Latest / Faster, Please! — The Podcast / ✨ My interview with Sebastian Mallaby, author of 'The Infinity Machine: Demis Hassabis, DeepMind, and the Quest for Superintelligence'
Transcript
- 0:00SPEAKER_01: Welcome to Faster, Please!, the podcast.
- 0:03SPEAKER_01: I'm Jim Pethokoukis,
- 0:04SPEAKER_01: and my guest today is journalist Sebastian Malby,
- 0:07SPEAKER_01: The Paul A.
- 0:08SPEAKER_01: Volk Volcker Senior Fellow for International Economics at the Council
- 0:11SPEAKER_01: of Foreign Relations and a widely read columnist for the Washington Post.
- 0:15SPEAKER_01: He is also author of the new bestselling book,
- 0:18SPEAKER_01: The Infinity Machine, Damesa Sabas, Sabas, DeepMind.
- 0:22SPEAKER_01: And the Quest for Super Intelligence.
- 0:26SPEAKER_01: Sebastian,
- 0:27SPEAKER_01: right there in the introduction of the book,
- 0:30SPEAKER_01: Sentence number three,
- 0:32SPEAKER_01: you write,
- 0:33SPEAKER_01: at some point in the not so distant future,
- 0:36SPEAKER_01: Artificial and intelligence will beat human intelligence
- 0:39SPEAKER_01: at almost every mental task,
- 0:41SPEAKER_01: and to say this marks a watershed would be a parody of understatement.
- 0:46SPEAKER_01: AI holds a transformation more profound than anything since Homo sapiens acquired
- 0:51SPEAKER_01: the capacity for, since abstract thought some 70,000 years ago.
- 0:56SPEAKER_01: In my view, you sell your like a true believer that this quest is Will Be Successful.
- 1:01SPEAKER_01: Yeah.
- 1:03SPEAKER_01: I the book?
- 1:04SPEAKER_00: Yeah. I mean, And I began the...
- 1:07SPEAKER_00: book in 2022, And you already have...
- 1:11SPEAKER_00: had a And a couple of real real breakthroughs.
- 1:15SPEAKER_00: One was the AlphaGo system that DeepMind had built, which surpassed all the Best human players.
- 1:21SPEAKER_00: And that was significant because obviously Go is this massive combinatorial space.
- 1:26SPEAKER_00: There are 19 by 19 squares on the board.
- 1:29SPEAKER_00: So the very first move, there's 361 options.
- 1:32SPEAKER_00: The second move, there's 360.
- 1:34SPEAKER_00: You multiply that out,
- 1:35SPEAKER_00: you get to a number Pretty close to infinity in terms of the number of permutations
- 1:40SPEAKER_00: that you could have.
- 1:42SPEAKER_00: And yet the machine figures out the pattern.
- 1:45SPEAKER_00: Patterns and figures out the best strategies And can must.
- 1:48SPEAKER_00: Master that infinity of data.
- 1:51SPEAKER_00: And then the same thing,
- 1:52SPEAKER_00: but actually even more extreme,
- 1:54SPEAKER_00: with protein filled folding in 2020, Another deep mindset.
- 1:57SPEAKER_00: DeepMind system, Deep Mind being this.
- 2:00SPEAKER_00: The startup that Demis Isabis, my main character, had started in London in 2010.
- 2:06SPEAKER_00: And, you know, again, protein folding, Just an annoying.
- 2:10SPEAKER_00: Enormous number of possible permutations and finding the way that a protein is going
- 2:14SPEAKER_00: to fold itself up like a self-executing origami model is,
- 2:18SPEAKER_00: you know, like searching for a needle in sort of, you know...
- 2:24SPEAKER_00: 10 to the 100.
- 2:25SPEAKER_00: No, 10 to the 300 hasty.
- 2:27SPEAKER_00: Haystacks.
- 2:29SPEAKER_00: It's just this incredibly big number of possibilities.
- 2:32SPEAKER_00: And so you had two instances where infinity had been conquered,
- 2:37SPEAKER_00: hence my title, The Infinity Machine.
- 2:39SPEAKER_00: And it felt in 2020.
- 2:41SPEAKER_00: 2022 when I began the project that this was going to break out.
- 2:44SPEAKER_00: I didn't know it would break out quite as fast as it did.
- 2:46SPEAKER_00: In fact, I pitched Demis Isabis on the idea of this book in November of 2022.
- 2:53SPEAKER_00: And at the end of that month is when ChatGPT came out.
- 2:56SPEAKER_00: So I was lucky that it broke out sort of immediately into the mainstream after I began.
- 3:02SPEAKER_00: But yeah, I mean...
- 3:04SPEAKER_00: um...
- 3:04SPEAKER_00: You know, I just think you look backwards, you see how fast the progress has been.
- 3:09SPEAKER_00: And, you know,
- 3:10SPEAKER_00: to merely extrapolate forwards is probably to undersell the speed at
- 3:13SPEAKER_00: which we'll accelerate in the future because there's a,
- 3:16SPEAKER_00: you know,
- 3:17SPEAKER_00: there's an accelerating phenomenon here where the more advanced you are,
- 3:22SPEAKER_00: the more easy it is to get to the next level.
- 3:25SPEAKER_01: Um, and I And I hope you don't object to my use of the phrase true believer
- 3:29SPEAKER_01: with its quasi-religious aspect, but You're, you.
- 3:34SPEAKER_01: You have confidence this quest will be successful.
- 3:36SPEAKER_01: Um.
- 3:38SPEAKER_01: Are you enthusiastic about that?
- 3:40SPEAKER_01: There are people who believe it will be successful who are not enthusiastic about
- 3:44SPEAKER_01: that,
- 3:44SPEAKER_01: for all manner of reasons Raised,
- 3:48SPEAKER_01: you know,
- 3:48SPEAKER_01: ranging from from they don't like data centers to they think I think
- 3:51SPEAKER_01: the robots will take
- 3:52SPEAKER_01: Take all the jobs and when they're done taking all the jobs, they'll kill us.
- 3:55ha, ha.
- 3:57SPEAKER_00: Well, that's a more complicated question and I do have mixed feelings about that.
- 4:01SPEAKER_00: I mean, I think You know,
- 4:02SPEAKER_00: in one sense,
- 4:03SPEAKER_00: you can say,
- 4:04SPEAKER_00: obviously,
- 4:05SPEAKER_00: there are huge upsides and huge downsides,
- 4:07SPEAKER_00: and anybody honest is going to recognize both,
- 4:09SPEAKER_00: but then also perhaps reckon that humans have always Face this trade.
- 4:14SPEAKER_00: Trade-off when confronting technological disruption.
- 4:17SPEAKER_00: Um, and And they go forward because that's what defines being human,
- 4:21SPEAKER_00: you know, the willingness to Accept the answer.
- 4:24SPEAKER_00: Uncertainty, the thrill, the adventure of a new technology coming down the pike.
- 4:27SPEAKER_00: I mean, we just never stop it.
- 4:29SPEAKER_00: We don't choose not to go forward.
- 4:31SPEAKER_00: Um...
- 4:33SPEAKER_00: And so, you know,
- 4:34SPEAKER_00: presumably we're going to navigate this one and we'll be okay and we'll manage
- 4:37SPEAKER_00: the downside somehow.
- 4:40SPEAKER_00: But, you know, on another level, one could debate the optimal speed of this transition.
- 4:46SPEAKER_00: I mean, maybe Society with would have a better shot at digesting this
- 4:50SPEAKER_00: if it could be stretched out.
- 4:53SPEAKER_00: You know,
- 4:53SPEAKER_00: if I project that Let's say we take the anthropic projection,
- 4:59SPEAKER_00: which is recursive self-improvement, i.e.
- 5:02SPEAKER_00: the machine codes the next version of the machine.
- 5:04SPEAKER_00: This is projected to arrive at bianthropic at end of 2028.
- 5:10SPEAKER_00: That's their base case.
- 5:11SPEAKER_00: Case.
- 5:11SPEAKER_00: So let's say they're a bit optimistic.
- 5:14SPEAKER_00: Let's say it's a couple of years after that.
- 5:16SPEAKER_00: But we are very close now to a situation where the frontier model codes the next frontier model.
- 5:23SPEAKER_00: Now, would I prefer that to be like 10 years in the future, not three or four?
- 5:28SPEAKER_00: Probably.
- 5:29SPEAKER_00: I think it would make a difference.
- 5:29SPEAKER_00: It easier for society to digest it.
- 5:32SPEAKER_01: That reminds me of a comment.
- 5:33SPEAKER_01: Conversation I had with the economist David Otter, who's written about the China trade shock.
- 5:39SPEAKER_01: And I recall asking him, Um, given the, uh...
- 5:42SPEAKER_01: disruption...
- 5:42SPEAKER_01: The disruption.
- 5:44SPEAKER_01: Would have been And better somehow.
- 5:46SPEAKER_01: Somehow as unlikely as it had been to not do that,
- 5:50SPEAKER_01: to not open with China, though obviously I think that would be impossible.
- 5:54SPEAKER_01: It's like,
- 5:54SPEAKER_01: I guess the only thing I would say is perhaps if we are kind
- 5:57SPEAKER_01: of slowed down somehow our exposure and exposing workers to this,
- 6:02SPEAKER_01: you know, Humongous others.
- 6:04SPEAKER_01: Other labor market and giving people more time to adjust,
- 6:09SPEAKER_01: perhaps that would have been a good idea.
- 6:10SPEAKER_01: Again, I don't know How, how, how light.
- 6:12SPEAKER_01: Likely that Um, uh.
- 6:14SPEAKER_01: With AI.
- 6:15SPEAKER_01: And having spoken once myself To Demisa Sabin.
- 6:21SPEAKER_01: Sabas, he's a very calming presence.
- 6:24SPEAKER_01: He does not seem wild I'm not eyed.
- 6:29SPEAKER_01: This was maybe, I don't know, maybe a year ago.
- 6:31SPEAKER_01: You sound like a person.
- 6:32SPEAKER_01: Person who picked his words with great care.
- 6:36SPEAKER_01: Given that,
- 6:37SPEAKER_01: and your your many conversations with him,
- 6:40SPEAKER_01: Did your feeling about the wisdom of this project change at all during the writing
- 6:47SPEAKER_01: of the book?
- 6:44SPEAKER_01: Dread.
- 6:47SPEAKER_01: Was there ever a point where you were like,
- 6:49SPEAKER_01: I'm not sure this is a good idea given all the risks,
- 6:53SPEAKER_01: but talking to him sort of maybe think, yes, one, it's inevitable and it'll be okay.
- 7:01SPEAKER_01: We'll figure out and forward we will go.
- 7:04SPEAKER_00: Hmm.
- 7:05SPEAKER_00: I would say,
- 7:06SPEAKER_00: actually,
- 7:06SPEAKER_00: I became a bit more pessimistic as my project moved along,
- 7:11SPEAKER_00: partly because you can read Demi Sosabes' story and the way I tell it in
- 7:16SPEAKER_00: the book really You know, tries to bring this out.
- 7:18SPEAKER_00: Out.
- 7:19SPEAKER_00: As sort of a morality tale of a good person who has,
- 7:23SPEAKER_00: you know, honest,
- 7:25SPEAKER_00: genuine desire to make the technology beneficial for humanity,
- 7:29SPEAKER_00: who starts By earlier.
- 7:31SPEAKER_00: Early on,
- 7:31SPEAKER_00: you know,
- 7:32SPEAKER_00: stipulating that the sale of his company to Google will only go through if there's
- 7:36SPEAKER_00: a safety and ethics oversight board to make sure So, the technology is rolling.
- 7:40SPEAKER_00: Rolled out in a good way.
- 7:42SPEAKER_00: Who then fights a three-year legal battle secretly with Mountain View and Alphabet
- 7:47SPEAKER_00: to try to enforce that agreement with which Alphabet kind of pulled back from.
- 7:51SPEAKER_00: Um...
- 7:53SPEAKER_00: You know, and so really not only talk the talk about safety, but walk the walk.
- 7:56SPEAKER_00: And yet at the end of my story...
- 7:58SPEAKER_00: He's locked in the this You know, crazy capitalist.
- 8:02SPEAKER_00: Capitalist race between multiple labs in China and multiple labs in the West,
- 8:06SPEAKER_00: where each of them is locked in by the race dynamic.
- 8:10SPEAKER_00: If Demis decided to close down Google DeepMind tomorrow
- 8:13SPEAKER_00: and do no further AI research,
- 8:15SPEAKER_00: it wouldn't really decelerate things at all because there are
- 8:17SPEAKER_00: so many other labs doing it.
- 8:19SPEAKER_00: So they're all locked in and none of them has control of the pace of deployment.
- 8:24SPEAKER_00: And that doesn't fit.
- 8:26SPEAKER_00: Feel like a situation in which control people can really do much.
- 8:31SPEAKER_01: But the reason you were given this kind of access to this company
- 8:33SPEAKER_01: Um...
- 8:36SPEAKER_01: And to Hassan.
- 8:37SPEAKER_01: Hassabis It was just the opposite,
- 8:39SPEAKER_01: so that people would understand the technology and they would,
- 8:43SPEAKER_01: I think to some degree,
- 8:45SPEAKER_01: understand that at least he was a thoughtful person that you could have confidence
- 8:50SPEAKER_01: in, would do the right thing so it didn't take all the jobs and kill us all.
- 8:54SPEAKER_01: Um, But if he became less confident,
- 8:58SPEAKER_01: then that was perhaps a mistake, him giving you this deep access.
- 9:02SPEAKER_00: Yeah, maybe.
- 9:03SPEAKER_00: I mean, you know,
- 9:03SPEAKER_00: I made a distinction between the fact that he's good,
- 9:06SPEAKER_00: which I do believe he is, and can he do good, which is a different matter.
- 9:12SPEAKER_00: And as he kind of admits to me at the end of the book,
- 9:15SPEAKER_00: you know,
- 9:15SPEAKER_00: he's optimistic still,
- 9:17SPEAKER_00: but without actually much objective reason for being optimistic because
- 9:20SPEAKER_00: as I say he's lost control of something that he thought at
- 9:23SPEAKER_00: the beginning He believed,
- 9:26SPEAKER_00: of the reason he lost control is that the early assumptions were naive,
- 9:29SPEAKER_00: He believed,
- 9:31SPEAKER_00: for example,
- 9:31SPEAKER_00: that there would be one frontier AI lab rolling out AI for all of humanity with
- 9:38SPEAKER_00: a kind of united team of scientists working together.
- 9:42SPEAKER_01: A Manhattan Project model?
- 9:44SPEAKER_00: Precisely. And sometimes it's called the singleton scenario in the community.
- 9:49SPEAKER_00: And under that scenario, you could be super careful.
- 9:51SPEAKER_00: You could not roll anything out until you tested it 25 times.
- 9:56SPEAKER_00: And, and maybe.
- 9:57SPEAKER_00: Maybe you could make it safer.
- 9:58SPEAKER_00: But right now, you can't do that because everybody's racing everybody else.
- 10:02SPEAKER_00: Um, and And And, you know, some of the races.
- 10:05SPEAKER_00: Racers are releasing their models on an open weight basis where You know, there's no killing.
- 10:10SPEAKER_00: Kill switch for the model once it's in the hands of a private actor.
- 10:13SPEAKER_00: If the private actor turns out to be a criminal group, you've got a big problem.
- 10:17SPEAKER_00: And yet, That's the way we're headed as a world.
- 10:19SPEAKER_00: And so, yeah, I mean, you're right.
- 10:23SPEAKER_00: They hope to have a story about the reasonableness of Demis and they got a story
- 10:28SPEAKER_00: about the reasonableness of Demis.
- 10:29SPEAKER_01: For sure.
- 10:30SPEAKER_01: I think that that was successful.
- 10:31SPEAKER_01: Yes.
- 10:32SPEAKER_00: Yeah.
- 10:33SPEAKER_00: But, you know, there's a limit to the consequences of that reasonableness.
- 10:37SPEAKER_01: I mean, that does sound incredibly naive that if you think the technology is going
- 10:43SPEAKER_01: to be This powerful,
- 10:45SPEAKER_01: not just,
- 10:46SPEAKER_01: you know,
- 10:46SPEAKER_01: not just sort of the next stage in the digital revolution had began.
- 10:50SPEAKER_01: Began with the PC and then the PC plus the internet and the smartphone,
- 10:53SPEAKER_01: but something quite beyond that.
- 10:56SPEAKER_01: To not think that the economic and incentives would be such That there would be
- 11:02SPEAKER_01: a mad race to develop this technology,
- 11:06SPEAKER_01: keep it progressing as fast as possible and distribute it as widely as possible.
- 11:11SPEAKER_01: Like that, that would be the.
- 11:12SPEAKER_01: The controlling dynamic.
- 11:14SPEAKER_01: That would be the controlling dynamic.
- 11:16SPEAKER_01: Not You've got a,
- 11:17SPEAKER_01: you know,
- 11:17SPEAKER_01: 20 - know,
- 11:18SPEAKER_01: 21st century version of a bunch of brilliant people in a,
- 11:22SPEAKER_01: in a in a compound somewhere, you know, Giving out this text.
- 11:25SPEAKER_01: Technology After it's been tripled.
- 11:28SPEAKER_01: Triple checked, you know, in little bits over a long period of time.
- 11:32SPEAKER_01: I find it remarkable that that was an actual belief that he had.
- 11:37SPEAKER_00: Yeah, I mean, you're right.
- 11:39SPEAKER_00: It is remarkable.
- 11:39SPEAKER_00: And in retrospect, it's crazy as well.
- 11:42SPEAKER_00: And all one can say is that ex ante,
- 11:46SPEAKER_00: you know,
- 11:46SPEAKER_00: the atmosphere in the community of AI builders when Demis began his company
- 11:51SPEAKER_00: in 2010 was that this was a thing that simply didn't work, right?
- 11:55SPEAKER_00: You know, AI could not recognize the photograph of a cat.
- 11:59SPEAKER_00: AI could do nothing.
- 11:59SPEAKER_00: It was deep AI winter.
- 12:01SPEAKER_00: And so under those conditions,
- 12:03SPEAKER_00: You could assemble the entirety of the world's strong AI believers in one conference
- 12:07SPEAKER_00: in San Francisco.
- 12:09SPEAKER_00: And it felt like a single community.
- 12:10SPEAKER_00: And so this sort of singleton scenario where you just have one lab,
- 12:14SPEAKER_00: it sort of was a natural outgrowth of that moment in time.
- 12:18SPEAKER_00: But of course,
- 12:20SPEAKER_00: the moment it started to work,
- 12:22SPEAKER_00: then you were going to have,
- 12:23SPEAKER_00: you know,
- 12:24SPEAKER_00: multiple rival attempts to build it because humans are tribal and disputatious
- 12:29SPEAKER_00: and jealous.
- 12:30SPEAKER_01: In the end then,
- 12:31SPEAKER_01: does it really matter that The company is Google DeepMind and it's not MetaDeep.
- 12:39SPEAKER_01: DeepMind or Musk DeepMind.
- 12:42SPEAKER_01: That the dynamics are such that The dynamics of moving forward.
- 12:46SPEAKER_01: Move forward, As rapidly as I can.
- 12:48SPEAKER_01: Rapidly as possible and get business businesses to adopt these technologies
- 12:53SPEAKER_01: and use them, that dynamic was going to be the case.
- 12:55SPEAKER_01: Because at one point, Mark Zuckerberg was maybe going to buy DeepMind.
- 13:01SPEAKER_01: They like Google more because Google seemed to A higher, plays a higher emphasis.
- 13:05SPEAKER_01: Emphasis on And safety.
- 13:06SPEAKER_01: Safety.
- 13:07SPEAKER_01: But in the end, does any of that really matter?
- 13:10SPEAKER_00: I think it matters a You know, a bit.
- 13:12SPEAKER_00: Bit.
- 13:13SPEAKER_00: I think there are differences that one can see between the different lab leaders.
- 13:18SPEAKER_00: I would say Mark Zuckerberg is the most sort of purely commercial of the big tech
- 13:24SPEAKER_00: of AI building.
- 13:26SPEAKER_00: Builders.
- 13:27SPEAKER_00: I'd say that Google,
- 13:29SPEAKER_00: because of its strange history where it had this effectively monopoly product, Um...
- 13:35SPEAKER_00: which just printed money like nobody's business.
- 13:38SPEAKER_00: A large part of their strategy was to sort of not be too overtly commercial
- 13:44SPEAKER_00: because they were by dint didn't have like where they wound up almost by accident,
- 13:50SPEAKER_00: already excessively commercial, they were making too much money.
- 13:53SPEAKER_00: And so I have to.
- 13:54SPEAKER_00: Plenty of stories.
- 13:55SPEAKER_00: And I'm sure you know,
- 13:55SPEAKER_00: similar stories about people inside Google,
- 13:58SPEAKER_00: you know,
- 13:59SPEAKER_00: kind of five,
- 14:00SPEAKER_00: 10 years ago,
- 14:02SPEAKER_00: at being at meetings saying, Okay, well,
- 14:04SPEAKER_00: it's a bit embarrassing,
- 14:05SPEAKER_00: our profits are up again,
- 14:06SPEAKER_00: the antitrust people are going to come after us, what are we going to do?
- 14:09SPEAKER_00: Who's got an idea of how we bury You know,
- 14:12SPEAKER_00: a bunch of money and somebody puts their hands up and says,
- 14:14SPEAKER_00: you know, new research park in Taiwan.
- 14:16SPEAKER_00: And they say, you got it.
- 14:17SPEAKER_00: Let's spend as much.
- 14:18SPEAKER_00: I mean, Google was sort of post economic in this sense and their primary Concern,
- 14:25SPEAKER_00: or or at least maybe not.
- 14:27SPEAKER_00: Maybe the equal Thank you very much.
- 14:28SPEAKER_00: At the of senior level.
- 14:30SPEAKER_00: Level is,
- 14:31SPEAKER_00: how do you run this cash printing machine in a way that is It's going
- 14:34SPEAKER_00: to be sufficient.
- 14:35SPEAKER_00: Sufficiently socially acceptable that you don't get killed by government fines?
- 14:40SPEAKER_00: And it's not just EU fines, by the way, obviously it's in the US as well.
- 14:43SPEAKER_00: And so I think that particular produces a company which is instinctively...
- 14:47SPEAKER_00: Respectful of social norms and societal limits.
- 14:53SPEAKER_00: And so when they build data centers today,
- 14:56SPEAKER_00: I know they go out of their way to make sure You know, they're building...
- 15:00SPEAKER_00: building new power sources so that there won't be high electricity prices on
- 15:04SPEAKER_00: the local community,
- 15:04SPEAKER_00: that if the local community wants that new school or that new clinic,
- 15:08SPEAKER_00: they're going to provide it.
- 15:09SPEAKER_00: I mean, they really do care about building Local buy -in.
- 15:15SPEAKER_00: Buy-in for the data center.
- 15:17SPEAKER_00: And so I think there is a difference between Google and Meta.
- 15:21SPEAKER_00: I'd rather have Google build this kind of technology than Meta.
- 15:26SPEAKER_00: But, you know, the bigger picture is everybody's racing.
- 15:30SPEAKER_00: And ultimately,
- 15:30SPEAKER_00: if you want control,
- 15:32SPEAKER_00: the only person who can impose restraint on all the participants in
- 15:37SPEAKER_00: the race It's government.
- 15:38SPEAKER_00: Government.
- 15:39SPEAKER_01: So the fact that Hassabis might have an initial different intent or different vision
- 15:46SPEAKER_01: than Sam Altman or Daryl Modey or anybody.
- 15:49SPEAKER_01: Anybody else,
- 15:50SPEAKER_01: not inconsequential, but certainly I think likely to be dominated by just the race.
- 15:58SPEAKER_00: Yeah, yeah.
- 15:59SPEAKER_00: And I think there's a big inflection point in 2022.
- 16:02SPEAKER_00: I mean, if you think about what happened before 2022,
- 16:05SPEAKER_00: Demis had the freedom,
- 16:07SPEAKER_00: because he was clearly the leader, to sort of define What the next project is.
- 16:11SPEAKER_00: Project should be.
- 16:12SPEAKER_00: And so he chose at one point to go and do this protein folding project.
- 16:15SPEAKER_00: Project, they start.
- 16:16SPEAKER_00: Sunk four years of work into that, and the result was a Nobel Prize winning project.
- 16:19SPEAKER_00: Winning Breakthrough in Structural structural biology, which is very good for Medical Advance.
- 16:24SPEAKER_00: Advances and maybe material sciences and so forth.
- 16:27SPEAKER_00: And that clearly is good for humanity.
- 16:29SPEAKER_00: This is kind of AI with a smiley face painted on it.
- 16:33SPEAKER_00: Whereas,
- 16:34SPEAKER_00: you know,
- 16:34SPEAKER_00: once the chatbot went viral at the end of 22,
- 16:38SPEAKER_00: ChattyPT Chattypt, then everybody had to pile in and build a competitor.
- 16:43SPEAKER_00: And there's a lot less leeway to define your own path.
- 16:47SPEAKER_00: So I think, you know, the agency of the individual was quite strong Until 2020.
- 16:54SPEAKER_00: 2022.
- 16:55SPEAKER_00: And thereafter, the power of the race dynamic takes over.
- 17:00SPEAKER_01: The listeners to this podcast love when I read.
- 17:03SPEAKER_01: So I'm going to read another sentence, another beautifully crafted sentence by you here.
- 17:10SPEAKER_01: The way Hasab is saw things,
- 17:12SPEAKER_01: true general intelligence would make almost anything possible,
- 17:16SPEAKER_01: surpassing the internet,
- 17:17SPEAKER_01: the printing press, or even the Industrial Revolution in importance.
- 17:21SPEAKER_01: It would usher in a post-scarcity world A Radical Abundance.
- 17:24SPEAKER_01: Abundance, resembling the future described in the science fiction he read as a teenager.
- 17:30SPEAKER_01: And then you quote him,
- 17:31SPEAKER_01: people aren't thinking ambitiously enough about what a post-AGI,
- 17:35SPEAKER_01: meaning artificial general intelligence, world will look like.
- 17:38SPEAKER_01: Now, Sebastian, I work with and I talk to a lot of economists.
- 17:42SPEAKER_01: That kind of stuff sets their eyes rolling.
- 17:46SPEAKER_01: They just don't see it.
- 17:48SPEAKER_01: They don't see it in any kind of reasonable timeline.
- 17:50SPEAKER_01: They've, they've...
- 17:51SPEAKER_01: They think that these technologists,
- 17:53SPEAKER_01: the people in San Francisco, the singularity people, just...
- 17:58SPEAKER_01: Well, yeah.
- 18:25SPEAKER_00: I think the frontier of the technology is accelerating really, really fast.
- 18:29SPEAKER_00: And if you look back from 2022 with the arrival of ChatGPT and
- 18:34SPEAKER_00: then you just don't think what's happened since then.
- 18:35SPEAKER_00: You go from a model that hallucinated nonstop to a model
- 18:39SPEAKER_00: that mostly stopped hallucinating when you plugged in GPT-4 a few months later
- 18:44SPEAKER_00: to model where the a context window where you could upload the entirety of
- 18:48SPEAKER_00: a Tolstoy novel and have, you know, query it about the novel.
- 18:53SPEAKER_00: And then you get reasoning so that logic and mathematics becomes possible.
- 18:57SPEAKER_00: Then you get agentic models.
- 18:58SPEAKER_00: You get these coding assistants.
- 19:01SPEAKER_00: You know, they can handle multimodal inputs and, you know, sound and video.
- 19:06SPEAKER_00: I mean, the number of improvements before even we talk about Mythos and
- 19:11SPEAKER_00: the side cyber capabilities of that one.
- 19:15SPEAKER_00: I mean, it's been an extraordinary ride in what is actually less than four years.
- 19:21SPEAKER_00: So I don't take back anything I say about the speed of their the advance of the frontier.
- 19:26SPEAKER_00: Now, that's different to the speed of the deployment.
- 19:29SPEAKER_00: And And there I have a lot of stuff.
- 19:31SPEAKER_00: Sympathy with with the economy.
- 19:32SPEAKER_00: Economists.
- 19:33SPEAKER_00: I mean, I think maybe, You know, they're honest.
- 19:35SPEAKER_00: Ismoglu,
- 19:35SPEAKER_00: who is quoted a lot on this because he's on the cautious end of his expectations
- 19:40SPEAKER_00: in terms of productivity growth.
- 19:42SPEAKER_00: He may be too pessimistic.
- 19:43SPEAKER_00: Having talked to him, I don't agree with him.
- 19:45SPEAKER_00: I think he's he's not quite He doesn't actually use any of this stuff himself, for example.
- 19:50SPEAKER_00: He doesn't have a feel For quite hard.
- 19:53SPEAKER_00: How electrifying it can be to use Claude Fable.
- 19:56SPEAKER_00: And I think if you don't use it yourself, your pronouncements should be discounted.
- 20:01SPEAKER_00: So So, I don't agree with that.
- 20:03SPEAKER_00: Agree with him,
- 20:03SPEAKER_00: but I think You know,
- 20:04SPEAKER_00: I would put the the sort of mainstream of the economics profession between him And
- 20:11SPEAKER_00: the assumptions you have.
- 20:11SPEAKER_00: Assumptions you get from the pure technologists in Silicon Valley.
- 20:14SPEAKER_00: In other words, You know, the road rollout is going to take a while.
- 20:18SPEAKER_00: It's going to take a while for corporations to figure out how to make this productive for them.
- 20:23SPEAKER_00: There are certain applications like writing code where we are already there.
- 20:28SPEAKER_00: It's already super helpful.
- 20:30SPEAKER_00: It's beginning in things like law And call center.
- 20:33SPEAKER_00: Centers and so on.
- 20:34SPEAKER_00: But, you know, if you want to sort of re-engineer your, You know, fun.
- 20:39SPEAKER_00: Fundamentally how you are.
- 20:40SPEAKER_00: Your business is structured and kind of closed down.
- 20:42SPEAKER_00: Down departments, we're not there yet.
- 20:45SPEAKER_00: And so that's going to take a while.
- 20:46SPEAKER_00: And it's,
- 20:46SPEAKER_00: you know,
- 20:47SPEAKER_00: instructive to remember how many years ago people said,
- 20:50SPEAKER_00: well, there'll be driverless taxis absolutely everywhere.
- 20:54SPEAKER_00: And it just takes much longer than people expect.
- 20:56SPEAKER_00: But that doesn't mean,
- 20:58SPEAKER_00: you know,
- 20:58SPEAKER_00: that doesn't rule out the possibility that super abundance comes in the end.
- 21:01SPEAKER_00: Once you've figured out how to use this Powerful frontier tech.
- 21:04SPEAKER_00: Technology and you've deployed it everywhere,
- 21:07SPEAKER_00: maybe you do have a,
- 21:08SPEAKER_00: you know,
- 21:09SPEAKER_00: fundamentally bigger GDP number,
- 21:11SPEAKER_00: a fundamentally Now,
- 21:14SPEAKER_00: I
- 21:16SPEAKER_01: was going to say,
- 21:16SPEAKER_01: in the end, we're all dead, but I know that many of these CEOs do not believe that.
- 21:13SPEAKER_00: Productivity.
- 21:20SPEAKER_01: They believe that death will become optional, so we can get rid of that saying.
- 21:24SPEAKER_01: Little But That's not that sort of belief.
- 21:28SPEAKER_01: Belabor the point, but...
- 21:30SPEAKER_01: This is not,
- 21:30SPEAKER_01: you know,
- 21:31SPEAKER_01: radically new economics,
- 21:33SPEAKER_01: the notion that it takes a while for technology to be effectively implemented into
- 21:39SPEAKER_01: the economy.
- 21:41SPEAKER_01: And yet you have CEOs until very recently where I think the message has gotten
- 21:46SPEAKER_01: through saying that everybody who works behind a computer We'll be out of a job.
- 21:50SPEAKER_01: Job in 18 months.
- 21:52SPEAKER_01: Now, whether that's exactly what he meant, that is certainly how it was interpreted.
- 21:56SPEAKER_01: So, you know, is it any wonder that people are going to oppose Use data centers.
- 22:03SPEAKER_01: Centers that even if,
- 22:04SPEAKER_01: you know, why should you even take the environmental or the risk of hiring.
- 22:08SPEAKER_01: Higher electricity prices for a technology that will basically destroy the world
- 22:13SPEAKER_01: as I know it, my life in 18 months?
- 22:17SPEAKER_01: I think this has been incredible.
- 22:18SPEAKER_01: Incredibly reckless by these CEOs.
- 22:20SPEAKER_01: And And and Domestis Opposites.
- 22:22SPEAKER_01: Has been one of the more cautious people about timelines,
- 22:26SPEAKER_01: though I will say that even he has sort of,
- 22:28SPEAKER_01: I think,
- 22:28SPEAKER_01: brought his timeline up,
- 22:30SPEAKER_01: though it does not make the those kinds of predictions, I don't think.
- 22:32SPEAKER_00: He doesn't say, you know, that 50% 90 % of entry level jobs.
- 22:36SPEAKER_00: Jobs will be disrupted in five years or by 2030, which is what I think Dario Amadei did say.
- 22:44SPEAKER_00: So that is a distinction between them.
- 22:46SPEAKER_00: And I agree that, you know, I think Dario's projection on that particular point is credible.
- 22:51SPEAKER_00: Um, not, not.
- 22:53SPEAKER_00: I think he's way exaggerating how quickly that changes,
- 22:57SPEAKER_00: even though I am inclined to believe him.
- 22:59SPEAKER_00: When he says that,
- 23:01SPEAKER_00: you know,
- 23:01SPEAKER_00: there'll be recursive self-improvement in terms of frontier model development inside
- 23:05SPEAKER_00: and throng.
- 23:06SPEAKER_00: Anthropic end of 2028.
- 23:08SPEAKER_00: He knows that much better than I do.
- 23:09SPEAKER_00: And I have no reason to Dispute his foreclose.
- 23:12SPEAKER_00: Forecast in terms of the advances in the computer science.
- 23:16SPEAKER_00: It's just the rollout that we're doing.
- 23:19SPEAKER_00: That I don't agree with him on.
- 23:21SPEAKER_00: And I think there are lots of examples.
- 23:22SPEAKER_00: You know,
- 23:24SPEAKER_00: Law is one thing which is projected to be disrupted enormously
- 23:29SPEAKER_00: because you've got these cognitive workers sitting at a screen and
- 23:35SPEAKER_00: that can be automated.
- 23:37SPEAKER_00: But when I look at the...
- 23:39SPEAKER_00: You know.
- 23:40SPEAKER_00: Corporate law firm at which my son is employed in London,
- 23:45SPEAKER_00: They've always had hired 80 Bright young kid.
- 23:47SPEAKER_00: Kids every year, and they've already always gotten rid of, you know, 65 out of the age.
- 23:53SPEAKER_00: A.T. by
- 23:53SPEAKER_00: Bye, six years later.
- 23:55SPEAKER_00: Later.
- 23:55SPEAKER_00: They always overhired.
- 23:56SPEAKER_00: Why?
- 23:57SPEAKER_00: Because,
- 23:58SPEAKER_00: you know,
- 23:58SPEAKER_00: it's nice to be able to pick the best ones,
- 24:00SPEAKER_00: but also it's good for network entrenchment when you kick them out.
- 24:03SPEAKER_00: They go work at your clients,
- 24:05SPEAKER_00: and then when they want business later, They turn around and hide.
- 24:08SPEAKER_00: Hire the law firm where they got trained.
- 24:10SPEAKER_00: So there are all kinds of reasons why people will hire young graduates.
- 24:15SPEAKER_00: And it's not as simple as saying, oh, because there's an AI, you won't hire anymore.
- 24:19SPEAKER_01: From what you've gathered, what sounds more difficult?
- 24:23SPEAKER_01: Building a super intelligence that will not do things we don't want it to do,
- 24:31SPEAKER_01: or is it more difficult to To regulate such a super intelligence,
- 24:36SPEAKER_01: Intelligent,
- 24:36SPEAKER_01: Such a if this is indeed a rapidly evolving,
- 24:39SPEAKER_01: powerful technology,
- 24:42SPEAKER_01: easy to copy,
- 24:43SPEAKER_01: globally distributed, huge economic incentives to keep it moving forward.
- 24:49SPEAKER_01: I guess the phrase that another in another way is that our efforts
- 24:53SPEAKER_01: at regulation basically going to be doomed.
- 24:56SPEAKER_01: Let's talk about your P doom.
- 24:57SPEAKER_01: This is the R doom, the regulation.
- 25:01SPEAKER_00: Yeah.
- 25:02SPEAKER_00: I'm actually quite optimistic in terms of the ability of a government agency
- 25:07SPEAKER_00: to Thank you very much.
- 25:09SPEAKER_00: I know that's perhaps not a fashionable thing to be telling an AEI podcast,
- 25:14SPEAKER_00: but here's my thinking.
- 25:16SPEAKER_00: People often You know, think of AI.
- 25:19SPEAKER_00: AI as a bunch of code that flies around cyberspace and you really can't control it.
- 25:22SPEAKER_00: But actually, It's also...
- 25:24SPEAKER_00: There's also a bunch of data centers, which are huge physical installations.
- 25:28SPEAKER_00: The government knows precisely where they are.
- 25:30SPEAKER_00: They can't be moved or hidden.
- 25:32SPEAKER_00: The owners are known that payment flows can be discovered and you can go as
- 25:36SPEAKER_00: the government,
- 25:36SPEAKER_00: You can go tell those data centers listen you're not going to train
- 25:40SPEAKER_00: the following model And if you do I'm shutting you down there is a choke point
- 25:45SPEAKER_00: in this technology so please first For example.
- 25:48SPEAKER_00: Get the kind of code flying around cyberspace, Quicksilver, impossible to track down.
- 25:54SPEAKER_00: Just please let's park that myth because actually This is a, there is a choke.
- 25:58SPEAKER_00: Choke point.
- 26:00SPEAKER_00: And then once you understand that there's a choke point,
- 26:02SPEAKER_00: the next thing is don't let models out into the wild without there being
- 26:08SPEAKER_00: a kill switch.
- 26:09SPEAKER_00: And that's another way of saying don't allow open weight models.
- 26:13SPEAKER_00: And to me, it's a no brainer that we're going to stop open weight models.
- 26:17SPEAKER_00: I mean, You know, the Trump...
- 26:18SPEAKER_00: Trump administration has just done a 180 since Mythos appeared on the scene And is controlled.
- 26:24SPEAKER_00: Controlling the release even of proprietary models.
- 26:28SPEAKER_00: So the notion that they would control or even forbid open-weight models,
- 26:33SPEAKER_00: which can be freely adapted by the user and can never be shut down,
- 26:37SPEAKER_00: Of course they're going to do it.
- 26:38SPEAKER_00: Do that.
- 26:40SPEAKER_00: And would China ever do it?
- 26:41SPEAKER_00: Well, there was a Reuters story this week saying that they were thinking about it,
- 26:44SPEAKER_00: and I don't know where they are,
- 26:45SPEAKER_00: I don't know if the Reuters story is correct,
- 26:47SPEAKER_00: but the logic in terms of the direction of travel is surely
- 26:51SPEAKER_00: that China loves controlling stuff.
- 26:53SPEAKER_00: They're absolutely happy to regulate the internet,
- 26:56SPEAKER_00: to tell their teenagers they can only do X amount of video game per day, whatever.
- 27:00SPEAKER_00: They have no problem intervening.
- 27:03SPEAKER_00: And they're going to say to them,
- 27:03SPEAKER_00: themselves,
- 27:04SPEAKER_00: do we want mythos level cyber hacking technology to be Freely circulated.
- 27:09SPEAKER_00: Circulating in our country so people can do cyber attacks?
- 27:11SPEAKER_00: Of course not.
- 27:12SPEAKER_00: They're going to shut it down too.
- 27:14SPEAKER_00: Now, the question is,
- 27:15SPEAKER_00: will they shut it down in terms of their models being used outside of China?
- 27:20SPEAKER_00: They might be quite happy to Does it destabilize the world?
- 27:22SPEAKER_00: West with open weight models,
- 27:24SPEAKER_00: which is why I think we need to talk to China and try and do
- 27:26SPEAKER_00: a non-proliferation deal with them.
- 27:29SPEAKER_00: But I do think...
- 27:31SPEAKER_00: That government fundamentally can regulate data centers,
- 27:36SPEAKER_00: forbid open weight releases, And it's some combination And what sort of food?
- 27:43SPEAKER_00: And drug administration model where before the drug is released,
- 27:46SPEAKER_00: it has to be tested and you can veto the release If it's not a safe trip.
- 27:50SPEAKER_00: Drug.
- 27:50SPEAKER_00: And then fed supervision where after, you know, in honor.
- 27:55SPEAKER_00: On an ongoing basis, you might have government Um, officials.
- 27:59SPEAKER_00: Officials embedded at the major age.
- 28:02SPEAKER_00: AI labs, the major AI using Um, companies.
- 28:05SPEAKER_00: Companies,
- 28:06SPEAKER_00: which have sort of some systemic risk factor
- 28:09SPEAKER_00: because they are running big operating systems on the internet
- 28:12SPEAKER_00: or whatever it might be.
- 28:15SPEAKER_00: You know you'd have ongoing supervision by government agents to try
- 28:18SPEAKER_00: to reduce screw ups now I'm fully aware that you know fed supervision often fails
- 28:24SPEAKER_00: but I think it might better than nothing but what I'm saying is
- 28:28SPEAKER_00: that I think there's too much defeatism
- 28:30SPEAKER_01: I feel like all that's unlikely to happen
- 28:32SPEAKER_01: within the next 18 That's 18 months.
- 28:33SPEAKER_01: Months.
- 28:33SPEAKER_01: And if you listen to some aggressive forecasts,
- 28:36SPEAKER_01: we have 18 months to do everything you just said.
- 28:39SPEAKER_00: Well, you might be right, but let me put it to you this way.
- 28:41SPEAKER_00: You know, in March of...
- 28:43SPEAKER_00: 2026, if you had asked anybody in the Trump administration,
- 28:47SPEAKER_00: are you going to ban the release of a model and then case by case restrict
- 28:53SPEAKER_00: which consumer is allowed to buy it,
- 28:55SPEAKER_00: they would have said, no way, we are absolutely in favor of accelerating AI rollout.
- 29:01SPEAKER_00: By late April, they'd flipped.
- 29:02SPEAKER_00: They did a 180.
- 29:03SPEAKER_00: They did something that they didn't predict That's what they would do.
- 29:05SPEAKER_00: Do.
- 29:05SPEAKER_00: So I think the capacity for Now, the technology.
- 29:08SPEAKER_00: Technology itself to create facts on the ground which oblige government
- 29:13SPEAKER_00: to respond should not be underestimated.
- 29:15SPEAKER_01: Yet I still wonder if all that's, I mean, I think about nuclear Nuclear, non.
- 29:19SPEAKER_01: Non-proliferation, Which, you know...
- 29:21SPEAKER_01: maybe regarded as regarded as success, but yet North Korea still has nuclear bombs.
- 29:26SPEAKER_01: And I think Iran could get a nuclear bomb quite easily.
- 29:29SPEAKER_01: And I'm wondering if in one way you could view that as not successful.
- 29:33SPEAKER_01: And there are certainly a lot of choke points with building a nuclear weapon.
- 29:38SPEAKER_01: And I wonder at the end of the day,
- 29:40SPEAKER_01: North Korea doesn't have its own super intelligent model,
- 29:43SPEAKER_01: just like everybody else does.
- 29:45SPEAKER_00: Yeah, but on the other hand, you know, when was the moment when A.Q.
- 29:48SPEAKER_00: Khan of Pakistan stole the blueprints from Holland or Denmark or whatever it was?
- 29:54SPEAKER_00: I mean, I don't, 1990 or something, right?
- 29:57SPEAKER_00: So there was a an operative
- 29:59SPEAKER_00: non -profit.
- 30:00SPEAKER_00: Non-proliferation regime roughly from 1970 1970 to 1990, yeah.
- 30:04SPEAKER_00: 1990.
- 30:04SPEAKER_00: You know, 20 years is better than nothing.
- 30:07SPEAKER_00: I'm not in favor of giving up before we try.
- 30:10SPEAKER_01: The quote I read at the start, the,
- 30:12SPEAKER_01: you know,
- 30:12SPEAKER_01: at some point in the not distant future,
- 30:14SPEAKER_01: AI will beat human intelligence and the biggest thing since we acquired the capacity
- 30:19SPEAKER_01: for abstract thought.
- 30:20SPEAKER_01: Now, when do you...
- 30:21SPEAKER_01: people question that?
- 30:22SPEAKER_01: And I mentioned a lot of economists question.
- 30:24SPEAKER_01: There are certainly technologists who question that.
- 30:26SPEAKER_01: When do you think that will be undeniably true that that sentence
- 30:29SPEAKER_01: about what this technology is and what it can do We'll be obviously.
- 30:33SPEAKER_01: Obviously true beyond dispute?
- 30:36SPEAKER_00: To be honest, I would say it's already true.
- 30:38SPEAKER_00: I mean, you try using Fable.
- 30:41SPEAKER_00: And if people are listening And,
- 30:42SPEAKER_00: you know,
- 30:42SPEAKER_00: if people are listening and they're inclined not to agree with me,
- 30:46SPEAKER_00: I just ask you, you know, spend And a couple of hours were...
- 30:49SPEAKER_00: with Claude Fable and then see if you disagree with me.
- 30:52SPEAKER_00: I mean, Claude Faber.
- 30:54SPEAKER_00: Fable is an astonishing jump from the capability that existed You know, a month or two ago.
- 30:59SPEAKER_00: Ago.
- 31:00SPEAKER_00: And to me, when I ask it something, I literally put this in.
- 31:03SPEAKER_00: I said, You know, design me.
- 31:06SPEAKER_00: Me a tax reform which seeks to balance three different objectives in
- 31:15SPEAKER_00: the United States, which are You know, a desire to...
- 31:18SPEAKER_00: slow displacement of workers by AI,
- 31:23SPEAKER_00: but not Decrease from frontier innovation in a AI and protect
- 31:29SPEAKER_00: the government's ability to collect taxes in an environment where Individual.
- 31:35SPEAKER_00: Labor, maybe.
- 31:35SPEAKER_00: Be a smaller part of GDP And it produced a very sophisticated policy proposal,
- 31:43SPEAKER_00: way more sophisticated than I could have come up with if I'd had a week and it did
- 31:46SPEAKER_00: in 10 minutes.
- 31:48SPEAKER_00: So I think it is smarter than me by quite a long shot on any topic I ask it about.
- 31:55SPEAKER_01: Let me flip it around and ask you And let's go this way.
- 31:58SPEAKER_01: Way.
- 31:58SPEAKER_01: If looking back seven 7 or 10 years from now.
- 32:00SPEAKER_01: Now, What are the odds that we'll - will view your book rather than one documenting
- 32:05SPEAKER_01: the rise of of super intelligence.
- 32:07SPEAKER_01: Superintelligence,
- 32:07SPEAKER_01: but rather one documenting a mindset and hubris that led to
- 32:12SPEAKER_01: the big biggest investment bubble And maybe let down.
- 32:15SPEAKER_01: Letdown of all time They're, they're...
- 32:18SPEAKER_01: it does not turn out to be like the last invention ever invented?
- 32:21SPEAKER_01: Is there a non-zoom?
- 32:22SPEAKER_01: Non-zero percent chance that will be true?
- 32:25SPEAKER_01: That that is the book you have in fact written, documenting hubris rather than success.
- 32:30SPEAKER_00: ha.
- 32:30SPEAKER_00: Thank you.
- 32:31SPEAKER_00: Well, look,
- 32:31SPEAKER_00: you know,
- 32:32SPEAKER_00: as well as I do,
- 32:32SPEAKER_00: that there are lots of cases where, you know, you have a railway bubble.
- 32:36SPEAKER_00: It doesn't mean that the railways are not a very good thing.
- 32:39SPEAKER_00: Very important technology, right?
- 32:41SPEAKER_00: And so, you know, I'm not saying there could never be a bubble in A .I.'s stocks.
- 32:49SPEAKER_00: I think that OpenAI as an individual company example is where
- 32:54SPEAKER_00: It's way overvalued.
- 32:55SPEAKER_00: Overvalued.
- 32:55SPEAKER_00: For example, I don't think it's got a business model.
- 32:57SPEAKER_00: It's got great technology,
- 32:58SPEAKER_00: but no business model that's going to sustain the valuation that it's engineered
- 33:03SPEAKER_00: for itself.
- 33:04SPEAKER_00: So I'm willing to say that, you know, of course, markets can overshoot.
- 33:09SPEAKER_00: I don't think that negates the forecast about the technology being really deeply transformative.
- 33:14SPEAKER_01: That you believe there is basically a 0% chance that this is not a transformative technology.
- 33:21SPEAKER_01: Whatever the impact on markets and stock prices,
- 33:25SPEAKER_01: setting all that aside,
- 33:27SPEAKER_01: this is a transformative technology and we can begin to plan based
- 33:22SPEAKER_01: And, you know.
- 33:32SPEAKER_01: on.
- 33:32SPEAKER_01: And on that assumption.
- 33:33SPEAKER_01: Assumption.
- 33:34SPEAKER_00: Yeah, I mean,
- 33:35SPEAKER_00: I think a very lowball assessment,
- 33:37SPEAKER_00: like the bottom end of my range for plausible Judgements about
- 33:41SPEAKER_00: about how important this is,
- 33:43SPEAKER_00: would be to say it's,
- 33:44SPEAKER_00: you know, it's like the internet plus the kind of first 10 years of applications.
- 33:48SPEAKER_00: I mean, it's so, which is big.
- 33:52SPEAKER_00: That's the lowest I can imagine being the case.
- 33:56SPEAKER_01: What does a good outcome here look like?
- 33:59SPEAKER_01: I'm very good and Hollywood is very good and People don't like data sets.
- 34:03SPEAKER_01: Centers are very good at pointing to like how this doesn't turn out to be something
- 34:08SPEAKER_01: that will have been worth it and makes our lives better and makes,
- 34:11SPEAKER_01: you know, not just people live in rich countries, but everybody's lives better.
- 34:14SPEAKER_01: What is what is the good outcome look like?
- 34:17SPEAKER_00: Look, I think the most obvious upside This upside is in in science,
- 34:22SPEAKER_00: you know, medical discoveries and material science discoveries.
- 34:27SPEAKER_00: The latter may be leading to breakthroughs in energy generation
- 34:33SPEAKER_00: that fundamentally reduce the speed of climate change.
- 34:40SPEAKER_00: So I think that that's the biggest upside.