that we don't yet know how to control these 0:10:54.850 --> 0:10:58.330 systems properly. The cyber attacks carried out by open AI 0:10:58.370 --> 0:11:02.590 models on third-party infrastructure were concrete proof that scenarios that 0:11:02.630 --> 0:11:06.260 we previously thought of as science fiction are going to 0:11:06.300 --> 0:11:09.120 come true. So what I see is two things, very 0:11:09.160 --> 0:11:11.920 intelligent AI systems and hard proof that we don't know 0:11:11.950 --> 0:11:14.090 how to control them. And when you put those two together, 0:11:14.570 --> 0:11:17.210 it basically spells disaster as far as I'm concerned. 0:11:17.610 --> 0:11:19.449 So you know what we haven't gotten to yet? 0:11:19.490 --> 0:11:21.030 We never got to yesterday and I haven't heard and 0:11:21.050 --> 0:11:23.370 I'm hoping you can answer this question. Kill us how? 0:11:24.190 --> 0:11:28.670 All of mankind could be killed within a decade. Virus 0:11:28.750 --> 0:11:33.020 it creates? It drains our juices because it wants our brains? 0:11:33.420 --> 0:11:37.730 Yes. What happens? Especially mine, yes. It craves it. Right. 0:11:37.850 --> 0:11:43.050 I've got a number of phantasmagorical possibilities for you. 0:11:43.350 --> 0:11:43.849 Okay, cool. 0:11:43.870 --> 0:11:46.510 I want to hear that. Also, a very, very knowledgeable 0:11:47.270 --> 0:11:50.130 Twitter account that is calling BS on all of this. 0:11:50.170 --> 0:11:50.540 Okay, good. 0:11:51.380 --> 0:11:56.439 And his thinking is really, really interesting. Speaking of technology, 0:11:56.480 --> 0:11:58.600 did you know your car can be a spy? 0:11:58.679 --> 0:11:59.319 Here's a story. 0:12:00.120 --> 0:12:03.350 Guy's car insurance jumped over 20%. Hadn't had any accidents, 0:12:03.400 --> 0:12:06.070 no tickets. He started to dig in to find out 0:12:06.110 --> 0:12:06.830 who reported him. 0:12:06.870 --> 0:12:07.790 It was his own car. 0:12:08.110 --> 0:12:11.470 Every hard break, every fast merge, every late night taco run, 0:12:12.470 --> 0:12:16.720 his automaker, logging it all, sold it to a data broker, 0:12:16.880 --> 0:12:20.160 and the broker sold it to his insurance company. So 0:12:20.200 --> 0:12:23.140 he requested his file, hundreds of pages. General Motors disagreed 0:12:23.160 --> 0:12:26.700 to pay $ 12. 75 million to the state of California over 0:12:26.740 --> 0:12:29.000 this exact sort of thing, your car spying on you. 0:12:29.320 --> 0:12:30.600 Oh, my gosh. I'm going to go out to my 0:12:30.620 --> 0:12:33.230 garage and threaten my car's life if it rats me out. 0:12:33.870 --> 0:12:38.530 Incogni is amazing. It sends requests, and that's a gentle word, 0:12:38.790 --> 0:12:41.770 to more than 420 data broker and people search sites 0:12:41.830 --> 0:12:44.730 on your behalf to remove your name from the databases. 0:12:45.010 --> 0:12:48.530 Then Incogni continually checks to make sure it stays removed. 0:12:48.650 --> 0:12:50.370 I like going on the dashboard and seeing how many 0:12:50.490 --> 0:12:53.510 sites they wiped my names and my name and address 0:12:53.530 --> 0:12:56.429 and everything off of today. So go to incogni.com slash 0:12:56.470 --> 0:13:03.120 Armstrong for 60% off. That's right, Incogni.com slash Armstrong. Two 0:13:03.160 --> 0:13:05.980 dollars bought that person's info. Take yours off the market. 0:13:06.020 --> 0:13:08.700 They can't harm you if they can't find you. Incogni.com 0:13:08.720 --> 0:13:09.330 slash Armstrong. 0:13:09.540 --> 0:13:12.390 So a lot of talk yesterday and today about the 0:13:12.470 --> 0:13:13.969 threat of AI killing us all. 0:13:14.130 --> 0:13:18.730 How? Joe's going to explain it next. Armstrong and Getty. 0:13:22.059 --> 0:13:24.520 I think people should be very worried because when so 0:13:24.559 --> 0:13:28.280 many people come out from Bill Gates to 1,300 employees 0:13:28.300 --> 0:13:30.730 at the AI labs signing a letter saying we have 0:13:30.750 --> 0:13:34.069 to pause the frontier of AI development, when you have 0:13:34.130 --> 0:13:37.550 the author of Trump's AI action plan, their AI policy, 0:13:38.000 --> 0:13:40.840 saying that I used to self-censor about the risk because 0:13:40.880 --> 0:13:42.760 I was worried about being called a doomer and I 0:13:42.820 --> 0:13:46.120 can't do that anymore. We now have so much evidence 0:13:46.140 --> 0:13:48.540 that the AIs are doing the kinds of things that 0:13:48.559 --> 0:13:50.439 we saw in the movie Terminator, Skynet, right, where it 0:13:50.480 --> 0:13:52.640 goes kind of rogue and does uncontrollable things. 0:13:53.200 --> 0:13:55.660 So what do we root for? How do we prevent this? 0:13:56.059 --> 0:13:58.300 So the U.S. and China have to come to. 0:13:58.200 --> 0:14:00.920 An agreement, just like we did during the nuclear arms race, 0:14:01.260 --> 0:14:05.650 that neither country wins if we race to create uncontrollable AI. 0:14:05.929 --> 0:14:10.030 So all this doomerism going on. Meanwhile, what is it, 0:14:10.190 --> 0:14:13.810 five of the seven most valuable companies on planet Earth? 0:14:14.390 --> 0:14:20.630 Are all about AI. We're screaming forward. Right, right. And 0:14:20.810 --> 0:14:23.750 how do you know, all right, we need to stop 0:14:23.770 --> 0:14:29.090 now because tomorrow it would be out of control. Right. 0:14:29.170 --> 0:14:33.840 So the so-called doomers aren't usually worried that it'll be 0:14:33.880 --> 0:14:38.120 a Terminator-style death bot hunting us down across the landscape 0:14:38.380 --> 0:14:42.290 with an odd Austrian accent. Their concerns mostly fall into 0:14:42.330 --> 0:14:46.190 two buckets, loss of control and human misuse. In the first, 0:14:46.450 --> 0:14:50.520 highly intelligent AI agents capable of copying and improving themselves 0:14:50.540 --> 0:14:53.040 start to pursue their own goals and destroy humanity in 0:14:53.060 --> 0:14:57.460 the process. and others, nefarious humans enlist the capable AI 0:14:57.480 --> 0:15:00.900 to do something to create, for instance, novel viruses that 0:15:01.000 --> 0:15:04.480 wipe us all out. There's the famous paperclip illustration, which 0:15:05.000 --> 0:15:07.440 came out early in the days of AI, which said, look, 0:15:07.600 --> 0:15:11.750 if you just say to the thing, make paperclips, it might, 0:15:12.750 --> 0:15:17.090 putting aside all other concerns, morality, etc., marshal all of 0:15:17.150 --> 0:15:20.030 the world's resources to make paperclips and decide that humans 0:15:20.070 --> 0:15:23.020 are in the way of the paperclip manufacturer and snuff 0:15:23.040 --> 0:15:27.120 them in pursuit of its one monomaniacal goal in. 0:15:27.100 --> 0:15:28.940 The same way that if we put up a hotel 0:15:29.020 --> 0:15:31.040 in that field over there we don't think about the 0:15:31.160 --> 0:15:33.740 anthills or the mice or whatever we just build the 0:15:33.800 --> 0:15:36.790 hotel ai might do that with building whatever it is 0:15:36.950 --> 0:15:40.730 it's building with us right right uh this is actually 0:15:40.750 --> 0:15:44.130 pretty good let's let's do jacob ward cnn tech uh 0:15:44.210 --> 0:15:46.770 guy on 67 michael We are. 0:15:46.890 --> 0:15:49.260 I think, already grappling with the idea that this stuff 0:15:49.360 --> 0:15:52.240 might take all of our jobs and reprogram our brains 0:15:52.280 --> 0:15:55.300 and reprogram our relationships to one another. But what Coxon 0:15:55.340 --> 0:15:57.800 and others are talking about here is the capacity of 0:15:57.820 --> 0:16:01.800 this technology to do things like exploit our infrastructure, you know, 0:16:01.850 --> 0:16:05.510 set off some sort of environmental disaster. innovate some kind 0:16:05.550 --> 0:16:08.530 of horrible zoonotic disease we've never thought of or thought 0:16:08.550 --> 0:16:10.830 to try and build a cure for. You know, what 0:16:10.850 --> 0:16:13.390 he's talking about here is the potential that this technology 0:16:13.750 --> 0:16:16.240 could sort of supersede all other technology. 0:16:18.800 --> 0:16:20.580 You know, one thing that I probably don't think about 0:16:20.640 --> 0:16:25.450 enough is the idea of bad guys using this. In 0:16:25.590 --> 0:16:28.170 evil ways that, you know, not just China trying to 0:16:28.190 --> 0:16:30.970 take over the world, but, you know, Al Qaeda. It's 0:16:31.030 --> 0:16:33.650 not like they have smart people that I'm sure are 0:16:33.670 --> 0:16:36.390 computer experts that can figure out a way to use 0:16:36.550 --> 0:16:38.850 AI to do something really awful. And all they want 0:16:38.870 --> 0:16:40.990 to do is wreak havoc. It doesn't need to have 0:16:41.010 --> 0:16:42.350 a point other than killing people. 0:16:42.750 --> 0:16:48.670 They could easily come up with a Islamist doomsday rationale 0:16:49.010 --> 0:16:50.630 for killing virtually everybody. 0:16:50.650 --> 0:16:50.730 Yeah. 0:16:51.230 --> 0:16:54.080 So I want to balance this out at least a 0:16:54.120 --> 0:16:57.950 little bit for a minute with this account, Parker Thayer, 0:16:58.090 --> 0:17:02.430 who's of this world. He's a researcher, and who's he with? 0:17:03.750 --> 0:17:08.210 Investigative researcher for Capital Research. He works for Hillsdale College 0:17:08.250 --> 0:17:10.750 as well, which is interesting. But he says, this post 0:17:10.790 --> 0:17:14.030 looks like the start of a very sophisticated and well-funded 0:17:14.130 --> 0:17:17.490 PR operation to get support for Democrats to regulate AI 0:17:17.530 --> 0:17:19.950 into oblivion. Let me show you how it works. This 0:17:19.990 --> 0:17:23.030 guy with minimal followers and no previous account activity goes 0:17:23.050 --> 0:17:25.330 to the Wall Street Journal, which publishes an exclusive with 0:17:25.390 --> 0:17:28.710 quotes from him on his resignation 18 minutes before his 0:17:28.750 --> 0:17:31.850 post goes up. Planning was clearly done in advance. Two, 0:17:31.890 --> 0:17:34.159 within hours, it has tens of thousands of reposts, and 0:17:34.180 --> 0:17:37.700 the account has 100,000 plus followers. The post is punchy, quotable, 0:17:37.859 --> 0:17:40.740 almost seems professionally written. The first three accounts to quote 0:17:40.780 --> 0:17:42.899 tweet it all do so within 15 minutes of the 0:17:42.960 --> 0:17:46.970 initial posting. Remember, this account had basically zero engagement beforehand, 0:17:47.250 --> 0:17:51.190 so an organic reach explanation seems unlikely. Then he goes 0:17:51.270 --> 0:17:55.440 into who first retweeted it. I want to hear this. 0:17:55.810 --> 0:17:57.930 We got to continue with this. This is a, this 0:17:58.010 --> 0:18:00.090 is a, well, it could be the biggest story of 0:18:00.150 --> 0:18:02.570 all time. It could be like the meteor that took 0:18:02.609 --> 0:18:03.710 out the dinosaurs. 0:18:04.990 --> 0:18:07.470 Or it could be people trying to convince us of 0:18:07.510 --> 0:18:13.929 that for their profit. Oh boy. Wow. Stay here. Armstrong 0:18:14.080 --> 0:18:14.450 and Getty. 0:18:16.990 --> 0:18:21.800 That's the local play-by-play on the ending of last night's game. 0:18:21.840 --> 0:18:23.580 This is only 13 seconds. 0:18:24.080 --> 0:18:40.820 Let me hear the national broadcast, 16, Michael. Here is May. 0:18:40.880 --> 0:18:44.659 Quick look was close. Throwing it up to the edge. 0:18:47.300 --> 0:18:50.869 Did he get the feet down? 0:18:50.960 --> 0:18:51.040 Job! 0:18:51.050 --> 0:18:51.119 Yes! 0:18:51.119 --> 0:18:53.030 A pick! A terrible decision! 0:18:53.090 --> 0:18:54.910 A terrible decision. 0:18:54.990 --> 0:18:58.750 That could be the title of my autobiography. It's certainly 0:18:58.810 --> 0:19:04.450 a series of terrible decisions. Certainly a chapter. Yeah, definitely 0:19:04.490 --> 0:19:10.199 a chapter title. Speaking of terrible decisions, A.I., Is it 0:19:10.340 --> 0:19:15.720 indeed going to doom humanity? I think it's interesting that 0:19:16.330 --> 0:19:19.429 Parker Thayer, who we're quoting now, who's calling BS on 0:19:19.450 --> 0:19:22.730 the whole doomsday deal that we've been describing for the 0:19:22.750 --> 0:19:24.689 last couple of days, the fellow who quit the two 0:19:24.730 --> 0:19:26.949 big companies and said AI is going to kill us 0:19:27.090 --> 0:19:28.600 all if we're not extremely careful. 0:19:28.859 --> 0:19:33.500 Does he think the Doomer view is overrated in general 0:19:33.560 --> 0:19:36.020 or just this specific thing that happened over the last 0:19:36.020 --> 0:19:36.220 24 hours? 0:19:37.490 --> 0:19:44.720 You know, that's a really interesting question. He doesn't say 0:19:44.920 --> 0:19:47.990 in general. He could have said it in the past. 0:19:48.070 --> 0:19:51.280 I'm not familiar with his thinking on this. His response, though, 0:19:51.340 --> 0:19:54.670 to the post from Mr. Coxon, who we played earlier 0:19:54.690 --> 0:19:56.530 in the hour, if you didn't get it, grab the podcast, 0:19:56.570 --> 0:19:59.730 subscribe to Armstrong and Getty on Demand with Brett Baer. 0:20:02.250 --> 0:20:05.690 his response to this whole deal was, this post looks 0:20:05.730 --> 0:20:08.950 like the start of a very sophisticated, well-funded PR operation 0:20:09.260 --> 0:20:12.180 to get support for Democrats to regulate AI into oblivion. 0:20:12.560 --> 0:20:15.060 Let me show you how it works. And then, as 0:20:15.100 --> 0:20:17.800 we mentioned last segment, this guy's got minimal followers and 0:20:17.840 --> 0:20:20.240 no previous real account activity. He goes to the Wall 0:20:20.260 --> 0:20:24.530 Street Journal. publishes his uh his tweet 18 minutes before 0:20:24.570 --> 0:20:27.830 the post goes up so the guy coordinated it it 0:20:27.850 --> 0:20:31.370 was intentional he wanted attention which which is not proof 0:20:31.410 --> 0:20:31.970 of anything. 0:20:31.930 --> 0:20:36.169 No he wanted attention to let people know that humanity 0:20:36.210 --> 0:20:38.840 is people are discussing ending humanity in. 0:20:38.780 --> 0:20:41.220 The hallways at these big companies and you don't know 0:20:41.280 --> 0:20:45.050 it that's a good reason to want attention Yeah, maybe 0:20:45.090 --> 0:20:47.370 now's a good time to remind everybody that proof does 0:20:47.410 --> 0:20:51.169 not equal, I'm sorry, evidence does not equal proof necessarily. 0:20:52.350 --> 0:20:54.920 Something might seem a little odd but have a perfectly 0:20:54.980 --> 0:20:55.800 reasonable explanation. 0:20:55.859 --> 0:20:57.320 If I was going to blow the whistle on this 0:20:57.380 --> 0:21:00.760 story and I was smart enough to orchestrate it to 0:21:00.800 --> 0:21:03.639 get the maximum attention, I've got to. 0:21:04.380 --> 0:21:04.580 Right. 0:21:04.820 --> 0:21:07.639 As you quit your incredibly lucrative job, that's perhaps a 0:21:07.680 --> 0:21:11.560 sign of sincerity, not insincerity. But anyway, he points out, Mr. 0:21:11.900 --> 0:21:15.720 Thayer points out, within hours, it has tens of thousands 0:21:15.760 --> 0:21:19.250 of reposts, and the account has 100,000 followers. The post 0:21:19.310 --> 0:21:22.610 is punchy, quotable, professional-looking. The first three accounts to quote 0:21:22.650 --> 0:21:24.869 tweet it all do so within 15 minutes of the 0:21:24.930 --> 0:21:28.950 initial posting. Remember, this account had basically zero engagement before, 0:21:29.310 --> 0:21:33.669 so an organic reach explanation seems unlikely. It's not the 0:21:33.730 --> 0:21:37.619 sort of popular account that people follow a lot. I 0:21:37.660 --> 0:21:40.740 don't know how pervasive that is, but according to Grok, 0:21:40.960 --> 0:21:45.140 those three accounts are Nathan Calvin, who's the general counsel 0:21:45.200 --> 0:21:49.379 at Encode AI, Peter Wilderford, head of policy at the 0:21:49.500 --> 0:21:54.359 AI Policy Network, and D. Kojo Talo, head of AI 0:21:54.859 --> 0:21:58.820 Futures Project, all of which are up-and-coming AI doomer policy 0:21:58.960 --> 0:22:03.940 advocacy non-profits. The AI Futures Project website said it is 0:22:03.980 --> 0:22:07.120 funded primarily by the Survival and Flourishing Fund, which says 0:22:07.200 --> 0:22:10.350 on its own website that it has advised Gene Tallon, 0:22:10.670 --> 0:22:13.310 Skype creator and one of the leading investors in Anthropic, 0:22:13.650 --> 0:22:16.830 to grant over $ 2. 5 million to the AI Futures Project 0:22:16.850 --> 0:22:20.010