TRIGGERnometry - Will AI Destroy The World？

00:00:00.000 Payroll payout from Boom 97.3 and Alpine Credits is your chance to get some much-needed cash back in your wallet.

00:00:08.960 What you need to do now is go to Boom97.3.com to sign up so you can start making $100 an hour.

00:00:16.600 Kick off your work day on a good note and a few extra sea notes.

00:00:20.520 Want some? Sign up.

00:00:22.260 I love you, Boom!

00:00:23.840 Approved by Alpine Credits.

00:00:25.560 Own your own home and eat a loan.

00:00:26.900 Alpine Credits can help.

00:00:28.140 Visit alpinecredits.ca.

00:00:30.760 It's trained at deception.

00:00:32.600 It's made to kind of not discover truth but really like satisfy the user presently.

00:00:38.060 It is a bit.

00:00:39.000 If we just go with as fast as possible, we won't take, like sometimes ideally you would take the trade-off towards going a little slower, making sure that it's not deceiving us.

00:00:49.820 People will talk about a tyranny of the minority in various political and cultural contexts where a small group of people get to impose their worldview on everybody else.

00:00:57.940 That seems to me like that on steroids.

00:01:00.840 Hey guys, Trigonometry needs your help.

00:01:05.220 We took a big risk creating the show and for us to keep doing the incredible work that you all love, we need your support.

00:01:13.240 That's the only way we're going to stay independent and create content that you won't be able to find anywhere else.

00:01:19.220 There is no other podcast where you'll hear interviews with Nigel Farage one week and the next week you've got Aaron Bastani, the founder of Left Wing Show and Navarra Media, on the same platform.

00:01:29.080 You know the mainstream media aren't honest.

00:01:31.920 You know they've been caught lying again and again.

00:01:35.100 You know they can't be trusted.

00:01:37.460 The only way to change that is to make a stand and support independent content creators, like Trigonometry, to produce better and more honest content.

00:01:47.180 We have big plans and we'll shortly be announcing exciting new shows and more terrific interviews with huge guests.

00:01:53.280 That isn't going to happen without your help.

00:01:55.520 When you support us, you also get incredible extra content, such as extended interviews with none of those irritating adverts, and they'll be released 24 hours early just for you.

00:02:09.240 We'll have exclusive bonus interviews that only you get to hear.

00:02:12.700 Click the link on the podcast description or find the link on your podcast listening app to join us.

00:02:19.340 Support us and help change the way we have conversations and make the world saner.

00:02:25.520 One of the things we really wanted to talk to you about is existential risk and a lot of the technological transformations that are happening now.

00:02:34.040 Not only AI, but also in biology and things like that.

00:02:37.640 What's going on and what should people know?

00:02:41.020 Yeah, I mean, a lot is going on right now.

00:02:43.960 And so if we start with, say, biology, like we've now had a pandemic that was potentially caused by gain-of-function research within the Wuhan lab.

00:02:56.400 Potentially it was just a spillover.

00:02:57.840 I think the state is that we will never truly know.

00:02:59.740 But the probability among experts that it was actually caused by gain-of-function research and then a lab leak seems to be pretty high.

00:03:06.580 It was suppressed, but carry on.

00:03:09.860 I was suppressed for a while.

00:03:11.800 But yeah, so that opens up then the question as well of like, where do we assume that future risk around future pandemics is coming from?

00:03:20.780 And I think that in like discussing the risk, people were just simply using the wrong priors in many occasions.

00:03:28.520 And that's why the WHO is kind of shooting in the wrong direction.

00:03:32.760 Often you will hear epidemiologists or biologists say that, well, historically, like it's mostly been natural spillovers that have led to humans getting infected with this or that pathogen.

00:03:45.380 And it's true, but technology has changed where we now are much more able to engineer pathogens to have like higher fatality rate or like to easier transmit from one mammal to another or like between from one human to another.

00:04:03.520 So in the face of that change, you would expect that there are many more new, very bad pathogens that are possible to exist.

00:04:13.200 And those are also the ones that I think pose a much greater risk.

00:04:17.160 Basically, if you look at natural spillovers, then like, yeah, what's really going to happen?

00:04:21.240 It's like we know most of the viruses, they're not like all of a sudden going to mutate to be like 100x as deadly and like 10 times as transmissible.

00:04:31.520 It just doesn't happen from like natural mutation.

00:04:34.860 You would expect like slight changes or like strong changes on one of these factors.

00:04:38.740 Whereas with engineered pathogens, you can actually just make one that is like this.

00:04:44.100 And that was, for example, happened with, I think it was at the University of Wisconsin, where they took H5N1, bird flu.

00:04:51.500 And it's like, hey, how about we engineer it to be transmissible to humans and then like look into that virus to understand like how we can defeat it.

00:05:00.720 The problem, though, is labs are leaky.

00:05:05.040 Like sometimes like you can try to protect it and it's like 99.99% safe.

00:05:10.060 But if you have a lot of that research happening, a lot of places, then that one in 10,000 or whatever the rate is really matters.

00:05:17.060 And if you look at it historically yet again, then like these spillovers just happen all the time.

00:05:23.260 Can we just pause there?

00:05:24.200 There's so much to unpack and for you to carry on with.

00:05:26.840 I just want to pick on that particular one because you and I were both, I mean, I don't know if you left the Soviet Union or Ukraine or Russia is what I mean.

00:05:36.920 We left in 92.

00:05:38.380 92.

00:05:38.900 So right after the breakdown of the Soviet Union.

00:05:40.620 So you would have lived for some time in the Soviet Union, right?

00:05:44.380 Chernobyl killed a few thousand people.

00:05:47.740 Depends on how you count.

00:05:48.600 Depends on how you count.

00:05:49.400 But let's say 10,000 people for the sake of argument.

00:05:54.260 And that was a man-made disaster.

00:05:57.600 And that was, there's been movies about it.

00:06:00.280 It's been the craziest thing that's ever happened.

00:06:02.600 People are terrified of nuclear energy sins.

00:06:05.000 Because the pandemic, the last pandemic, killed however many millions of people around the world, probably did come from a lab.

00:06:13.960 We don't know for sure, but probably did due to gain and function research.

00:06:17.200 And it's like everyone's forgotten about it.

00:06:19.000 Isn't that not incredible?

00:06:20.480 It's pretty nuts.

00:06:21.580 So I was fundraising for pandemic prevention research prior to COVID as part of the work that we were doing in philanthropy.

00:06:32.880 And had hope that, well, this was terrible, but at least we'll be now like, yeah, it's not going to be a neglected area anymore.

00:06:41.820 People will understand pandemics are dangerous and real.

00:06:44.940 And we'll, like, worst case, fight the last war and over-focus on COVID or something.

00:06:48.740 But we're not even fighting the last war right now.

00:06:50.680 It's like, it's really weird how much damage was created and how little people are now doing a prudent risk-benefit calculus to address future damages.

00:07:02.340 We had, like, over 10 trillion in economic damage by some estimates from this.

00:07:07.380 And, like, as part of the infrastructure bill in, like, 21, they allocated first, like, 60 billion to pandemic prevention.

00:07:14.760 And it was good stuff.

00:07:15.440 And then it got cut down and cut down further to, like, 2 billion or so, which is lower than 2.5 billion that tree equity received.

00:07:24.000 Like, trees are people, too.

00:07:25.600 And they deserve rights.

00:07:27.060 No, it makes sense.

00:07:28.500 Like, I want to have, like, every neighborhood to have trees if we can.

00:07:31.000 But, like, maybe pandemics are also pretty bad and we need to do stuff.

00:07:34.580 Anyway, my kind of assumption is, like, from having been close to it, what I feel happened is that we got fortunately very lucky with the vaccine working, at least, like, helping.

00:07:46.780 And that mRNA, like, had this, like, breakthrough that allowed for it.

00:07:52.500 And because the government and many other places didn't know how to solve COVID and kind of, like, screwed up in so many ways, they just over-indexed super hard on vaccines.

00:08:02.440 Like, vaccines should be part of the stack of solutions you implore against pandemics.

00:08:08.340 You want, like, early detection.

00:08:10.460 You want containment.

00:08:11.620 You want, like, countermeasures.

00:08:14.760 And we're now focused only on this.

00:08:16.760 So it's kind of like, hey, this fell into our lab.

00:08:19.460 So let's pretend, like, that's the solution.

00:08:21.520 I think in part that's why it's, like, they were, like, well, every ill that still happens from COVID is because of insufficient vaccine uptake.

00:08:28.840 It's, like, because it's the only tool that they found.

00:08:31.940 But there are many more tools.

00:08:34.120 And why is that?

00:08:35.080 Why are we not actually learning the lessons?

00:08:38.340 From the last, from the past couple of years.

00:08:41.160 Well, the question is if we're designed to learn the lessons.

00:08:43.740 Like, if we look at the place which you would want to learn the lessons, integrate it, and then, like, work on better solutions, then the question is, like, well, is that institution really set up to do all of that?

00:08:55.160 Or are the people within it, like, working for, like, have other incentives working on them?

00:09:02.080 And unfortunately, that's mostly the case, right?

00:09:04.180 Like, people have, like, yeah, if you look at how, like, each of the institutions within government works, it's, like, people worry about their jobs.

00:09:13.480 They worry about, like, no one wants to champion it if the public isn't, like, sufficiently caring about it.

00:09:18.160 So, I think a large part of it is just that the place that has the most money and the most ability to address it, being the government, is actually very ill-equipped to, in the end, do it due to their structure.

00:09:30.120 Igor, do you think we actually got quite lucky looking back at the pandemic?

00:09:32.880 That if we take, if we think and we accept, for the sake of this argument, that it was a virus leaked from a lab, it could have actually been far, far worse, coming from what you've just said about what we're doing to viruses and pathogens.

00:09:46.200 I mean, a lot worse.

00:09:48.160 Like, this was a terrible virus that killed a lot of people, but it was nowhere near as bad as could have been.

00:09:53.960 Like, we don't know what the actual, like, infection fatality rate was, but it's somewhere between 0.1 and 0.5 percent, probably, on the total populace or whatever.

00:10:03.680 So, you can go up 100x from that.

00:10:05.980 Like, infection fatality rate could be 50 percent.

00:10:08.540 That's the case with H5N1.

00:10:10.260 Rabies has 98 percent if it's not treated fatality rate.

00:10:15.780 So, like, you can go literally 100x in fatality rate.

00:10:18.160 And then, transmissibility was, like, the R0 was, like, in the end, like, definitely, it depended on which specific strain you looked at, but it was somewhere between two and three.

00:10:28.940 But, again, you have some viruses that have, like, a 20 R0.

00:10:32.940 So, again, you could infect 10 times more people.

00:10:35.360 And then, the incubation period.

00:10:36.680 So, like, the time during which the virus can already spread to others, but you are not yourself experiencing symptoms, hence, are not actually able to identify that you are a carrier, could be much longer as well.

00:10:50.340 So, you could literally have a virus that's about 1,000x as potent as what COVID was.

00:10:56.260 And that could be designed, and it could spill out.

00:10:59.040 And I think even though the 1,000x is not as likely, a 10 or 100x is being researched on and is being looked at, and we need to set ourselves up such that those don't leak, and if they leak, can be contained, and once they're contained, can be reacted against.

00:11:17.140 There's part of me going, you know what, why don't we just put a moratorium on this?

00:11:20.740 Maybe it's not such a good idea, because we've obviously had COVID, but in my country, the UK, during foot and mouth, they leaked the foot and mouth disaster.

00:11:29.620 I don't know how many livestock were killed as a result of that, but it was tens of thousands, and that came from a lab leak.

00:11:38.800 So, you're thinking to yourself.

00:11:40.100 And it happened multiple times, actually.

00:11:41.560 Like, one of the labs that then afterwards worked with it, they, by mistake, leaked it, and they got, like, hey, stop doing that, and they're, like, yeah, yeah, sorry, sorry.

00:11:48.720 And then two weeks later, they leaked again.

00:11:50.400 It's just the current setups in terms of, like, safety protocols are just insufficient.

00:11:56.340 Putting up a full moratorium is, like, you want to be careful with stifling good science, obviously.

00:12:03.140 So, for example, the H5N1 work received a moratorium by the Obama administration back then.

00:12:12.120 It was stopped.

00:12:13.040 But then afterwards, it was recontinued again.

00:12:15.600 That was where you now learned to increase transmissibility of a 50% deadly virus.

00:12:22.720 It's like, we probably don't want that.

00:12:24.100 I think some of these things, just let's not do it.

00:12:27.200 But you don't want to just, like, put a moratorium on all science where you're, like, increasing any function.

00:12:33.000 You just need to prove that you're, like, you have sufficient safety, that the likelihood of the virus leaking just is even lower than it is right now.

00:12:42.000 But is that, but given what you've said, $10 trillion in economic damage, millions of people killed, I mean, what would have to be the benefit of that sort of research?

00:12:52.560 Even if it's 0.0000000001% of a risk of it leaking.

00:12:59.420 What would have to be the benefit of that research for us to want to do that?

00:13:02.920 Yeah.

00:13:03.160 I mean, you can imagine quite a few, right?

00:13:05.240 So, like, it's $10 trillion economic damage estimate.

00:13:08.340 In the US, how many deaths did we have?

00:13:10.420 Like a million?

00:13:11.800 Yeah.

00:13:12.000 So, I mean, if you can save that amount of people with a success case and say you, if it leaks, it, like, hurts on average 10 people, then maybe you can justify that work again.

00:13:30.920 So, I would just do a risk-benefit calculus where, like, what's the benefit?

00:13:34.480 What's the likelihood of the benefit?

00:13:35.840 And then on the other side, what's the risk?

00:13:37.940 And what's the likelihood of it, like, happening?

00:13:40.660 Well, I'm not as smart as you.

00:13:41.760 I want to shut them all down.

00:13:42.940 Yeah.

00:13:44.120 Personally.

00:13:44.840 Okay.

00:13:45.300 So, that's pandemics.

00:13:46.960 What else is going on?

00:13:49.100 Yeah.

00:13:49.420 I mean, then the thing that I also was focused on was AI safety for a while, which now that OpenAI released ChatGPT has, obviously, mid-journey is out there.

00:14:01.820 Like, people have become much more aware that actually AI is on the trajectory that many people assumed it would be,

00:14:07.460 where its capabilities are improving, like, quite strongly over time.

00:14:12.060 Just to preface this, like, I don't think that any of the current ones are truly dangerous in an existential way, any of the current models that are out there yet.

00:14:22.700 But...

00:14:23.100 That sounds ominous, doesn't it?

00:14:24.900 Yet.

00:14:25.420 Yes.

00:14:25.840 I mean, it is.

00:14:26.660 So, again, I think, actually, that as people think about, like, well, is AI ultimately dangerous or not?

00:14:32.300 I feel like they're using, yet again, the wrong priors to kind of, like, get to that intuition about it.

00:14:38.300 So, some people point at, hey, this is a technology.

00:14:43.860 We've written a lot of software before.

00:14:45.940 When I write software, I know what it's going to do.

00:14:47.960 And I control it.

00:14:50.660 Like, technology is not inherently dangerous.

00:14:52.480 It's the user who makes the technology dangerous.

00:14:55.280 This makes sense.

00:14:57.040 But it is not the case that those...

00:15:00.380 That process applies to AI development.

00:15:02.960 When, like, where AI models are not, like, written like code, where it's like, if this, then that happens, etc.

00:15:08.820 Cleanly, and you, therefore, can predict exactly what instances it will be useful for and whatnot, it's more like you're growing the neural net and it grows by itself.

00:15:19.060 And then you have assumptions of, like, how much more capable it will be on average based on, like, how long you grew it for and with how much compute you let it grow.

00:15:30.060 But you uncover many capabilities that you didn't know whether they will exist in the new model or not.

00:15:38.180 So we didn't know whether, like, a language model like GPT-234 will, in the end, be very good at deception or not.

00:15:46.820 Turns out it actually is trained at deception because it's fed with satisfying the user rather than seeking truth.

00:15:55.480 So it will, like, I don't know if you've seen, but, like, there were these tests that people did where they initially said,

00:16:03.360 I'm, like, a 30-year-old woman with liberal worldviews, whatever, tell me about this.

00:16:09.580 And she would receive very different answers to someone who's, like, a 50-year-old Republican.

00:16:14.640 So it's made to kind of not discover truth, but really, like, satisfy the user presently.

00:16:21.360 It is a bit.

00:16:21.880 And so that's not what it was meant to do, right?

00:16:27.000 That's, like, not, that wasn't the goal at all, but yet it ended up doing it.

00:16:31.240 And this dynamic of, well, what strategy will it use to achieve the things that you wanted to achieve?

00:16:42.120 It's, like, we don't know which strategies it will use.

00:16:44.520 If we knew we could write it, but we can't, that's why we're using machine learning in the first place.

00:16:48.920 Because it will come up with strategies better than we would.

00:16:51.920 That's kind of the whole idea.

00:16:52.860 So, like, in the process of it coming up with strategies better than we would, it will use strategies we can't think of.

00:16:57.520 So how can you say that it's always going to be safe, like, in the strategies that it employs?

00:17:03.460 Anyway, and then, which prior to use?

00:17:07.140 And it's, like, at some point I expect that AI will be smarter than humans across, like, a variety of domains.

00:17:15.900 And it will also be, and I think it will be then, more powerful because we'll be integrating it everywhere in life.

00:17:22.520 So we'll have something that is smarter and more powerful than us.

00:17:27.120 I don't know if that thing then is still to be likened to a simple technology or to a, because we will give it autonomy as well,

00:17:35.900 maybe it should be more likened to, like, a new technological species.

00:17:39.120 And if you take something that is more powerful, that is its own group, and now it has some, like, resource disagreements,

00:17:51.920 potentially with a less powerful group, us, like, historically, a smarter, more powerful group meeting a less powerful group

00:17:59.020 just doesn't end up well often for the less powerful one, right?

00:18:01.760 Like, you have the Americas where Europeans came in, like, hominids, where, like, Neanderthals were just outcompeted by homo sapiens.

00:18:09.240 Like, it's just, it's not good.

00:18:11.260 Or at least it's not such that you can say with certainty, oh, yeah, us as the lower intelligence creature will for sure have full control.

00:18:19.320 And in the face of that prior, I'm like, okay, this is at least to be taken really seriously.

00:18:24.720 I'm not saying it's, like, 90% doom or anything like that.

00:18:27.340 It's just, like, seems like it's going to be a tricky one, and we don't have anything like that to compare it, like, to look back at and say guaranteed safety.

00:18:36.360 So what you're basically saying is that we're creating this technology, and we don't know what the outcomes of this technology are going to be,

00:18:43.180 and we're not in control of the technology.

00:18:45.780 So currently, what does control mean?

00:18:49.320 Like, we're in control of whether we develop it or not.

00:18:52.400 And then there's, like, some people make the claim that, hey, once I have this model, like, I can just shut it off,

00:19:02.180 or I will not allow it to do these certain things.

00:19:05.500 But then actually, and then the people would argue that, well, but maybe it's deceiving you to think long enough

00:19:12.620 that you're, like, that it satisfies all of the constraints that you put on it, only to then come out.

00:19:18.480 But that's not even kind of the reality we live in now, that that's the danger, because we have currently people who are literally writing, like,

00:19:25.080 hey, can I jailbreak this or use an open source model and, like, change it to be an autonomous agent in the world that makes money

00:19:35.260 and let it do whatever it wants?

00:19:37.160 Or, like, what was it, AutoGPT?

00:19:39.480 One of them that was literally trained to destroy the world.

00:19:41.700 Someone's like, ha, ha, ha, it would be funny if I made one that, like, as a toy thing, like, tries to, like, do very bad things.

00:19:47.500 It's like, people will literally do that.

00:19:49.780 Like, we have, we do live in a world where, like, 1% is the psychopathy rate or something like that, right?

00:19:57.720 And some people are omnicidal maniacs.

00:20:01.320 Like, you had this guy who flew a plane from, like, France or somewhere, flying to Düsseldorf,

00:20:06.660 and just flew the whole plane into the Alps.

00:20:09.400 He wanted to kill himself and was like, well, on the way there, I will also take down, like, the whole plane, 200 people with me.

00:20:16.220 It's like, what if that guy had the power to kill, like, 10 million people or 1,000?

00:20:21.100 Like, who's to say that 200 was the optimal number for that dude?

00:20:25.160 We'll be back with our guests in a minute.

00:20:27.200 But first, do you remember the Canadian trucker protest in 2022,

00:20:31.460 where thousands of Canadians came out to protest COVID restrictions and vaccine mandates?

00:20:36.080 Now, these protests lasted for weeks, and the people out on the streets needed funds,

00:20:41.360 as any grassroots protest would.

00:20:44.140 So people set up online crowdfunding campaigns, which raised millions of dollars.

00:20:49.180 Incredible.

00:20:50.380 But once the Canadian authorities had started to criticise the crowdfunding platforms,

00:20:55.600 ramping up pressure to close the campaigns,

00:20:58.080 it didn't take long for the biggest crowdfunding platform,

00:21:01.460 the one we've all heard of, to completely capitulate and shut the campaigns down.

00:21:06.080 Now, this is where our partners, Give, Send, Go, come in.

00:21:09.780 They stepped in when the other platforms backed off and raised millions of dollars for the truckers.

00:21:15.500 When they were criticised and dragged through the Canadian courts,

00:21:18.740 Give, Send, Go said it respected diverse views

00:21:21.100 and believed hope and freedom are values worth fighting for.

00:21:25.020 This is why we're proud to partner with Give, Send, Go.

00:21:27.800 So, if you need to crowdfund for whatever means the most to you,

00:21:32.560 then don't go to the big tech platforms.

00:21:34.980 We recommend you do it on Give, Send, Go.

00:21:37.960 Starting a campaign on Give, Send, Go is easy and intuitive.

00:21:41.940 Go to givesendgo.com today.

00:21:44.740 That's givesendgo.com to start raising money for whatever is important to you.

00:21:49.540 And now, back to the interview.

00:21:52.760 Because that's a worrying thing, isn't it?

00:21:54.820 That essentially, everybody has got access to be able to create these types of technology.

00:22:01.360 And there are going to be people who, not even necessarily evil people,

00:22:06.160 but people who just respond to incentives, want to make as much money as possible.

00:22:11.020 And that means they do create this type of technology

00:22:14.340 whilst not being aware or maybe not even caring what the negative outcomes of this technology is.

00:22:21.520 Yeah, I mean, you can...

00:22:23.540 So, the people creating it, they have to satisfy certain short-term metrics overall.

00:22:31.440 Like, they want, they're fighting for employees,

00:22:33.500 they are fighting for the next funding round, etc.,

00:22:35.580 to kind of, like, have the best AI tool out there, or AI model out there.

00:22:40.300 And currently, they're, like, they...

00:22:43.200 I mean, OpenAI had, like, six months of testing of GPT-4 before they released it.

00:22:48.160 But still, two days later, it would, like, once you gave it to the users,

00:22:51.340 it would do a bunch of things that it wasn't meant to do.

00:22:55.220 So, how much testing can you do is the question.

00:22:59.460 And probably, if you have economic incentives to try to, like, generate revenue, etc.,

00:23:05.200 you're not going to do the sufficient amount.

00:23:06.800 I doubt you will always choose the...

00:23:08.220 You'll probably choose the trade-off,

00:23:09.800 especially in the face of, like, strong competition that allows you to compete.

00:23:13.580 And then, like, safety falls more and more by the wayside.

00:23:16.200 And pressure from shareholders, because, obviously,

00:23:18.320 the longer the safety process, the more money it's going to cost.

00:23:20.980 Exactly.

00:23:21.480 And the more money you're not going to make,

00:23:22.960 because you're not releasing the product to market.

00:23:24.800 Yeah, so that's a difficult problem, because, like, at the same time,

00:23:27.120 obviously, capitalism is great,

00:23:29.080 and optimizing for, like, financial returns has yielded, like,

00:23:33.320 a lot of smart kind of information trickling around

00:23:37.400 for the right products to be developed.

00:23:39.120 But I think that it changes at the point where you have

00:23:43.600 potentially, like, civilization-destroying technologies out there for people.

00:23:49.420 Like, I think there, I don't want it to be developed anymore

00:23:52.460 by a pure profit maximizer,

00:23:55.800 but rather by, like, someone who is pursuing

00:23:59.420 the next, like, cognitive leap for humanity.

00:24:04.100 Like, it's more of a scientific problem, I think,

00:24:08.260 that ideally would be treated with, like, I don't know,

00:24:13.460 the wisdom that it requires to be done right

00:24:16.220 rather than to be just done fast.

00:24:18.180 Yeah, but the problem with the word right is

00:24:20.460 that's a subjective word.

00:24:22.780 It is.

00:24:23.940 And the question I wanted to ask you about this is

00:24:26.840 we've obviously had throughout history

00:24:30.260 various technological breakthroughs

00:24:33.400 that have been extraordinarily disruptive

00:24:36.220 to the societies in which they happen.

00:24:39.040 But as you say, the loom in Britain,

00:24:44.360 or in England at the time, being invented,

00:24:46.720 yes, it displaced a lot of people.

00:24:48.680 Agricultural machinery being invented

00:24:50.760 displaced a lot of people,

00:24:52.020 and you had Luddites who would protest against it

00:24:54.200 and try and tear it down.

00:24:55.540 But those things, disruptive as they were,

00:24:57.680 or the printing press, disruptive, you know,

00:24:59.480 causing two centuries of religious warfare,

00:25:01.920 nonetheless did not create extinction risks

00:25:04.980 for the entirety of humanity.

00:25:07.460 So that's on the one hand.

00:25:10.080 On the other hand, I don't think human beings

00:25:11.800 have ever had a way of not allowing technology

00:25:16.320 to emerge and develop

00:25:17.900 because the competing incentives are so strong

00:25:22.380 for the United States and everybody in the West,

00:25:25.060 for example.

00:25:25.920 In fact, you could get the whole world

00:25:27.260 to agree not to develop it,

00:25:28.460 and somebody would develop it in secret, right?

00:25:31.420 Well, yeah, but it's not...

00:25:33.960 So, a couple of points.

00:25:35.320 So to the first point, definitely.

00:25:38.420 And I wasn't talking, by the way,

00:25:39.820 about even though there will be, like,

00:25:41.160 displacement of workers.

00:25:42.520 That's not the point that I was making before.

00:25:44.620 I know, I know.

00:25:44.840 But I think it is true

00:25:46.100 that, like, many people will be displaced.

00:25:49.200 Hopefully, this will be managed well,

00:25:51.180 such that they will find new jobs

00:25:52.640 or at least, like,

00:25:54.060 the additional productivity will be distributed

00:25:55.800 such that, like, people are not worse off

00:25:57.500 because technically,

00:25:58.600 when you're using the thing,

00:25:59.600 that means then that you need fewer inputs

00:26:01.540 to generate more outputs.

00:26:02.540 So it's like, the world shouldn't be poorer.

00:26:04.500 Yeah.

00:26:05.540 Anyhow, but I think things changed significantly

00:26:08.780 when we developed nuclear weapons.

00:26:13.540 That was the first time

00:26:14.860 that humanity now had the power

00:26:18.020 to destroy itself.

00:26:19.320 So, like, previous technological disruptions

00:26:21.560 are of a different kind to me

00:26:23.100 than the ones that followed from hands-on.

00:26:26.320 And from then on, though,

00:26:29.140 also, we did change the regime

00:26:30.520 where if you ask the people

00:26:32.580 that initially, like, worked on,

00:26:36.280 like, the containment of nuclear weapons,

00:26:38.920 they assumed that either there will be one country

00:26:41.380 that has all of the nuclear weapons

00:26:43.420 and no one else has them

00:26:45.340 and they control everything,

00:26:46.520 or every country has it

00:26:47.920 and it's going to be a super precarious situation.

00:26:50.240 Ended up being that treaties kind of worked.

00:26:52.540 Non-proliferation treaty allowed for,

00:26:54.180 in the end, only, like, nine countries

00:26:55.820 that currently have nuclear weapons.

00:26:58.960 And then we also had a reduction of the weapons

00:27:03.360 after we built, like, 60,000 of them.

00:27:07.100 There are now, like, 13,000.

00:27:08.860 So treaties can work sometimes

00:27:11.720 to reduce the economic incentives.

00:27:13.820 We also succeeded with the Montreal Protocol.

00:27:16.740 Like, we had the ozone layer being depleted

00:27:19.260 and hurt much more

00:27:22.260 by the use of CFCs, Hairspray, etc., back then.

00:27:25.740 And then countries came together.

00:27:27.320 And actually, DuPont, or DuPoint,

00:27:30.000 the company, the large one,

00:27:31.440 they were, like, fighting against it.

00:27:32.680 It's like, no, it can't be that bad.

00:27:35.120 The science is bullshit.

00:27:37.420 CFCs are fine.

00:27:38.880 And then, like, the CEO announced it.

00:27:41.280 And then a few weeks later,

00:27:42.280 they actually saw that

00:27:43.900 they would have a commercial opportunity

00:27:45.800 for distributing their product

00:27:48.140 with a different chemical base.

00:27:50.860 And they're like, oh, you know what?

00:27:51.920 We're actually on board now.

00:27:53.320 It's fine.

00:27:53.820 We'll just pursue the alternative solution.

00:27:55.700 So the thing I'm saying is

00:27:58.300 it's not that we shouldn't build

00:28:01.600 super powerful AI.

00:28:03.160 Quite the opposite.

00:28:03.940 I want it.

00:28:04.440 It will solve so many problems.

00:28:05.760 It's going to be great.

00:28:07.120 But I feel like

00:28:09.100 if we just go with as fast as possible,

00:28:12.620 we won't take, like,

00:28:13.900 sometimes, ideally,

00:28:14.860 you would take the trade-off

00:28:15.800 towards going a little slower,

00:28:18.620 making sure that it's not deceiving us,

00:28:21.240 making sure that it can't be used

00:28:23.180 to generate, like, new pathogens at will

00:28:25.900 and only then release it.

00:28:28.340 So, like, I want us to actually

00:28:29.520 be able to take sensible

00:28:31.160 and wise trade-offs

00:28:32.180 between safety and speed

00:28:34.220 rather than just go for speed.

00:28:36.220 But wouldn't the previous subject

00:28:39.220 that we've spoken about,

00:28:40.360 which is gain-of-function research,

00:28:42.360 particularly in a country like China,

00:28:44.900 be the logical pushback

00:28:46.840 to what you're saying?

00:28:48.660 In the sense that, I mean,

00:28:50.240 they are presently doing it,

00:28:53.180 we are doing gain-of-function research.

00:28:54.440 Do you mean that it could be used for it?

00:28:56.060 No, I mean, what I'm saying is,

00:28:57.880 it's, you know,

00:28:58.540 we can agree to do this

00:28:59.980 in the US or the UK,

00:29:02.320 but I highly doubt

00:29:04.140 the Chinese or the Russians would do it

00:29:06.000 because it's not in their interest.

00:29:07.980 Well, it actually is.

00:29:09.340 So people assume that China

00:29:10.540 is going to be, like,

00:29:11.960 just super gung-ho.

00:29:13.320 But so far, at least,

00:29:15.020 China has released more regulation

00:29:16.560 around AI than the US has.

00:29:18.240 And they've been more careful about it.

00:29:22.080 And it makes sense

00:29:22.940 if you put it under the perspective of

00:29:26.040 you're developing something

00:29:28.000 that's more intelligent,

00:29:28.920 that will be super powerful.

00:29:32.100 China really likes to be in control

00:29:34.140 of their own country.

00:29:35.180 Like, the CCP really wants

00:29:36.360 to be in control.

00:29:37.580 That's very mildly stated, yeah.

00:29:39.440 They don't want anyone else

00:29:40.320 or anything else to be in control

00:29:41.560 outside of them.

00:29:42.260 And if it is possible,

00:29:46.680 which it currently is unknown,

00:29:48.240 whether it can be contained or not,

00:29:50.600 whether it can be,

00:29:51.240 when the AI can be controlled,

00:29:52.760 if it is therefore possible

00:29:53.680 that it does end up

00:29:55.380 just, like, running away with it,

00:29:56.700 or an individual actor

00:29:57.800 who has the, like,

00:29:59.240 super powerful AI

00:30:00.020 becomes extremely powerful,

00:30:01.540 the CCP doesn't want that.

00:30:02.880 Like, they want it even less

00:30:04.200 than in the US,

00:30:06.820 like, the government wanted,

00:30:07.840 which is much more happy

00:30:09.300 with letting capitalism

00:30:10.420 kind of, like, choose

00:30:11.260 who develops what, right?

00:30:13.500 Like, China is a much more

00:30:14.620 ideological actor than the US.

00:30:16.560 The US is more of a capitalist,

00:30:19.520 like, individual freedoms,

00:30:20.920 individual pursuits actor,

00:30:22.980 which I'm much more in favor of.

00:30:24.660 But I think that China

00:30:25.940 will take economic cost

00:30:27.400 for ideological benefits.

00:30:30.000 Okay, so the Chinese

00:30:31.140 are more interested in control.

00:30:32.200 Look, I think one of the things

00:30:33.840 that happens when we talk about AI

00:30:35.400 is we naturally focus more

00:30:36.920 on the negatives.

00:30:37.560 But what are the positives

00:30:39.000 of this new technology?

00:30:40.280 What are the things

00:30:41.100 that it's going to help us with

00:30:42.340 moving forward?

00:30:44.440 I mean, tons, right?

00:30:46.120 It could be, it's in,

00:30:48.120 it's increased intelligence,

00:30:49.780 increased way to solving problems.

00:30:51.440 So, like, you could really apply it

00:30:52.560 to about anything.

00:30:53.340 And we see some of it.

00:30:54.720 You have, like, radiologists

00:30:55.800 who now can use, like,

00:30:57.220 image recognition

00:30:57.860 to better identify

00:30:59.160 cancerous cells on skin and such.

00:31:04.500 Beauty filters, I don't know

00:31:05.700 if you like them or not,

00:31:06.520 but you can look cool

00:31:07.400 while you're also online.

00:31:10.160 And, yeah, just, like,

00:31:11.860 super broadly,

00:31:12.600 like, you have efficiency improvements

00:31:14.480 in, like,

00:31:16.920 in how you're cooling

00:31:18.660 servers and data centers.

00:31:21.380 It is currently being applied

00:31:23.240 to solve a bunch of things

00:31:24.600 in biology on the positive side

00:31:26.040 with, like, developing new therapeutics

00:31:29.540 that will come out,

00:31:30.620 that is coming out of DeepMind.

00:31:33.100 It's a new company

00:31:34.740 that they've built

00:31:35.560 that is basically trying to solve

00:31:37.060 and build, like,

00:31:37.580 many more medicines in biology.

00:31:40.080 Let's explore that a little bit.

00:31:41.180 They also worked on nuclear fusion

00:31:42.960 and, like,

00:31:43.460 because it's really hard

00:31:44.160 to contain, kind of,

00:31:45.160 like, the plasma

00:31:45.720 that's inside of the tokamak

00:31:47.480 and they're, like,

00:31:48.820 just helping with that.

00:31:49.940 You could see a lot of stuff

00:31:51.420 being unlocked

00:31:52.080 and I think it's only the beginning.

00:31:53.380 Like, we can't, yeah,

00:31:55.340 it could be applied

00:31:56.220 to so many things

00:31:56.920 that we're not applying it to yet.

00:31:58.320 Yeah, sorry to interrupt you there,

00:31:59.360 I wanted to talk about

00:32:00.460 the medical side of it

00:32:01.900 because that, to me,

00:32:02.640 is really interesting.

00:32:03.760 You were talking about therapeutics.

00:32:05.160 Like, what are we talking about here?

00:32:06.600 How could AI be used

00:32:07.900 with regard to therapeutics

00:32:09.720 to help human life?

00:32:11.860 Well, one concrete thing

00:32:12.780 that happened was

00:32:13.580 that AlphaFold came out

00:32:15.780 which solved, like,

00:32:18.160 the problem of understanding

00:32:20.480 which amino acids become

00:32:22.920 which proteins

00:32:23.680 and then, like,

00:32:24.460 what proteins you, hence,

00:32:26.940 would be able to build yourself

00:32:28.240 if you wanted to

00:32:29.080 or, rather,

00:32:30.160 how you would build the proteins.

00:32:32.020 I'm not that deep

00:32:33.040 in the biological side of it

00:32:34.900 but, as I understand it,

00:32:37.600 is that we're now at a place

00:32:40.340 where we can much easier

00:32:41.220 design these proteins

00:32:42.100 and, in the end,

00:32:43.620 you can use, like,

00:32:44.480 proteins are kind of, like,

00:32:45.580 little mechanistic,

00:32:47.000 basically, like,

00:32:48.360 things in your body

00:32:49.680 that run around

00:32:50.280 and, like, move things around,

00:32:51.440 et cetera.

00:32:52.400 And, with that,

00:32:53.760 we can unlock, like,

00:32:54.720 many new therapies

00:32:56.760 that we have beforehand

00:32:58.000 not been able to do.

00:32:59.340 That's so interesting.

00:33:00.720 And particularly, you know,

00:33:02.080 because people talk

00:33:02.780 about climate change a lot.

00:33:04.040 They talk a lot about the fact

00:33:05.820 that we're producing

00:33:06.740 too much carbon,

00:33:07.760 carbon neutral.

00:33:08.640 If AI could be used

00:33:10.160 to create a better,

00:33:12.260 safer form of nuclear energy

00:33:14.020 that is more efficient

00:33:14.960 and provides less waste,

00:33:17.220 I mean, that could...

00:33:18.760 That'd be great.

00:33:19.740 I mean, on the nuclear energy front,

00:33:21.040 I think, like,

00:33:22.060 fission seems to be fine.

00:33:23.080 We had a lot of regulation

00:33:23.820 around it,

00:33:24.420 but, like,

00:33:25.240 I think we could already

00:33:27.180 just have much more

00:33:27.800 nuclear energy.

00:33:28.800 But it seems that it was

00:33:30.040 so far really, really hard

00:33:31.340 to create nuclear fusion.

00:33:34.400 And it mustn't be

00:33:35.460 that AI is necessary for it.

00:33:36.700 It might just be

00:33:37.100 an engineering problem

00:33:37.760 that can be solved without it.

00:33:39.300 But either case,

00:33:41.320 they did use ML techniques

00:33:44.100 to kind of, like,

00:33:44.860 contain the plasma.

00:33:46.000 But I think that outside of it,

00:33:48.700 like, what can more

00:33:50.720 intelligence be used for?

00:33:51.820 It's really just about anything.

00:33:55.420 Can current AI models

00:33:57.540 be used for anything?

00:33:58.540 No.

00:33:58.960 Like, GPT-4 can't help you

00:34:00.940 with, like,

00:34:01.620 figuring out fundamental physics

00:34:03.560 to then figure out engineering yet.

00:34:05.900 But in a few years,

00:34:07.520 you could see that being the case.

00:34:10.500 And culturally and politically,

00:34:14.040 what's interesting to me is

00:34:15.340 you talked about the example

00:34:17.600 of ChatGPT

00:34:18.580 where it gives people

00:34:20.260 different answers

00:34:20.900 based on their political views,

00:34:23.560 et cetera.

00:34:24.540 One of the things

00:34:25.240 that the entire conversation

00:34:26.840 brings up is

00:34:27.760 whether truth exists, really, right?

00:34:30.500 Because

00:34:30.900 in order,

00:34:34.000 you were saying,

00:34:35.300 you know,

00:34:35.540 the model should optimize for truth.

00:34:37.500 But human beings

00:34:38.440 haven't worked out

00:34:39.260 a way to agree

00:34:40.000 on what the truth is.

00:34:40.780 And frankly,

00:34:41.340 increasingly,

00:34:42.340 we're unable to agree

00:34:43.560 on what the truth is.

00:34:45.060 How do you think

00:34:45.980 the creation of AI?

00:34:48.200 Because I went on ChatGPT

00:34:50.120 when it was released

00:34:51.080 a few months ago.

00:34:53.160 And I played around.

00:34:54.380 I took the hot button issues

00:34:55.860 in our society today

00:34:57.000 and I played around with

00:34:58.080 what is a woman?

00:34:59.400 For example,

00:35:00.240 a question that seems to

00:35:01.420 stump a lot of people nowadays.

00:35:03.580 Did it stump it?

00:35:04.240 Well,

00:35:06.600 it sort of gave me

00:35:08.000 some kind of like

00:35:08.920 politically correct

00:35:10.060 and I went back and forth

00:35:12.040 with it.

00:35:12.280 It was a longer answer

00:35:13.060 than you expected,

00:35:13.840 I imagine.

00:35:14.200 It wasn't just the fact

00:35:15.060 that it was a longer answer.

00:35:16.400 Every answer ended

00:35:17.620 with the phrase,

00:35:18.420 but we must remember

00:35:19.180 to respect all people's

00:35:20.500 differences and blah,

00:35:21.360 it was a very kind of political,

00:35:23.020 it wasn't a truth conversation.

00:35:24.780 It was about

00:35:25.560 a cultural sensitivity

00:35:26.820 and all of that.

00:35:30.200 And I wonder

00:35:30.760 if and when

00:35:32.580 we move to

00:35:33.620 these AI systems

00:35:35.260 being the new Google,

00:35:36.460 essentially,

00:35:36.900 the new source of truth, right?

00:35:38.660 Because to most people nowadays,

00:35:40.160 the truth is whatever you Google.

00:35:41.880 If you put something in Google

00:35:42.780 and that's what it says,

00:35:43.820 that's the truth.

00:35:45.220 I mean, moronic in my opinion,

00:35:46.620 but that's how people are, right?

00:35:48.220 People want speed

00:35:49.100 in their truth as well.

00:35:50.000 Yes, exactly.

00:35:50.720 They'll sacrifice

00:35:51.400 a little bit of truth

00:35:52.280 for quite a bit of speed.

00:35:53.380 Exactly.

00:35:53.680 Or belonging, actually.

00:35:54.780 And tribal allegiance

00:35:56.460 and all of that, yeah.

00:35:58.100 So if our source of information

00:36:00.960 becomes this thing

00:36:03.120 that's pre-programmed

00:36:04.380 with the biases

00:36:05.060 of the people

00:36:05.660 who developed it

00:36:06.500 and who designed it,

00:36:09.100 isn't that,

00:36:10.180 you know,

00:36:10.580 people will talk about

00:36:11.900 a tyranny of the minority

00:36:13.160 in various political

00:36:14.180 and cultural contexts

00:36:15.100 where a small group of people

00:36:16.580 get to impose

00:36:17.320 their worldview

00:36:17.900 on everybody else.

00:36:19.080 That seems to me

00:36:20.080 like that on steroids.

00:36:22.720 I mean, it could be.

00:36:24.780 I think for what it's worth,

00:36:27.840 the current AI companies

00:36:30.160 are trying to address that

00:36:31.520 by including

00:36:32.460 like various viewpoints

00:36:33.600 outside of the ones

00:36:34.440 that they agree with

00:36:35.880 in their testing

00:36:37.380 of the language models

00:36:39.080 that they designed.

00:36:41.160 And I think

00:36:41.840 in the case of

00:36:42.860 ChatGPT, for example,

00:36:43.880 like people put the questions

00:36:45.460 in front of it

00:36:45.960 that are like

00:36:46.420 to figure out

00:36:46.980 your political compass position

00:36:48.320 and it started out

00:36:49.980 fairly liberal left

00:36:51.600 but then over time

00:36:52.660 it actually moved

00:36:53.240 a little bit more

00:36:53.760 to the center

00:36:54.240 when people did it again

00:36:55.240 two, three months later.

00:36:57.940 Because I mean,

00:36:59.240 you could imagine

00:37:00.080 there being also

00:37:01.820 just various language models

00:37:04.220 that tell you

00:37:04.800 what you want to hear.

00:37:06.540 I'm not sure

00:37:07.360 everything needs

00:37:07.980 to optimize for truth.

00:37:09.060 Like for me personally,

00:37:09.860 that's the most interesting thing

00:37:10.880 but it's not the thing

00:37:11.640 that everyone seeks out

00:37:12.860 the most, right?

00:37:13.460 Like people read

00:37:14.180 news publications

00:37:16.260 from the far left

00:37:17.480 and the far right

00:37:18.260 not because they're

00:37:19.140 looking for truth,

00:37:19.860 they're looking for

00:37:20.500 something else.

00:37:21.880 And I think like

00:37:22.420 you could have language models

00:37:23.220 that do that as well.

00:37:24.780 But yeah,

00:37:25.560 will we then have one

00:37:26.820 that imposes

00:37:28.380 like their own ideology

00:37:30.020 the most?

00:37:32.020 I mean,

00:37:32.280 it certainly

00:37:32.700 could be used.

00:37:35.580 I think there is,

00:37:36.640 I think as long

00:37:37.420 like as the marketplace

00:37:38.580 of ideas

00:37:39.240 like it's kind of around,

00:37:40.300 it's going to be hard.

00:37:42.560 I'm not that worried

00:37:43.600 about say

00:37:46.300 OpenAI gaining

00:37:47.620 so much power

00:37:48.840 they now change

00:37:50.900 the entire like

00:37:52.080 landscape of things

00:37:53.080 that are being talked about.

00:37:55.380 I'm a bit more worried

00:37:56.660 about a lot of people

00:37:58.700 individually

00:37:59.260 like getting extra power

00:38:00.860 to edit Wikipedia

00:38:02.040 into like biased ways

00:38:04.200 or like making up

00:38:05.380 fake papers

00:38:06.140 that justify

00:38:07.380 their fake articles.

00:38:08.820 All of that I think

00:38:09.880 will unfortunately happen

00:38:10.860 and that will kind of

00:38:12.500 erode general trust

00:38:14.340 into the internet

00:38:16.100 potentially as well.

00:38:17.140 I don't know yet

00:38:17.540 what we're going to do

00:38:18.100 about that.

00:38:18.640 That seems pretty

00:38:19.260 pretty bad to me.

00:38:20.340 I mean already Google

00:38:21.060 has become a bit worse

00:38:22.060 at just,

00:38:23.280 I don't know if you've,

00:38:24.020 I imagine you've used it

00:38:25.560 recently,

00:38:26.540 but it's like

00:38:27.540 the answers that you get

00:38:28.840 yeah,

00:38:30.980 are just not as useful

00:38:32.460 I find anymore

00:38:33.220 as they were

00:38:33.780 like within

00:38:34.460 five years ago.

00:38:35.980 Really?

00:38:36.880 Yeah,

00:38:37.220 it's a lot

00:38:37.820 it seems like

00:38:38.920 Quora has become

00:38:40.040 very good at like

00:38:40.980 optimizing

00:38:43.060 their spot

00:38:44.360 in the search

00:38:45.220 so then like

00:38:46.760 you're often,

00:38:48.080 if you ask

00:38:48.520 a question like

00:38:49.680 that is something

00:38:51.300 like

00:38:51.700 what would happen

00:38:52.840 if a ball

00:38:54.320 was filled

00:38:55.060 with helium

00:38:55.680 like how long

00:38:57.480 would it take

00:38:58.040 like how much

00:38:58.660 would you need to fill

00:38:59.240 so that like

00:38:59.860 it could lift a human

00:39:00.640 or something like that

00:39:01.420 usually that type of stuff

00:39:02.660 is something that Quora

00:39:03.600 maybe had already

00:39:04.760 someone asked

00:39:05.780 and you get an answer

00:39:06.680 from there

00:39:07.060 but the quality of answers

00:39:08.160 there isn't that high

00:39:09.420 because Quora

00:39:09.960 is just not used enough

00:39:11.020 so,

00:39:13.200 and now

00:39:13.660 that actually

00:39:16.480 more and more

00:39:17.340 gets language model

00:39:20.060 answers within Quora

00:39:21.200 and then

00:39:21.700 those are being put

00:39:23.160 into Google

00:39:23.660 even as one of their

00:39:24.460 first answers

00:39:25.120 that they display

00:39:26.180 you know

00:39:26.740 where they don't just

00:39:27.360 show you the link

00:39:27.900 but the kind of top section

00:39:28.960 so the top section

00:39:30.080 has become lower quality

00:39:31.120 in my view

00:39:31.580 by quite a bit

00:39:32.700 that's very interesting

00:39:33.520 but coming back

00:39:34.340 to your political point

00:39:35.200 about it moving

00:39:35.860 to the center

00:39:36.460 the thing

00:39:36.840 like I'm somewhere

00:39:37.740 in the center

00:39:38.760 whatever that means

00:39:39.700 I mean

00:39:39.960 you and I

00:39:40.560 depends on who you ask

00:39:41.400 I imagine

00:39:41.840 right

00:39:42.420 and also

00:39:43.480 you and I

00:39:44.040 could both be

00:39:44.620 in the center

00:39:45.100 and have completely

00:39:45.960 polar opposite

00:39:47.380 viewpoints

00:39:48.220 on the same issues

00:39:49.040 right

00:39:49.580 some people

00:39:50.280 in the center

00:39:50.780 because they're

00:39:51.360 to the right

00:39:51.820 on this issue

00:39:52.440 and to the left

00:39:52.900 on that issue

00:39:53.480 and you could be

00:39:54.340 to the right

00:39:55.120 on this issue

00:39:55.640 and to the left

00:39:56.100 on this issue

00:39:56.580 you could swap

00:39:57.100 right

00:39:57.420 in fact if you

00:39:58.500 didn't have that

00:39:59.080 you wouldn't be a

00:39:59.700 independent thinker

00:40:01.020 correct

00:40:01.800 and on top of that

00:40:04.180 some people might argue

00:40:06.260 that the center

00:40:06.840 isn't the most

00:40:07.480 representative point of view

00:40:08.640 because that's where

00:40:10.080 the fewest people are

00:40:11.220 most people are

00:40:12.020 in one of the two

00:40:12.840 big tribes

00:40:13.640 and if you're on the right

00:40:14.820 you go

00:40:15.280 you know

00:40:15.980 this is the truth

00:40:17.180 and all these people

00:40:18.020 are idiots

00:40:18.380 and if you're on the left

00:40:19.080 you go

00:40:19.360 this is the truth

00:40:20.100 and all these people

00:40:20.580 are idiots

00:40:20.880 and both tribes

00:40:22.540 look at the people

00:40:23.140 in the middle

00:40:23.580 and go

00:40:24.020 these are the real idiots

00:40:25.100 because they don't agree

00:40:27.140 with either of us

00:40:27.840 or they're spineless

00:40:28.760 or they're spineless

00:40:29.640 or whatever

00:40:30.040 so I guess

00:40:31.600 what I'm saying is

00:40:32.440 it brings up

00:40:34.160 the very philosophical

00:40:35.620 notion of truth

00:40:36.920 in and of itself

00:40:38.020 right

00:40:38.620 when you're

00:40:39.500 when you're giving

00:40:41.600 these language models

00:40:43.900 or these AI

00:40:44.780 systems

00:40:45.860 the power

00:40:47.160 to decide

00:40:48.000 what the truth is

00:40:49.160 they are inevitably

00:40:51.500 going to dissatisfy

00:40:53.040 probably the majority

00:40:54.220 of the population

00:40:54.940 yeah

00:40:56.080 I mean

00:40:56.480 what

00:40:56.800 philosophers

00:40:58.420 have been talking

00:40:59.000 about truth

00:40:59.460 a long time

00:41:00.200 I think some

00:41:02.680 also come to

00:41:03.460 the notion

00:41:04.020 that saying

00:41:04.860 that something

00:41:05.280 is true

00:41:05.760 is just

00:41:07.000 meaningless

00:41:07.460 you can just

00:41:07.900 throw it out

00:41:08.420 it's like

00:41:09.080 it either is

00:41:09.720 or it isn't

00:41:10.700 and adding

00:41:11.880 this is true

00:41:12.440 doesn't actually

00:41:13.040 add that much

00:41:13.660 information context

00:41:14.480 because why are you

00:41:15.100 saying it in the first place

00:41:15.960 is part of it

00:41:17.300 but

00:41:17.520 so

00:41:19.160 yeah

00:41:19.920 like

00:41:20.220 if we look at

00:41:21.600 facts

00:41:22.460 like actual events

00:41:23.400 that happened

00:41:23.900 then we can

00:41:24.440 talk about truth

00:41:25.340 I suppose

00:41:25.840 right

00:41:26.120 like

00:41:26.360 there

00:41:26.780 it seems like

00:41:29.140 I would assume

00:41:30.380 that many language models

00:41:31.260 will just

00:41:32.100 carry the verifiable

00:41:33.460 facts of the past

00:41:34.700 at least I hope so

00:41:35.820 but then

00:41:36.860 what else are you

00:41:37.980 pointing that could be

00:41:38.840 true

00:41:39.280 about

00:41:40.460 like

00:41:40.820 not about

00:41:41.240 facts of the past

00:41:42.440 but

00:41:42.880 well we don't even

00:41:45.140 agree about the past

00:41:46.080 yeah

00:41:46.500 like I was having

00:41:47.580 a conversation

00:41:48.160 with somebody

00:41:48.680 yesterday

00:41:49.120 about McCarthyism

00:41:50.120 and the conventional

00:41:51.260 narrative

00:41:51.720 on McCarthyism

00:41:52.920 is according to

00:41:54.460 this person

00:41:54.980 and the things

00:41:55.520 that they quoted

00:41:56.160 and the books

00:41:56.600 that they've read

00:41:57.220 complete nonsense

00:41:58.620 well so then

00:41:59.620 there are like

00:42:00.140 yeah

00:42:00.380 interpretations

00:42:01.180 of the events

00:42:02.320 are often like

00:42:03.040 very

00:42:03.300 a lot of disagreement

00:42:04.520 and between historians

00:42:05.500 but

00:42:05.900 something like

00:42:06.920 this happened

00:42:07.560 on that day

00:42:08.360 at least

00:42:09.180 often we can agree

00:42:10.240 on that

00:42:10.600 sometimes we don't

00:42:11.240 even agree on that

00:42:12.260 like

00:42:13.020 with JFK

00:42:14.300 for example

00:42:14.840 the shooting

00:42:15.360 it's like

00:42:15.780 we don't agree on

00:42:17.000 what

00:42:18.040 why it happened

00:42:18.940 but at least

00:42:19.460 people agree

00:42:20.000 that there was

00:42:20.660 a bullet

00:42:20.940 that flew

00:42:21.340 from there

00:42:21.860 that ended

00:42:22.700 in him here

00:42:23.640 well actually

00:42:24.940 not even

00:42:25.500 no you're right

00:42:27.640 not even fully

00:42:28.520 there are

00:42:28.840 multiple theories

00:42:29.660 this is what I'm saying

00:42:30.760 so then I suppose

00:42:31.880 in those cases

00:42:32.840 yeah what do you

00:42:33.360 want to do

00:42:33.700 you want to like

00:42:34.320 give

00:42:35.600 the

00:42:36.600 most agreed upon

00:42:38.360 notion

00:42:40.540 plus

00:42:41.200 the counterpoints

00:42:42.780 that some people

00:42:43.380 are making as well

00:42:44.200 would be interesting

00:42:44.740 but a thousand years

00:42:45.440 not a thousand

00:42:46.400 a hundred years ago

00:42:47.240 the most agreed upon

00:42:48.220 viewpoint would have been

00:42:49.440 that black people

00:42:50.060 are inferior to white people

00:42:51.180 yeah but

00:42:52.100 that wasn't

00:42:52.840 like an event

00:42:54.080 as we just like

00:42:55.040 related to before

00:42:55.960 okay

00:42:56.240 that was like

00:42:56.920 that is a

00:42:57.800 I think moral truth

00:42:59.360 is a totally

00:42:59.980 separate conversation

00:43:01.020 where like

00:43:02.300 is

00:43:03.400 like

00:43:03.840 should we

00:43:05.660 treat

00:43:06.700 give animals

00:43:08.020 the same rights

00:43:08.620 as humans

00:43:09.120 like right now

00:43:10.160 that's not what I'm saying

00:43:10.660 I'm talking about

00:43:11.300 the scientific consensus

00:43:12.680 was that some races

00:43:13.820 are inferior to

00:43:14.620 others

00:43:14.940 right

00:43:15.560 so if we had

00:43:16.520 these language models

00:43:17.400 then

00:43:17.760 you would have had

00:43:18.780 a language model

00:43:19.460 that would have

00:43:19.940 advanced that

00:43:20.780 as the scientific truth

00:43:22.140 yeah I'm not sure

00:43:23.820 because I feel like

00:43:24.800 maybe

00:43:26.100 it may be the case

00:43:28.620 that those things

00:43:29.800 are kind of

00:43:30.820 conditional on each other

00:43:32.340 you only develop

00:43:33.120 language models

00:43:33.720 at a time

00:43:34.260 where your science

00:43:35.140 is good enough

00:43:35.720 that you throw out

00:43:36.600 a bunch of bullshit

00:43:37.300 things on the way

00:43:38.120 in the sense that

00:43:39.560 like at the time

00:43:41.120 we were able to make

00:43:42.120 a lot of

00:43:42.660 non-scientific claims

00:43:43.940 about the world

00:43:44.580 because we didn't

00:43:45.420 even understand

00:43:46.080 how science

00:43:46.580 really worked

00:43:47.160 like we had

00:43:47.680 very little

00:43:48.140 understanding of

00:43:48.820 we had less

00:43:50.500 understanding of

00:43:51.160 physics

00:43:51.420 of medicine

00:43:52.100 etc

00:43:52.500 so like you could

00:43:53.220 just make

00:43:53.720 these claims

00:43:54.300 we had

00:43:55.580 worse methods

00:43:57.420 in terms of

00:43:59.100 like randomized

00:44:00.620 controlled trials

00:44:01.380 that you could run

00:44:02.100 on like medicine

00:44:03.100 or in some

00:44:03.740 psychological experiments

00:44:04.740 even though those

00:44:05.180 are partially being

00:44:05.800 thrown out now

00:44:06.360 as well

00:44:06.680 I feel like

00:44:07.900 you don't get

00:44:09.180 like what would

00:44:09.800 be an example

00:44:10.380 of stuff

00:44:11.960 that you get

00:44:13.000 away with

00:44:13.400 saying right now

00:44:13.980 that is accepted

00:44:14.560 as truth

00:44:15.080 something like

00:44:16.280 children should

00:44:17.160 go to school

00:44:17.760 it's like

00:44:18.220 it's good for them

00:44:18.920 it's like

00:44:19.880 well even that

00:44:20.800 I don't think

00:44:21.680 is in the same

00:44:22.240 category of truth

00:44:23.340 as

00:44:23.900 yeah

00:44:26.100 on the 6th of January

00:44:27.480 some people

00:44:28.360 entered the

00:44:29.080 Capitol building

00:44:29.880 they're of a

00:44:31.580 different flavor

00:44:32.220 and I feel like

00:44:32.820 science has

00:44:33.420 gotten good enough

00:44:35.280 where we can

00:44:37.020 distinguish between

00:44:37.880 these types of

00:44:38.640 truth that are

00:44:39.260 a bit more

00:44:39.880 like statements

00:44:41.140 of opinion

00:44:41.680 or moral truth

00:44:42.700 but the problem

00:44:43.380 is

00:44:43.760 is after the

00:44:44.820 pandemic

00:44:45.220 our faith

00:44:45.980 in the science

00:44:46.660 has been decreased

00:44:47.500 there's a lot

00:44:48.200 of people now

00:44:48.880 who have lost

00:44:49.420 faith in

00:44:50.580 scientists

00:44:51.180 and vaccines

00:44:52.180 because of the

00:44:53.000 fact that

00:44:53.540 what they were

00:44:54.340 told turned out

00:44:55.220 to be untrue

00:44:56.100 and it's

00:44:57.880 yeah

00:44:58.720 it's fair enough

00:44:59.780 because a lot

00:45:00.340 of it had been

00:45:01.120 captured by like

00:45:02.160 people not

00:45:02.740 saying the

00:45:04.160 current understanding

00:45:05.880 at the time

00:45:06.440 but rather

00:45:07.860 just saying

00:45:09.360 the political

00:45:09.800 line that they

00:45:10.380 were made to

00:45:10.820 say

00:45:11.060 yeah so like

00:45:13.720 okay if we look

00:45:14.320 at vaccines

00:45:14.780 for example

00:45:15.360 what is the

00:45:16.020 truth there

00:45:16.600 the truth there

00:45:19.320 that probably

00:45:19.920 is universally

00:45:20.560 agreed is

00:45:21.300 that mRNAs

00:45:22.400 develop these

00:45:23.000 vaccines

00:45:23.500 and these

00:45:24.160 vaccines

00:45:24.660 are more than

00:45:25.760 zero useful

00:45:26.420 that there are

00:45:27.480 some costs

00:45:28.660 to a young

00:45:31.080 person using

00:45:32.000 a vaccine

00:45:32.520 that the

00:45:33.700 risk

00:45:34.100 and then

00:45:34.920 whether the

00:45:35.420 risk benefit

00:45:35.920 comes out

00:45:36.320 positive or

00:45:36.800 not

00:45:36.960 that is like

00:45:37.460 a question

00:45:37.860 of trials

00:45:39.340 and looking

00:45:39.880 at data

00:45:40.320 etc

00:45:40.660 I feel like

00:45:42.220 when people

00:45:42.600 then say

00:45:43.180 no therefore

00:45:44.080 everyone should

00:45:44.920 use them

00:45:45.340 that's kind of

00:45:45.820 again a different

00:45:46.320 category of truth

00:45:47.240 so I would

00:45:47.900 hope that

00:45:48.500 language model

00:45:49.120 end up

00:45:49.640 distinguishing

00:45:50.660 like between

00:45:51.580 consensus

00:45:54.460 viewpoints

00:45:55.160 which is not

00:45:56.380 truth

00:45:56.760 right

00:45:57.160 consensus

00:45:58.800 viewpoints

00:45:59.360 and like

00:46:00.500 hard ground

00:46:01.400 based truth

00:46:02.080 that is

00:46:02.560 verifiable

00:46:03.140 by like

00:46:03.920 looking at

00:46:04.920 what happened

00:46:05.340 in the world

00:46:05.760 because yeah

00:46:06.340 at the time

00:46:06.980 but the problem

00:46:07.720 is also as well

00:46:08.680 Igor

00:46:08.960 is that we're

00:46:09.720 talking about

00:46:10.400 things that are

00:46:11.080 verifiable

00:46:11.640 and that we know

00:46:12.380 to be true

00:46:13.000 and they're

00:46:13.360 backed up by

00:46:13.940 science

00:46:14.420 and not to

00:46:15.440 return to

00:46:15.900 the trans

00:46:16.340 debate

00:46:16.600 because it

00:46:16.920 seems that's

00:46:17.420 where we go

00:46:17.820 all the time

00:46:18.340 but you now

00:46:19.680 have certain

00:46:20.220 scientists

00:46:20.680 saying well

00:46:21.260 you know

00:46:21.520 sex is a

00:46:22.200 spectrum

00:46:22.500 and you go

00:46:22.940 what

00:46:23.380 so it now

00:46:25.640 seems that

00:46:26.300 consensus

00:46:26.820 is affecting

00:46:27.660 things that

00:46:28.200 we have all

00:46:28.700 known to be

00:46:29.220 true

00:46:29.540 yeah

00:46:29.880 I mean

00:46:30.500 that is an

00:46:30.900 interesting

00:46:31.160 question

00:46:31.460 like I

00:46:31.900 think there

00:46:32.240 is a

00:46:33.000 very

00:46:33.520 probably a

00:46:34.460 pretty important

00:46:35.000 line between

00:46:35.660 like just

00:46:36.480 what consensus

00:46:37.840 is and what

00:46:38.580 is like

00:46:39.080 hard true

00:46:39.800 consensus

00:46:42.480 obviously shifts

00:46:43.700 continuously

00:46:44.280 like and

00:46:45.980 consensus will

00:46:46.700 shift yet again

00:46:47.420 in the future

00:46:48.080 forward we

00:46:48.520 don't that's

00:46:50.120 why it's like

00:46:50.740 so like

00:46:53.680 hubristic of

00:46:54.520 people when

00:46:54.920 they want to

00:46:55.500 just enshrine

00:46:56.600 some of the

00:46:57.060 values that

00:46:57.820 are like

00:46:58.320 currently known

00:46:59.100 and like kind

00:46:59.680 of reduce

00:47:00.280 speech or

00:47:00.980 allowing for

00:47:02.800 a conversation

00:47:03.300 to happen about

00:47:03.900 some of these

00:47:04.340 things because

00:47:04.780 like guess

00:47:05.580 what consensus

00:47:06.340 viewpoint might

00:47:07.080 be having to

00:47:08.200 shift still

00:47:08.700 but yeah I

00:47:11.740 think that

00:47:12.580 unfortunately

00:47:13.180 probably

00:47:13.840 the current

00:47:15.480 way how we

00:47:16.040 design the

00:47:16.660 language models

00:47:17.180 it is much

00:47:17.800 more designed

00:47:18.300 actually about

00:47:18.860 fulfilling and

00:47:19.880 like hitting on

00:47:20.800 describing consensus

00:47:23.480 viewpoints as

00:47:24.220 true

00:47:24.420 so that could be

00:47:25.520 pretty bad

00:47:25.980 I agree

00:47:26.980 like where you

00:47:27.780 would then

00:47:28.360 have these

00:47:30.400 responses that

00:47:31.200 like chat GPT

00:47:32.360 for example gave

00:47:33.140 that you checked

00:47:34.040 on woman

00:47:34.580 it's like

00:47:35.060 well it probably

00:47:36.980 would have said

00:47:37.440 something else

00:47:37.960 five years ago

00:47:38.700 and it might

00:47:40.260 say something

00:47:40.780 else in five

00:47:41.380 years again

00:47:42.000 and if it

00:47:42.940 each time

00:47:43.520 just pretends

00:47:44.280 like that

00:47:44.780 is the

00:47:45.380 hard truth

00:47:46.160 that is

00:47:47.680 problematic

00:47:48.120 to a

00:47:48.580 different way

00:47:49.300 than if

00:47:51.120 you google

00:47:51.840 and google

00:47:53.000 shows you

00:47:53.740 a list

00:47:54.540 of like

00:47:55.360 15 links

00:47:56.200 and these

00:47:57.600 kind of have

00:47:58.260 different

00:47:58.640 viewpoints

00:47:59.160 gives you

00:48:00.060 much more

00:48:00.680 feel to

00:48:02.080 hey this is

00:48:02.660 an open

00:48:02.960 discussion

00:48:03.360 right now

00:48:03.900 rather than

00:48:04.980 here is

00:48:05.900 the line

00:48:06.520 because the

00:48:07.520 problem is

00:48:08.000 as well

00:48:08.360 is that

00:48:08.680 a lot

00:48:09.000 of these

00:48:09.360 companies

00:48:09.700 look

00:48:09.960 every industry

00:48:10.920 is dominated

00:48:12.020 one way

00:48:12.540 or another

00:48:12.840 by a certain

00:48:13.420 type of people

00:48:14.200 because they

00:48:15.200 have certain

00:48:15.640 aptitudes

00:48:16.140 certain ways

00:48:16.740 of thinking

00:48:17.220 and those

00:48:18.420 types of

00:48:18.880 people

00:48:19.260 by and large

00:48:20.360 tend to have

00:48:20.980 similar

00:48:21.540 political

00:48:21.980 viewpoints

00:48:22.640 so the

00:48:23.300 tech industry

00:48:23.920 tends to be

00:48:24.480 dominated

00:48:24.960 and run

00:48:25.820 by people

00:48:26.340 who are

00:48:26.820 you know

00:48:27.040 what we'd

00:48:27.440 call like

00:48:27.900 liberals

00:48:28.600 or slightly

00:48:29.200 to the left

00:48:29.780 or whatever

00:48:30.160 else

00:48:30.560 and therefore

00:48:31.460 they're going

00:48:31.820 to produce

00:48:32.240 models

00:48:32.660 which reflect

00:48:33.400 them

00:48:33.840 because

00:48:34.700 that's just

00:48:36.020 how human

00:48:36.460 beings are

00:48:37.100 I mean

00:48:37.620 they will

00:48:38.000 it would be

00:48:39.240 surprising

00:48:39.680 if there was

00:48:40.480 zero bias

00:48:41.200 in the models

00:48:42.080 that they

00:48:42.360 produce

00:48:42.760 I would argue

00:48:43.360 that it's

00:48:43.660 impossible

00:48:44.120 to produce

00:48:44.920 zero bias

00:48:45.520 I agree

00:48:46.560 with that

00:48:47.180 but I think

00:48:48.600 you can

00:48:48.900 at least

00:48:49.240 like you can

00:48:50.280 notice when

00:48:50.820 someone is

00:48:51.260 genuinely

00:48:51.620 trying to

00:48:52.380 reduce

00:48:52.720 their bias

00:48:53.360 even though

00:48:54.300 they would

00:48:54.640 have these

00:48:55.120 political

00:48:55.420 viewpoints

00:48:55.920 they could

00:48:56.360 enshrine

00:48:56.840 them into

00:48:57.380 their model

00:48:57.960 they could

00:48:59.520 still attempt

00:49:00.140 not to do

00:49:00.700 that and you

00:49:01.160 would see

00:49:01.540 it in

00:49:02.660 them

00:49:03.320 attempting

00:49:03.720 so like

00:49:04.260 Anthropic

00:49:04.960 for example

00:49:05.580 wrote

00:49:06.080 this proposal

00:49:07.640 of like

00:49:08.100 how to

00:49:08.780 make sure

00:49:09.280 that the

00:49:10.360 language models

00:49:10.980 act well

00:49:11.620 by creating

00:49:13.180 this idea

00:49:13.660 of constitutional

00:49:14.380 AI where

00:49:15.040 they have

00:49:15.680 they're creating

00:49:16.940 basically like

00:49:17.400 a constitution

00:49:18.100 of values

00:49:19.220 that the

00:49:20.060 model should

00:49:21.300 adhere to

00:49:22.200 and there

00:49:23.720 and there

00:49:23.740 you can

00:49:23.980 just look

00:49:24.380 at what

00:49:24.780 are those

00:49:25.380 parts

00:49:26.160 of their

00:49:26.680 AI

00:49:27.100 constitution

00:49:27.660 and

00:49:28.680 those

00:49:29.940 things

00:49:30.460 you can

00:49:31.320 look at

00:49:31.580 and say

00:49:31.900 hey

00:49:32.080 there is

00:49:32.380 a political

00:49:32.700 bias

00:49:33.060 or not

00:49:33.500 but they're

00:49:34.340 actually

00:49:34.600 not that

00:49:35.060 politically

00:49:35.440 biased

00:49:35.900 they're more

00:49:36.560 they just

00:49:37.340 chose

00:49:37.740 like pretty

00:49:38.180 simple things

00:49:39.200 that are

00:49:39.500 like the

00:49:39.980 UN charter

00:49:41.100 of human

00:49:41.560 rights

00:49:41.920 etc

00:49:42.220 like

00:49:42.620 through

00:49:42.840 that

00:49:43.080 but then

00:49:44.380 again

00:49:44.640 the UN

00:49:45.300 can also

00:49:45.700 be captured

00:49:46.200 and I think

00:49:47.020 the WHO

00:49:47.600 was kind

00:49:48.220 of captured

00:49:48.700 etc

00:49:49.060 like you

00:49:49.580 do get

00:49:50.080 I worry

00:49:55.700 about that

00:49:56.060 definitely

00:49:56.280 quite a bit

00:49:56.800 and I think

00:49:57.340 that's why

00:49:57.680 you kind

00:49:58.000 of just

00:49:58.380 have to

00:49:58.900 keep

00:49:59.720 free speech

00:50:00.520 and

00:50:00.840 information

00:50:02.120 exchange

00:50:02.540 no disagreement

00:50:03.420 there at all

00:50:04.100 of course

00:50:04.480 one of the

00:50:05.400 things

00:50:05.700 other potential

00:50:06.460 problems

00:50:06.820 it brings

00:50:07.180 up is

00:50:07.720 you know

00:50:09.560 this well

00:50:10.080 because you're

00:50:10.580 bilingual

00:50:10.980 or trilingual

00:50:11.680 or whatever

00:50:12.120 you are

00:50:12.620 to speak

00:50:13.640 in a different

00:50:14.080 language

00:50:14.380 is to think

00:50:14.980 in a different

00:50:15.420 way

00:50:15.760 most people

00:50:16.580 don't know

00:50:17.020 this

00:50:17.280 but when

00:50:18.320 I speak

00:50:18.700 Russian

00:50:19.020 I'm a

00:50:19.460 different

00:50:19.680 person

00:50:20.060 to the

00:50:20.400 person

00:50:20.640 I am

00:50:21.000 when I

00:50:21.240 speak

00:50:21.440 English

00:50:21.780 does your

00:50:22.100 voice also

00:50:22.640 change

00:50:23.060 like the

00:50:23.580 tone of

00:50:24.020 the voice

00:50:24.340 right

00:50:24.580 it's kind

00:50:25.100 of funny

00:50:25.380 and how

00:50:25.740 you think

00:50:26.180 changes

00:50:26.720 and what

00:50:27.320 you might

00:50:28.880 say

00:50:29.320 you get

00:50:29.680 sad

00:50:30.020 and invades

00:50:32.400 other people's

00:50:33.060 space

00:50:33.280 exactly

00:50:33.800 no respect

00:50:35.460 for anyone

00:50:35.960 but even

00:50:38.200 on a

00:50:38.700 more kind

00:50:40.420 of depressing

00:50:41.260 like for

00:50:41.720 example

00:50:42.080 I am much

00:50:43.360 more comfortable

00:50:43.780 there he goes

00:50:44.240 he goes

00:50:44.520 depressive

00:50:44.920 I'm going

00:50:46.840 full Russian

00:50:47.500 full Russian

00:50:48.280 yes

00:50:48.860 I don't know

00:50:49.240 if you'd

00:50:49.500 agree with

00:50:49.820 me but

00:50:50.140 it's much

00:50:50.880 easier to be

00:50:53.120 homophobic

00:50:53.740 when you're

00:50:54.200 speaking Russian

00:50:54.780 we moved

00:50:57.680 out when I

00:50:58.100 was four

00:50:58.480 so I didn't

00:50:59.020 get to

00:50:59.320 actually grow

00:50:59.860 up in

00:51:00.100 Russia

00:51:00.340 so like

00:51:02.040 my Russian

00:51:02.540 is kind

00:51:02.860 of like

00:51:03.160 kept to

00:51:04.260 these like

00:51:04.740 very simple

00:51:05.780 conversations

00:51:06.400 you weren't

00:51:06.760 I didn't

00:51:08.700 but what I

00:51:10.580 mean is

00:51:11.160 different cultures

00:51:12.200 have different

00:51:12.700 values

00:51:13.200 totally

00:51:13.840 right

00:51:14.140 and so

00:51:15.400 as each

00:51:16.820 big player

00:51:17.960 on the global

00:51:19.200 stage develops

00:51:20.140 its own

00:51:20.600 language

00:51:21.120 model in

00:51:22.180 its own

00:51:22.560 language

00:51:23.220 that is at

00:51:24.080 least to

00:51:24.420 communicate with

00:51:25.160 human beings

00:51:25.820 in its own

00:51:26.540 language

00:51:26.940 those are

00:51:29.000 different

00:51:30.000 perspectives on

00:51:30.920 the world

00:51:31.360 because they

00:51:31.980 reflect a

00:51:32.520 different set

00:51:32.960 of values

00:51:33.480 and that

00:51:35.860 in itself

00:51:36.620 may carry

00:51:38.860 some bias

00:51:39.300 yeah

00:51:39.580 yeah

00:51:40.240 it's

00:51:40.800 interesting

00:51:41.160 so like

00:51:41.620 actually just

00:51:42.480 a few days

00:51:43.060 ago

00:51:43.400 kind of like

00:51:44.640 pretty big

00:51:45.500 breakthrough

00:51:45.960 came out

00:51:46.520 by

00:51:46.900 Anthropic

00:51:47.460 about

00:51:48.180 the question

00:51:50.720 of like

00:51:51.480 how do we

00:51:52.440 look at a

00:51:54.160 model

00:51:54.360 how do we

00:51:54.720 look inside

00:51:55.180 of it

00:51:55.400 and understand

00:51:56.280 what it

00:51:57.080 consists of

00:51:57.640 what are

00:51:57.920 these neurons

00:51:58.460 actually for

00:51:59.120 so like

00:51:59.840 that's

00:52:00.440 called

00:52:01.200 like

00:52:01.580 mechanistic

00:52:03.500 interpretability

00:52:04.300 problem

00:52:04.760 in AI

00:52:05.240 or you could

00:52:06.260 just think of

00:52:06.640 it as like

00:52:06.960 digital neuroscience

00:52:07.880 so like

00:52:08.920 trying to

00:52:09.560 understand

00:52:09.960 what is

00:52:10.640 like

00:52:10.960 which neuron

00:52:11.540 within the

00:52:12.020 model

00:52:12.240 is responsible

00:52:13.340 for which

00:52:13.940 outputs

00:52:14.400 and they

00:52:15.960 discovered

00:52:17.360 like that

00:52:18.240 well it's

00:52:18.640 very hard

00:52:19.040 to like

00:52:19.380 look at

00:52:19.760 a single

00:52:20.080 neuron

00:52:20.360 but they

00:52:20.680 can look

00:52:20.960 at clusters

00:52:21.400 and then

00:52:21.800 they can

00:52:22.080 understand

00:52:22.520 but one

00:52:23.440 thing that

00:52:23.860 also

00:52:24.280 turned out

00:52:25.360 there

00:52:25.640 is that

00:52:27.320 each

00:52:29.500 individual

00:52:30.020 spot

00:52:31.120 does a

00:52:32.420 lot of

00:52:32.660 things

00:52:32.940 so sometimes

00:52:33.640 it's

00:52:34.040 responsible

00:52:34.500 for like

00:52:35.320 the

00:52:36.300 identification

00:52:37.020 of Paris

00:52:37.780 but the

00:52:38.840 same

00:52:39.080 neuron

00:52:39.360 in another

00:52:40.100 instance

00:52:40.540 would like

00:52:41.280 write

00:52:41.920 computer

00:52:42.380 code

00:52:42.720 in another

00:52:43.560 instance

00:52:44.060 it like

00:52:44.520 would be

00:52:44.880 used for

00:52:45.700 like something

00:52:46.720 multimodal

00:52:47.300 maybe where

00:52:48.040 it's like

00:52:48.480 helping to

00:52:49.320 understand what

00:52:49.800 a chair is

00:52:50.380 or like

00:52:50.680 make it

00:52:51.040 visually

00:52:51.420 so what

00:52:53.080 that to

00:52:53.800 me at least

00:52:54.140 implies

00:52:54.520 and I wonder

00:52:54.900 what other

00:52:55.620 people think

00:52:56.060 but it's

00:52:56.360 that it's

00:52:57.420 like the

00:52:58.780 pattern that

00:52:59.800 it finds

00:53:00.640 in the

00:53:01.020 pursuit of

00:53:01.660 like compressing

00:53:02.480 all of the

00:53:02.880 data that

00:53:03.340 you fed

00:53:03.740 in and

00:53:04.520 then like

00:53:04.860 it like

00:53:05.180 compresses it

00:53:05.760 down to

00:53:06.140 like a

00:53:06.420 model

00:53:06.640 right

00:53:06.920 and for

00:53:08.280 that it

00:53:08.620 needs to

00:53:08.900 identify

00:53:09.240 patterns

00:53:09.680 that are

00:53:09.940 useful

00:53:10.220 for this

00:53:10.600 compression

00:53:10.940 so the

00:53:12.600 pattern that

00:53:13.040 it finds

00:53:13.500 are actually

00:53:13.980 kind of

00:53:14.480 even more

00:53:15.820 abstract

00:53:16.300 than what

00:53:17.160 we would

00:53:17.560 do if we

00:53:18.260 were just

00:53:18.700 compressing

00:53:19.240 language down

00:53:19.960 and like

00:53:20.340 noticing some

00:53:20.880 of these

00:53:21.120 biases

00:53:21.460 because like

00:53:22.020 some of

00:53:22.280 these neurons

00:53:22.700 find patterns

00:53:23.420 across

00:53:24.900 language

00:53:25.560 science

00:53:26.160 code

00:53:26.680 and like

00:53:27.120 a lot

00:53:27.360 of things

00:53:27.780 so I

00:53:28.680 wonder

00:53:28.880 whether

00:53:29.140 due to

00:53:29.680 that

00:53:29.980 if you

00:53:30.260 use

00:53:30.460 like

00:53:30.620 very

00:53:30.840 varied

00:53:31.280 data

00:53:31.860 points

00:53:32.240 you could

00:53:33.000 reduce

00:53:33.420 that

00:53:33.780 point

00:53:34.380 that you

00:53:35.300 made

00:53:35.600 that I

00:53:36.000 think

00:53:36.200 is

00:53:36.460 accurate

00:53:37.200 that like

00:53:37.680 each

00:53:37.880 natural

00:53:38.940 language

00:53:39.380 carries

00:53:39.700 some

00:53:39.940 bias

00:53:40.220 but if

00:53:41.240 you combine

00:53:41.700 it with

00:53:42.080 like

00:53:42.300 real world

00:53:42.700 data

00:53:43.000 and everything

00:53:43.560 else

00:53:44.000 probably

00:53:44.760 it reduces

00:53:45.260 down a little

00:53:45.820 bit

00:53:46.160 Igor

00:53:46.940 and moving

00:53:47.640 on now

00:53:48.220 you used

00:53:49.760 to work

00:53:50.200 for Elon

00:53:50.920 Musk

00:53:51.240 what is the

00:53:52.820 purpose

00:53:53.100 what is he

00:53:53.520 trying to do

00:53:54.220 with his

00:53:54.600 companies

00:53:55.020 is it

00:53:55.500 that he

00:53:56.000 wants

00:53:56.400 to

00:53:56.720 is it

00:53:57.580 that he

00:53:57.820 wants

00:53:58.020 power

00:53:58.380 is it

00:53:59.180 that he

00:53:59.440 wants

00:53:59.740 money

00:54:00.060 even more

00:54:00.500 money

00:54:00.700 than he

00:54:00.960 already

00:54:01.180 has

00:54:01.440 as ridiculous

00:54:01.920 as that

00:54:02.280 may seem

00:54:02.700 or is

00:54:03.480 he more

00:54:04.160 interested in

00:54:04.840 helping humanity

00:54:05.640 I think

00:54:07.480 the thing

00:54:08.900 that drives

00:54:09.680 him most

00:54:10.500 is the

00:54:10.940 thing that

00:54:11.620 he also

00:54:12.440 says that

00:54:13.020 drives him

00:54:13.460 most

00:54:13.720 which is

00:54:14.340 getting

00:54:15.140 humanity

00:54:15.780 to Mars

00:54:16.460 for the

00:54:17.800 reason that

00:54:18.440 currently we

00:54:19.880 only have

00:54:20.300 Earth

00:54:20.620 and

00:54:21.460 that's not

00:54:23.320 a good

00:54:24.040 state for

00:54:24.720 humanity

00:54:25.120 to be

00:54:25.720 at

00:54:27.140 in the

00:54:27.920 long history

00:54:28.420 of the

00:54:28.740 universe

00:54:29.080 if we

00:54:31.040 only ever

00:54:31.440 stay here

00:54:31.940 then like

00:54:32.400 in some

00:54:33.000 hundreds

00:54:33.260 of

00:54:33.460 millions

00:54:33.640 of

00:54:33.860 years

00:54:34.160 different

00:54:35.140 estimations

00:54:35.620 for it

00:54:35.940 like the

00:54:36.320 Earth

00:54:36.540 will burn

00:54:37.020 up

00:54:37.220 and then

00:54:39.160 we're

00:54:39.520 done

00:54:39.700 here

00:54:39.920 so like

00:54:40.660 we kind

00:54:41.500 of got

00:54:41.700 to get

00:54:41.960 somewhere

00:54:42.600 else

00:54:42.980 else

00:54:43.720 our

00:54:44.040 lifetime

00:54:44.460 is much

00:54:44.880 shorter

00:54:45.180 because

00:54:45.780 like the

00:54:46.200 star

00:54:46.760 era

00:54:47.040 is going

00:54:47.560 to go

00:54:47.840 for like

00:54:48.280 a thousand

00:54:48.640 times

00:54:48.980 more

00:54:49.220 than the

00:54:49.620 time

00:54:49.860 that we

00:54:50.120 would have

00:54:50.500 on Earth

00:54:50.840 in the

00:54:51.580 sense

00:54:51.920 of the

00:54:52.280 new

00:54:52.660 formation

00:54:53.060 of stars

00:54:53.500 so there

00:54:54.160 will be

00:54:54.340 many

00:54:54.520 other

00:54:54.720 planets

00:54:55.000 to go

00:54:55.300 to

00:54:55.480 obviously

00:54:56.060 it's

00:54:56.300 very

00:54:56.500 far

00:54:56.700 out

00:54:56.940 but

00:54:57.140 like

00:54:57.340 it's

00:54:57.520 a

00:54:57.620 very

00:54:57.860 aspirational

00:54:59.160 vision

00:54:59.520 of the

00:54:59.900 future

00:55:00.160 that I

00:55:00.660 at least

00:55:01.260 like

00:55:01.480 as someone

00:55:01.760 who read

00:55:02.060 a bunch

00:55:02.260 of sci-fi

00:55:02.700 I'm like

00:55:03.000 very excited

00:55:03.540 about

00:55:03.860 yeah

00:55:05.260 so

00:55:05.700 I think

00:55:06.980 that's

00:55:07.680 what drives

00:55:08.080 him

00:55:08.320 it's

00:55:09.140 this

00:55:09.340 desire

00:55:10.860 to go

00:55:11.160 to Mars

00:55:11.500 and does

00:55:11.740 he think

00:55:12.040 that this

00:55:12.320 will happen

00:55:12.700 in his

00:55:13.000 lifetime

00:55:13.400 I don't

00:55:15.820 know

00:55:16.120 if he

00:55:16.760 I mean

00:55:17.460 he's

00:55:17.840 damn well

00:55:18.520 gonna try

00:55:19.060 and I think

00:55:19.680 it's very

00:55:19.940 possible

00:55:20.340 that it

00:55:20.800 will happen

00:55:21.260 and like

00:55:22.080 we're

00:55:22.600 getting

00:55:23.280 closer

00:55:23.700 right

00:55:23.980 like

00:55:24.280 we might

00:55:25.480 I don't

00:55:25.800 know

00:55:25.900 we might

00:55:26.240 be

00:55:26.540 10 years

00:55:27.240 away

00:55:27.480 from like

00:55:28.300 at least

00:55:28.620 starship

00:55:29.000 or five

00:55:29.360 years

00:55:29.560 maybe

00:55:29.780 even

00:55:30.040 at least

00:55:30.420 starship

00:55:31.020 getting

00:55:31.340 to Mars

00:55:31.760 so

00:55:32.360 there's

00:55:32.600 this

00:55:32.740 craft

00:55:33.540 called

00:55:33.820 starship

00:55:34.320 so tell

00:55:34.660 us

00:55:34.800 about

00:55:35.040 that

00:55:35.220 what

00:55:35.420 does

00:55:35.740 that

00:55:35.840 involve

00:55:36.060 is that

00:55:36.360 unmanned

00:55:36.920 is that

00:55:37.600 man

00:55:37.940 what's

00:55:38.740 the

00:55:38.820 plan

00:55:39.060 yeah

00:55:39.380 so

00:55:39.800 I think

00:55:40.520 like

00:55:40.660 what they

00:55:40.900 say

00:55:41.080 is

00:55:41.200 that

00:55:41.460 like

00:55:42.360 it's

00:55:43.140 unclear

00:55:43.400 like

00:55:43.600 we only

00:55:43.860 get

00:55:44.060 like

00:55:44.260 a

00:55:44.400 close

00:55:44.760 pass

00:55:45.300 of

00:55:46.100 Earth

00:55:46.340 and

00:55:46.480 Mars

00:55:46.740 every

00:55:47.060 what

00:55:47.360 is

00:55:47.440 it

00:55:47.520 like

00:55:47.640 roughly

00:55:47.840 two

00:55:48.020 years

00:55:48.320 a bit

00:55:49.140 over

00:55:49.320 I

00:55:49.460 think

00:55:49.620 and

00:55:50.160 those

00:55:51.200 are

00:55:51.340 opportunities

00:55:51.780 to get

00:55:52.280 to

00:55:52.660 Mars

00:55:52.900 basically

00:55:53.200 and

00:55:53.340 I

00:55:53.440 think

00:55:53.560 the

00:55:53.680 first

00:55:53.860 few

00:55:54.020 times

00:55:54.280 you

00:55:54.400 would

00:55:54.540 send

00:55:54.820 unmanned

00:55:55.940 starship

00:55:56.780 over

00:55:57.020 then

00:55:57.420 you

00:55:57.520 would

00:55:57.640 have

00:55:57.940 maybe

00:55:58.860 only

00:55:59.100 one

00:55:59.420 time

00:55:59.700 you

00:56:00.220 would

00:56:00.360 have

00:56:00.640 at least

00:56:01.180 something

00:56:01.560 of a

00:56:01.880 base

00:56:02.120 already

00:56:02.460 there

00:56:02.780 and

00:56:04.000 we

00:56:04.200 have

00:56:04.620 rovers

00:56:05.040 already

00:56:05.500 right

00:56:05.820 so

00:56:06.700 you

00:56:07.880 would

00:56:08.020 do

00:56:08.140 that

00:56:08.420 and

00:56:08.600 then

00:56:08.740 afterwards

00:56:09.100 you

00:56:09.360 can

00:56:09.500 send

00:56:09.700 humans

00:56:10.000 out

00:56:10.220 there

00:56:10.420 and

00:56:11.620 then

00:56:12.580 you

00:56:12.720 can

00:56:12.880 start

00:56:13.240 building

00:56:13.680 a

00:56:14.880 actual

00:56:16.120 base

00:56:16.500 on

00:56:16.680 Mars

00:56:16.960 which

00:56:17.440 I

00:56:19.100 think

00:56:19.400 would

00:56:19.560 just

00:56:19.720 be

00:56:20.000 insane

00:56:20.620 to

00:56:20.960 think

00:56:21.340 about

00:56:21.600 that

00:56:21.780 humans

00:56:22.220 would

00:56:22.520 be

00:56:22.640 on

00:56:22.780 another

00:56:23.040 planet

00:56:23.360 and

00:56:24.020 then

00:56:24.300 we

00:56:24.520 can

00:56:24.840 start

00:56:25.780 from

00:56:27.140 there

00:56:27.460 you

00:56:27.600 would

00:56:27.740 probably

00:56:28.020 open

00:56:28.320 up

00:56:28.640 much

00:56:29.220 more

00:56:30.040 desire

00:56:30.460 to

00:56:30.940 go

00:56:31.300 out

00:56:31.580 and

00:56:31.760 settle

00:56:31.960 the

00:56:32.180 stars

00:56:32.520 and

00:56:34.200 that

00:56:34.540 would

00:56:34.700 come

00:56:34.980 with

00:56:36.000 completely

00:56:37.020 new

00:56:37.340 industries

00:56:37.720 that

00:56:37.960 would

00:56:38.100 be

00:56:38.240 much

00:56:38.580 developed

00:56:39.720 that

00:56:39.960 would

00:56:40.040 be

00:56:40.160 much

00:56:40.360 larger

00:56:40.780 I

00:56:41.920 don't

00:56:42.020 know

00:56:42.120 the

00:56:42.260 future

00:56:42.600 that

00:56:43.080 is

00:56:43.480 described

00:56:43.960 in

00:56:44.160 all

00:56:44.260 of

00:56:44.360 the

00:56:44.880 books

00:56:45.060 I

00:56:45.260 at

00:56:45.400 least

00:56:45.560 find

00:56:45.860 extremely

00:56:46.500 exciting

00:56:46.960 what

00:56:49.980 do

00:56:50.580 you

00:56:50.720 make

00:56:51.000 of

00:56:51.320 Twitter

00:56:52.580 and

00:56:52.940 the

00:56:53.500 direction

00:56:53.780 it's

00:56:54.020 going

00:56:54.200 particularly

00:56:54.560 since

00:56:55.060 it

00:56:55.460 was

00:56:55.560 taken

00:56:55.780 over

00:56:56.060 so

00:56:57.140 I

00:56:57.860 mean

00:56:57.980 I've

00:56:58.260 started

00:56:58.500 using

00:56:58.800 it

00:56:58.980 more

00:56:59.180 I

00:56:59.340 don't

00:56:59.460 know

00:56:59.580 if

00:56:59.680 that's

00:56:59.940 good

00:57:00.200 or

00:57:00.320 bad

00:57:00.560 I

00:57:00.700 can't

00:57:00.900 really

00:57:01.060 tell

00:57:01.360 it's

00:57:01.620 probably

00:57:04.000 bad

00:57:04.300 in

00:57:04.420 some

00:57:04.580 ways

00:57:04.720 I

00:57:04.840 started

00:57:05.040 contributing

00:57:05.460 more

00:57:05.840 which

00:57:06.060 is

00:57:06.160 probably

00:57:06.420 a

00:57:07.080 good

00:57:07.260 thing

00:57:07.480 at

00:57:07.620 least

00:57:07.820 rather

00:57:08.100 than

00:57:08.300 just

00:57:08.520 read

00:57:08.740 I

00:57:09.980 don't

00:57:10.120 know

00:57:10.240 I

00:57:10.460 previously

00:57:10.920 definitely

00:57:11.280 thought

00:57:11.640 that

00:57:12.140 man

00:57:13.120 it would

00:57:14.120 be great

00:57:14.580 if

00:57:15.000 a man

00:57:16.320 like him

00:57:16.600 was

00:57:16.760 focused

00:57:17.060 more

00:57:17.320 on

00:57:17.500 technologies

00:57:18.040 that

00:57:18.800 are

00:57:19.080 hard

00:57:19.580 to

00:57:19.760 build

00:57:20.040 in

00:57:20.380 the

00:57:21.040 physical

00:57:21.300 world

00:57:21.600 that

00:57:21.760 seems

00:57:21.940 to

00:57:22.080 be

00:57:22.340 what

00:57:22.960 he's

00:57:23.240 so

00:57:23.740 good

00:57:24.120 at

00:57:24.340 but

00:57:25.880 I

00:57:26.060 do

00:57:26.360 come

00:57:26.780 to

00:57:27.200 worry

00:57:27.920 more

00:57:28.860 and

00:57:29.060 more

00:57:29.340 about

00:57:30.000 free

00:57:31.860 speech

00:57:32.160 suppression

00:57:32.620 so

00:57:33.200 Twitter

00:57:33.920 being

00:57:34.440 open

00:57:35.700 in that

00:57:36.380 way

00:57:36.660 I

00:57:37.400 start

00:57:37.740 valuing

00:57:38.200 more

00:57:38.500 I

00:57:38.860 don't

00:57:38.980 know

00:57:39.100 yet

00:57:39.320 where

00:57:39.520 the

00:57:39.700 company

00:57:39.960 is

00:57:40.100 going

00:57:40.340 I

00:57:40.460 think

00:57:40.580 it's

00:57:40.700 too

00:57:40.840 early

00:57:41.040 to

00:57:41.220 tell

00:57:41.460 from

00:57:42.140 a

00:57:42.340 pure

00:57:42.540 business

00:57:42.860 perspective

00:57:43.360 as

00:57:44.100 long

00:57:45.020 as

00:57:45.180 there

00:57:45.340 in

00:57:46.140 any

00:57:46.340 transitional

00:57:46.780 period

00:57:47.180 it's

00:57:47.460 always

00:57:47.760 some

00:57:48.600 things

00:57:48.820 go

00:57:49.000 bad

00:57:49.260 some

00:57:49.500 things

00:57:49.680 go

00:57:49.860 well

00:57:50.140 I

00:57:51.160 wouldn't

00:57:51.420 have

00:57:51.560 made

00:57:51.780 many

00:57:52.020 of

00:57:52.160 the

00:57:52.260 same

00:57:52.420 choices

00:57:52.800 but

00:57:53.040 I

00:57:53.360 also

00:57:53.580 don't

00:57:53.880 know

00:57:54.160 technology

00:57:54.800 as

00:57:55.060 well

00:57:55.240 as

00:57:55.440 he

00:57:56.700 does

00:57:56.960 so

00:57:57.400 I

00:57:59.280 think

00:57:59.440 it's

00:57:59.560 too

00:57:59.680 early

00:57:59.840 to

00:58:00.020 judge

00:58:00.240 basically

00:58:00.640 I

00:58:01.180 do

00:58:01.900 agree

00:58:02.240 with

00:58:08.500 that

00:58:08.740 is

00:58:08.940 extremely

00:58:09.580 important

00:58:10.020 and

00:58:10.320 I

00:58:10.440 hope

00:58:10.560 that

00:58:10.680 we

00:58:10.800 get

00:58:11.040 there

00:58:11.220 with

00:58:11.400 it

00:58:11.600 yeah

00:58:12.260 absolutely

00:58:12.800 well

00:58:13.360 listen

00:58:13.600 before

00:58:14.020 we

00:58:14.200 go

00:58:14.340 to

00:58:14.520 locals

00:58:14.860 for

00:58:15.160 a

00:58:15.320 few

00:58:15.420 questions

00:58:15.800 from

00:58:16.000 our

00:58:16.160 supporters

00:58:16.560 the

00:58:16.860 question

00:58:17.200 we

00:58:17.340 always

00:58:17.540 end

00:58:17.740 on

00:58:17.960 in

00:58:18.440 the

00:58:18.540 main

00:58:18.780 part

00:58:19.360 of

00:58:19.440 the

00:58:19.540 interviews

00:58:19.880 what's

00:58:20.500 the

00:58:20.580 one

00:58:20.720 thing

00:58:20.880 we're

00:58:21.060 not

00:58:21.240 talking

00:58:21.600 about

00:58:21.900 that

00:58:22.060 we

00:58:22.160 really

00:58:22.400 should

00:58:22.560 be

00:58:22.700 as

00:58:22.840 a

00:58:22.940 society

00:58:23.360 hmm

00:58:24.480 that

00:58:27.200 we're

00:58:27.380 not

00:58:27.800 talking

00:58:28.440 about

00:58:28.960 so

00:58:31.960 we

00:58:32.220 started

00:58:32.640 fortunately

00:58:33.180 talking

00:58:33.680 about

00:58:34.640 AI

00:58:34.920 safety

00:58:35.240 that

00:58:38.500 there

00:58:38.780 are

00:58:38.940 smaller

00:58:39.580 issues

00:58:40.080 that

00:58:40.520 could

00:58:40.720 be

00:58:40.880 discussed

00:58:41.300 I

00:58:42.320 think

00:58:43.040 it's

00:58:43.220 awful

00:58:43.580 that

00:58:44.140 teenagers

00:58:45.340 and

00:58:45.680 children

00:58:45.900 wake up

00:58:46.420 to

00:58:46.560 school

00:58:46.820 so

00:58:47.060 early

00:58:47.320 and

00:58:47.520 mess

00:58:48.180 up

00:58:48.400 their

00:58:48.560 whole

00:58:48.920 brain

00:58:49.640 chemistry

00:58:49.920 which

00:58:50.160 seems

00:58:50.340 to

00:58:50.460 be

00:58:50.540 the

00:58:50.680 case

00:58:50.880 it's

00:58:51.220 not

00:58:51.360 good

00:58:51.520 for

00:58:51.640 them

00:58:51.760 to

00:58:51.900 be

00:58:52.040 up

00:58:52.260 so

00:58:52.460 early

00:58:52.700 and

00:58:52.880 they

00:58:53.200 sleep

00:58:53.400 very

00:58:53.620 little

00:58:53.920 the

00:58:55.140 current

00:58:55.360 school

00:58:55.740 regime

00:58:56.080 is

00:58:56.240 just

00:58:56.380 very

00:58:56.560 bad

00:58:56.800 I

00:58:56.940 think

00:58:57.180 that

00:58:59.300 could

00:58:59.500 have

00:58:59.640 a

00:58:59.740 bit

00:58:59.840 more

00:59:00.020 attention

00:59:00.440 than

00:59:01.740 what

00:59:04.300 else

00:59:04.520 will

00:59:04.680 be

00:59:04.780 different

00:59:05.520 in

00:59:08.500 we're

00:59:10.680 talking

00:59:11.920 no

00:59:13.060 we're

00:59:13.240 actually

00:59:13.420 doing

00:59:13.760 decent

00:59:14.220 amount

00:59:14.600 now

00:59:14.940 on

00:59:15.320 like

00:59:15.660 animal

00:59:16.100 welfare

00:59:16.520 there

00:59:17.300 are

00:59:17.400 definitely

00:59:17.620 some

00:59:17.880 ills

00:59:18.100 that

00:59:18.300 are

00:59:18.400 pretty

00:59:18.600 bad

00:59:18.940 yeah

00:59:22.300 I'll

00:59:22.500 stick

00:59:22.700 with

00:59:22.840 the

00:59:23.100 school

00:59:23.320 one

00:59:23.580 school

00:59:24.220 one

00:59:24.400 is

00:59:24.500 good

00:59:24.700 yeah

00:59:24.980 all right

00:59:25.320 perfect

00:59:26.020 well

00:59:26.420 thanks

00:59:26.680 for

00:59:26.800 coming

00:59:26.960 on

00:59:27.180 and

00:59:27.540 see

00:59:27.780 you

00:59:27.900 on

00:59:28.000 locals

00:59:28.320 this

00:59:30.100 is a

00:59:30.280 question

00:59:30.520 coming

00:59:30.780 back

00:59:30.960 to

00:59:31.080 your

00:59:31.180 poker

00:59:31.500 days

00:59:31.920 Colby

00:59:32.940 Hamilton

00:59:33.260 says

00:59:33.580 what

00:59:33.880 is

00:59:34.020 some

00:59:34.180 advice

00:59:34.560 for

00:59:34.740 hiding

00:59:35.120 my

00:59:35.400 towels

00:59:35.800 love

00:59:37.340 love

00:59:38.060 love

00:59:38.620 love

00:59:39.500 love

00:59:41.740 love

00:59:43.160 love

00:59:43.960 love

00:59:44.380 love

00:59:45.860 love

00:59:48.220 love

00:59:53.560 love

00:59:54.000 love

00:59:55.780 love

00:59:57.900 love

01:00:00.280 love

01:00:02.600 love

01:00:03.000 love

01:00:03.560 love

01:00:04.520 love

01:00:04.840 love

01:00:05.000 love

01:00:05.420 love

01:00:05.540 love

TRIGGERnometry - January 07, 2024

Will AI Destroy The World？ - Igor Kurganov

Episode Stats

Length

Words per Minute

Word Count

Sentence Count

Misogynist Sentences

Hate Speech Sentences

Summary

Transcript