Theo Jaffee on New Media, Doomerism, and Forecasting

Transcript:

[00:10–00:19] all right all right welcome back to the super cycle today our guest is theo jaffe who

[00:19–00:24] previously had his own podcast called the theo jaffe podcast and then went to work on the new

[00:24–00:33] media team for for a16z and now is the co-host of mts or monitoring the situation which is one of

[00:33–00:40] of the best tech live streams. It’s probably one of the only tech live streams, but I watch it all

[00:40–00:47] the time. And it’s a part of the main places where I get my AI news, not only from their stream,

[00:47–00:51] but from their newsletter as well, which I definitely recommend subscribing to. And there,

[00:51–00:58] you can find them at mts.now. And so really excited to be able to get to chat with Theo today

[00:58–01:06] about new media and his opinions on AI takeoff and forecasting, et cetera.

[01:06–01:07] So welcome.

[01:07–01:09] Yeah, Eli, thanks for having me on.

[01:09–01:10] You know, I’m 22 years old.

[01:10–01:13] I’m the youngest person in the room in SF most of the time.

[01:13–01:15] Eli is one of the few people that makes me feel old.

[01:16–01:17] Yeah.

[01:17–01:18] Oh, yeah.

[01:18–01:23] And I forgot to mention we’re here at the Overhang Conference in Washington, D.C.,

[01:23–01:28] which is run by, it’s a project of Sparrow Infrastructure,

[01:29–01:32] which is also kind of light cone in which it’s like the manifest crowd,

[01:33–01:35] but sponsored by MNX.

[01:36–01:41] And so really excited to be here and start talking about this.

[01:41–01:44] And so for those who don’t know, can you explain A16Z,

[01:45–01:48] because they’re the sponsor of MTS, what’s their thesis about new media

[01:48–01:51] and why is it important to have new media?

[01:51–02:05] Well, A16Z is an investor in MTS. We were sort of incubated, spun out of the A16Z new media team. A lot of the original MTS team members, including me, were previously on the A16Z new media team.

[02:05–02:23] I think the theory behind new media at A16Z is like for the longest time, trad media, especially trad tech media, stuff like, well, to say nothing of the New York Times and the Wall Street Journal and whatnot, which are generally pretty bad on tech coverage, especially the New York Times.

[02:23–02:33] But even, you know, tech media like The Verge and TechCrunch and Wired have become like extremely negative to, hostile towards technology.

[02:34–02:41] Most of what you read in the Verge Wired Tech Crunch and so on is written by people who kind of hate technology.

[02:41–02:43] And they don’t really hide that.

[02:43–02:52] You know, to them, journalism is about, quote, speaking truth to power, which they interpret as just like being really critical and negative of everything.

[02:52–03:06] And so, you know, E16Z is an investment firm, Andreessen Horowitz, that invests in all kinds of amazing disruptive technologies from, you know, AI to space to biotech to software to like everything you can imagine, really.

[03:07–03:15] And so the idea behind new media is just to create media that is unapologetically positive towards technology.

[03:15–03:26] And not that this is dishonest or anything, like it is just true that technology is like one of the greatest levers that humanity has ever had for improving its own material living conditions.

[03:27–03:35] And so like to have media that is unapologetically pro-tech and also, you know, novel and interesting in other ways is very important.

[03:35–03:59] So while I agree with you that in general, in almost all cases, new innovation and new technologies are a net positive, but I’m wondering if you might feel that like traditional media has swung to bias negative on technology gradually.

[03:59–04:04] gradually and now they hold this bias, it’s something that they’re kind of like, they have

[04:04–04:10] to conform to. So like, even if there’s a single pro-tech writer at New York Times, they have to,

[04:10–04:14] you know, kind of agree with the PAC and say, you know, data centers are bad for the environment,

[04:14–04:18] which is obviously not true. Do you think the same thing could happen kind of in the other

[04:18–04:24] direction where you become so committed to the pro-tech premise that if it looked like, for

[04:24–04:29] example, even if you don’t think it’s the case now, if it looks like there’s a lot more AI

[04:29–04:34] safety incidents in the near future and you might like not want it to cover them honestly do you

[04:34–04:38] think it’s possible that it could become too biased in the other direction yes and like you

[04:38–04:44] do see this with some of the pro tech people on twitter who basically deny that ai could ever

[04:44–04:49] pose a risk of any kind especially an existential risk which is a very different position than ai

[04:49–04:55] as a technology will be immensely net positive i believe ai as technology in expectation will be

[04:55–04:59] immensely that positive but that doesn’t mean that there are absolutely no risks associated

[04:59–05:05] with ai it doesn’t mean that existential risk is complete sci-fi nonsense like it’s definitely a

[05:05–05:12] possibility at the minimum and and like there are people on x who kind of just deny this altogether

[05:12–05:18] they’re they’re just like they’re too accelerationist it it becomes ideological

[05:18–05:22] and it you know messes with their epistemics it messes with their model of the world

[05:22–05:28] like the the correct take here is like technology is good almost all the time

[05:28–05:34] all like almost all the time meaning there are very very very few edge cases where technology

[05:34–05:38] might not be good can you give an example of one of this yeah existential risk from ai like i think

[05:38–05:47] that if ai labs with the current level of if you want to call it safety that’s one thing but

[05:47–05:53] techniques to ensure that AI systems are controllable, steerable, reliable, and interpretable.

[05:53–05:58] If they were to vastly rapidly increase capabilities through recursive self-improvement

[05:58–06:05] without a corresponding increase in our capabilities to control, steer, and interpret

[06:05–06:10] these systems, then you could end up in a very, very bad situation where you have like

[06:10–06:14] immensely powerful technology that we’re not able to adequately steer.

[06:14–06:20] do you do you believe that that that kind of takeoff scenario is the path that we’re on

[06:20–06:26] currently it’s hard to say i think like a lot of ai forecasting sort of assumes that

[06:26–06:31] we’ll just like speed through a takeoff by default i don’t think that’s really the case i think that

[06:31–06:37] there are pretty strong incentives for companies to if they’re about to undergo like ai like full

[06:37–06:43] pretty hard takeoff for personal self-improvement and they have not yet

[06:43–06:51] received enough assurance that the resulting technology will go well for them like I think

[06:51–06:55] the incentive is just to not do that right like OpenAI doesn’t want to blow up the world

[06:55–07:01] even in the most like selfish narrow sense you can’t IPO if you’re dead

[07:01–07:31] Yeah, that’s true.

[07:31–07:37] be like 70% that we like achieve like a $30 trillion market cap or something based on the

[07:37–07:43] expected value for him that might be that might kind of contradict that no I think not I mean I

[07:43–07:48] think in like first of all if that happened like the researchers a lot of the researchers would

[07:48–07:55] just rebel like they would not be okay with kicking off full RSI if they thought there was a 10%

[07:55–08:03] chance of human extinction secondly i think that sam all that wouldn’t do that like i i

[08:03–08:04] You’re saying St. Albany wouldn’t do that?

[08:05–08:06] I think St. Albany wouldn’t do that.

[08:06–08:09] I think that like what opening I did after the hugging face incident,

[08:09–08:14] where they decided to pause their frontier RL training and implement a bunch of

[08:14–08:19] new safeguards and monitoring measures and whatnot was the correct and

[08:19–08:21] responsible thing to do.

[08:21–08:23] The hugging face incident itself, by the way,

[08:23–08:25] it was not a big deal at all.

[08:25–08:26] Like it was a big deal in,

[08:26–08:29] in its implications for the alignment of future systems,

[08:29–08:32] but in terms of what it actually was like opening,

[08:33–08:33] I hacked hugging face.

[08:33–08:36] It was a felony, but you know, no one was hurt.

[08:36–08:38] There was not a lot of monetary damage.

[08:39–08:48] If Sam Allman were like a full maximum accelerationist, he would have said like, oh, well, this, you know, safety incident clearly was just like a little bit of a monitoring failure.

[08:48–08:50] We’re not going to pause our frontier RL training.

[08:50–08:52] We’re not going to invest more resources in alignment.

[08:53–08:53] We’re going to go full speed ahead.

[08:54–08:54] And he didn’t do that.

[08:54–08:57] So I like I reject the premise sort of.

[08:57–09:10] Okay, and do you think, how effective currently do you think we can actually have safeguards in place? What’s your personal opinions on how well these are actually helping?

[09:10–09:17] I think they’re helping pretty substantially. I think that our level of...

[09:18–09:22] of alignment techniques are not sufficient to fully trust

[09:22–09:25] that if we went through a hard takeoff,

[09:25–09:27] things on the other end would be all right.

[09:27–09:32] But I think that like in the vast majority of cases,

[09:33–09:34] models do behave in aligned ways

[09:34–09:37] and in the vast majority of deployments, right?

[09:37–09:39] Like you don’t have to worry about your codex

[09:39–09:42] or your cloud code agent, you know,

[09:42–09:44] hacking into other companies.

[09:44–09:46] The worst it will do is sort of mislead you

[09:46–09:48] about the nature of its work.

[09:48–09:51] It will tell you it has done something that it hasn’t.

[09:51–09:56] But the misalignment of models in normal deployments

[09:57–09:58] is not a huge problem.

[10:01–10:04] Remember, companies are incentivized to spend a lot

[10:04–10:07] making sure that their models are well aligned

[10:07–10:09] in normal deployments.

[10:09–10:11] Most of the serious misalignment incidents

[10:11–10:13] in the hugging face incident, for example,

[10:13–10:15] that we’ve seen were in evals.

[10:16–10:35] And many of the other incidents that we’ve seen between Anthropic, OpenAI, Meta, and so on, aside from the Hugging Face incident, had the common factor of dealing with a cybersecurity firm called Irregular, which makes testing environments for the labs, which did a really shitty job making these testing environments.

[10:35–10:47] And so if you are training your models, you know, to oversimplify, if you’re training your models in faulty testing environments, it will learn to exploit those faults, which is what reward hacking is.

[10:48–10:54] And so you don’t want your models to be trained in sloppy, vibe-coded environments.

[10:55–10:56] And, okay, so if I could,

[10:57–11:01] I want to start talking a little bit about how you think,

[11:01–11:05] just to get back to our prediction market angle here with the super cycle,

[11:05–11:10] how do you think current AI forecasting has been?

[11:10–11:18] maybe AI 2027 and how do you think that forecasting is going to play a role in advising labs and the

[11:18–11:26] government about how to proceed with AI development what do you mean by curt like

[11:26–11:31] like that the forecast like AI forecasters like Dan Cook, Tylo and other people who are making

[11:31–11:37] predictions about how takeoff is good like different takeoff scenarios right yeah so I have a huge

[11:37–11:44] amount of respect for Daniel Povetajlo, Eli Liflund, and the rest of the AI Futures project

[11:44–11:52] team, Thomas Larson, Romeo Dean, and so on, everyone else involved. I think that they

[11:52–12:00] produce some of the most rigorous forecasts of anyone in AI. Like, there are definitely

[12:00–12:06] things to criticize about the way they arrive at their conclusions. But, you know, the critiques

[12:06–12:12] are definitely not that they lack rigor or effort or quantitative chops. Like they’re really,

[12:12–12:18] really good at this stuff. So the AI Futures project has produced three main things. AI 2027,

[12:18–12:24] which is their original scenario forecast of what things might look like if humanity sort of

[12:24–12:32] maximally raced to superintelligence with pretty aggressive estimates for both timelines and takeoff.

[12:32–12:48] Then they produced the AI Futures model, which is their quantitative model of AI takeoff with a whole bunch of different parameters you can check out, like, for example, how much uplift do coding models give to researchers today, for example.

[12:48–12:54] And you can change the parameters in the model and it will produce different estimates for timelines and takeoff.

[12:55–13:00] And then AI 2040, their most recent piece is kind of a mix of a scenario forecast.

[13:00–13:07] I think they explicitly said it’s not really a forecast, but it’s just their target scenario for Planet.

[13:08–13:13] Yeah, it’s a forecast under specific assumptions that they don’t expect to actually happen.

[13:13–13:14] Okay, yeah, that’s fair.

[13:14–13:20] And it’s a combination of a scenario forecast and policy recommendation paper.

[13:22–13:28] And AI 2040 is, I think, one of the best AI policy papers that has been written.

[13:28–13:33] there’s like there’s definitely things to critique about it it’s a little bit too authoritarian for

[13:33–13:38] my taste but like it involves ai actually continuing to get better and diffuse throughout

[13:38–13:44] the economy and produce sort of a slow rolling singularity rather than a lot of ai policy that’s

[13:44–13:50] basically just shut down ai which would be terrible so so i know you personally have a low p doom but

[13:50–13:57] now you’re also giving a lot of credit to the ai futures project which has a very doomer outlook

[13:57–14:02] can you explain that discrepancy between your PDOOM and the PDOOM of the AI Futures Project,

[14:02–14:06] whereas if you think there’s a lot of credibility in their forecasts?

[14:08–14:14] The PDOOM of the different AI Futures Project members varies. I think Daniel Kropotajlo is

[14:14–14:20] like 70%. I think Eli Liflund is like 30%, something like that. I think some of the others

[14:20–14:28] or lower but like i i don’t how do i say this like your estimations of timelines and takeoff

[14:28–14:34] don’t have like a ton to do with pdoom like pdoom is largely about like part of it is like

[14:34–14:39] is the ai capable enough to kill everyone i mean but ultimately their forecasts end with human

[14:39–14:49] extinction like sort of 2027 and 2027 in in in the the takeoff scenario in the race it sounds

[14:49–14:56] like we we’re we are going to be in a race so no i don’t think so why not because like the

[14:56–15:02] incentives aren’t really there like nobody wants to risk it like all of them well i mean i think

[15:02–15:07] it would be kind of unfair to say the incentives aren’t there it’s trillion tens of trillions of

[15:07–15:11] dollars of market cap well there’s tens of trillions of dollars of market cap that you can

[15:11–15:17] get with a slower takeoff that will lower debune yeah and of course there are all kinds of exogenous

[15:17–15:23] political pressures as well, which we’re starting to see now. Can you explain a scenario in which

[15:23–15:31] there’s a bilateral slowdown between the U.S. and China, where we can verify that China would

[15:31–15:35] actually be slowing down and how that would actually lead to a lower P2? Because I kind of

[15:35–15:41] just don’t see that. Well, they describe such a scenario in AI 2040, which is the U.S. and China,

[15:41–15:47] both sort of spooked by AI capabilities and the potential for very rapid RSI decide to make a

[15:47–15:55] bilateral agreement that’s like both leaders understand RSI would be really bad. If one of us

[15:55–16:00] decides to go full RSI, we risk like completely losing control rather than gaining unilateral

[16:00–16:05] dominance over the other. I think this is kind of where things are headed. Like I think

[16:05–16:12] politicians on both sides of the Pacific are increasingly worried about what would happen if

[16:12–16:21] we did very rapid RSI. And of course, like, we’re going to go leaders for a second, like so many

[16:21–16:26] American politicians, like American society in general is pretty primed to reject new technology.

[16:26–16:31] Like, look at how difficult it’s been to get like Waymo’s on the road in blue cities like New York.

[16:31–16:37] yeah yeah it’s like and a waymo is like one of the most ev positive technologies you can imagine

[16:37–16:43] i was saying you can you can go to prison for for murdering someone but mom donnie can’t go to

[16:43–16:48] prison for letting all those car taxi deaths happen when you get hit by a taxi exactly like

[16:48–16:54] waymo is literally like a policy button that you can press to not have tens of thousands of people

[16:54–17:00] die every year and yet like the u.s finds it so difficult to do that so if it’s if there’s there’s

[17:00–17:05] much political pressure against these like very clearly ev positive technologies of course there’s

[17:05–17:10] going to be massive political backlash against the technology whose ev is much more unclear which is

[17:10–17:18] you know artificial super intelligence right but i would say there’s like for for waymo it’s kind

[17:18–17:24] of like a smaller a smaller thing that’s like okay there’s like these one or two kind of

[17:24–17:30] There’s like a Google, it was Waymo, and then Tesla, who are trying to put the robo-taxis on the road.

[17:30–17:40] But then AI is kind of like a totally different reference class with much more pressures on all sides and much more financial incentives going around to accelerate.

[17:41–17:45] But I mean, you do have a point, but that would just be my pushback.

[17:45–17:51] The financial incentives don’t push towards maximum acceleration if you believe in existential risk, basically.

[17:51–17:59] right but the ev that doesn’t mean the ev wouldn’t be positive necessarily for the ai lab

[17:59–18:06] leaders well like open ai for example which has been criticized many times for not caring about

[18:06–18:11] safety like has slowed down in costly ways like pausing their for trl training you don’t think

[18:11–18:17] it’s marketing no why not why would it be marketing be because they want to show because

[18:17–18:21] because there’s gonna be so much backlash, as you were saying, they want to look like they’re slowed down.

[18:21–18:23] there are ways that they could-

[18:23–18:30] could signal slowdowns in non-costly ways that don’t actually involve slowdowns like they could

[18:30–18:35] you know put out some paper claiming like they’re going to increase safety with like very vague

[18:35–18:39] details then just not do it but like pausing frontier training runs is like very costly

[18:39–18:45] like a lot is riding on okis ability to produce better models that help create more economic

[18:45–18:50] value that help them pay back all their compute commitments for example okay yeah but like like

[18:50–18:57] You’ve got to remember, the people who are launching these labs really believe in, they’re really AI-pilled.

[18:57–18:58] They’re really AI-pilled.

[18:58–18:59] They believe in the singularity.

[18:59–19:00] They believe in existential risk.

[19:00–19:11] If you look at Sam Ullman’s blog post from 2015, before he started OpenAI, he was very clear that he thinks that the largest existential threat to humanity over the coming decades is artificial intelligence.

[19:11–19:14] Obviously, it’s also the largest opportunity for humanity.

[19:14–19:21] but these are not you know total mercenary businessmen who don’t understand the potential

[19:21–19:28] of the risk of AI and Dario even more so what do you think I I watched your your podcast

[19:28–19:33] with Liron Shapira and one of the things you asked him which I actually kind of argued when I

[19:33–19:39] was on his podcast in debates is that there’s there’s little rigor in in like just throwing

[19:39–19:46] out a p doom totally and do you think that it kind of lowers credibility for the doomers that

[19:46–19:57] many of them do that yes basically like i the the differences between what causes you know

[19:57–20:03] evan hubbinger from anthropic to arrive at 10 or greater than 10 p doom versus daniel

[20:03–20:05] to arrive at 70% PDOOM.

[20:06–20:09] I don’t understand how you would actually

[20:09–20:12] lay out those cruxes in kind of

[20:12–20:15] a useful, rigorous, and quantitative way.

[20:15–20:18] I definitely think it detracts from the credibility of the Doomers

[20:18–20:21] to some extent, and all the more so if they

[20:21–20:24] then make up a PDOOM number and then use it

[20:24–20:27] as a parameter in some

[20:27–20:28] EV calculation equation.

[20:30–20:32] That’s how you get these Pascal’s mugging

[20:32–20:38] kind of things where it’s like oh you know even if pdoom is like five percent it’s like the worst

[20:38–20:45] thing in the universe and so we just slow down around like basically if you’re going to like

[20:45–20:50] come up with policy recommendations based on like quantitative measured quantitative estimates of

[20:50–20:55] things you should ensure that those quantitative estimates are pretty rigorous and like it’s much

[20:55–21:01] easier to do rigorous forecasting of ai timelines of ai takeoff because like it’s it’s a lot of

[21:01–21:07] extrapolation from stuff we’ve seen in the past, like the rate of growth and effective compute,

[21:07–21:13] for example, the rate of algorithmic improvements, the rate of progress in Moore’s law, you know,

[21:13–21:18] transistor density per chip or performance per watt. Like these, these are all things that are

[21:18–21:23] much easier to forecast. And like, if we do have artificial super intelligence, like,

[21:23–21:29] would it want to kill us? And if so, like, how would it do it? And if so, like, how long would

[21:29–21:35] to take and so on and so on these are all very like hard to answer questions yeah i think i i

[21:35–21:40] kind of agreed with with lear on at the end of the podcast that while i wasn’t going to form a

[21:40–21:46] specific pdoom estimate i think what he calls the sane zone a 10 to 90 pdoom sounds

[21:46–21:53] pretty reasonable to me yeah i don’t know 90 sounds pretty unreasonable unless we’re talking

[21:53–22:00] about very long time presence but like 90% PDOOM by the end of the decade seems like implausible

[22:00–22:06] yeah it’s it’s possible I just think there’s like way too much uncertainty to to pin down any sort of

[22:06–22:12] range it makes sense yeah and like you know into the medium term future like I think PDOOM

[22:12–22:19] under 10% totally makes sense maybe I don’t know I might push back a bit there but what do

[22:19–22:27] you think about, especially over the past week or two weeks, like the big push into the public,

[22:27–22:33] into the mainstream that the doomers have had, what do you think about how that’s going to impact

[22:33–22:39] the discourse among the general public? It’s going to make it suck.

[22:39–22:48] like i hmm you know i i am concerned about ai risks i’m not even sure that raising public

[22:48–22:52] salience about these things broadly is the best way to address them like i think it’s just going

[22:52–22:57] to end up with massive backlash basically butlerian jihad against the technology and it’s

[22:57–23:02] going to suck it’s like we already have like bernie sanders saying we’re going to do a

[23:02–23:06] moratorium on data centers and we’re going to pause all ai training and we’re going to permanently

[23:06–23:11] banned superintelligence, where superintelligence is defined as, I’m going to mess up the exact

[23:11–23:13] wording, but something like...

[23:13–23:15] Yeah, so what was I saying?

[23:15–23:17] AI discourse is gonna suck.

[23:17–23:19] Oh yeah, you were saying what Bernie- AI discourse is gonna suck.

[23:19–23:21] Yeah. Super intelligence,

[23:21–23:23] how Bernie defines it. How Bernie defines-

[23:23–23:25] Yeah, it’s just like,

[23:25–23:28] I tweeted the other day, this is the least stupid AI discourse we’ll ever be.

[23:28–23:40] I think that, you know, those of us who have been in this world for some time around a lot of SF people, lab people, forecasting people, rationalists, very rigorous sort of world models and probability estimations.

[23:40–23:56] Like all of this is just going to get swept away and AI is going to become just another political topic with, you know, dumb people saying stupid things, mostly for the sake of their own aura farming rather than, you know, genuine desire to arrive at a sort of shared truth.

[23:57–24:01] It’s like, it’s going to be so bad. It already has become much worse than it was.

[24:02–24:07] Like, I think that the epistemics around this stuff are just going to like kind of get poisoned.

[24:07–24:16] and you know like this is part of the vision of mts actually is like we’re trying to help fix

[24:16–24:20] the discourse we’re trying to help steer the discourse in a positive direction that helps

[24:20–24:28] people understand the truth about things that are going on yeah yeah that makes total sense

[24:28–24:46] I saw a Politico poll a couple days ago that showed 17% of Americans think that it’s almost certain, like their PD is essentially, they say, almost certain that AI is going to kill everyone, which seems kind of extreme to me.

[24:46–24:47] Well, it’s extreme.

[24:47–24:53] Even the most doom-pilled of doomers wouldn’t say it’s almost certain.

[24:53–24:57] Well, the most doom-pilled of doomers, like Ginkowski would say 99%.

[24:57–24:59] I think you’d count it like 90%.

[24:59–25:01] Of the quantitatively rigorous...

[25:01–25:05] 75th percentile of rationalists wouldn’t say almost certain.

[25:05–25:08] Yeah, I mean, of the quantitatively rigorous kind of doomers,

[25:08–25:12] like Daniel Kovatajla, his PDO is 70%, which is not almost certain.

[25:13–25:15] But regardless, it’s also silly because...

[25:15–25:17] You say Yudkowsky is not quantitatively rigorous?

[25:18–25:19] Not really.

[25:20–25:21] But what are your thoughts on it?

[25:21–25:26] I think Yudkowsky’s beliefs on this are sort of more religious.

[25:27–25:33] he, you know, he has this very like kind of elegant theory of the way intelligence works.

[25:33–25:40] you know, like me, Yuri has had a lot of like very, like thoughtful writing on this

[25:40–25:45] that I think is just like wrong. I think it’s kind of pretty paradigmatic, so to speak. Like,

[25:45–25:51] I don’t think that, building AI is akin to like summoning alien intelligence, unless you like

[25:51–25:57] do it really really wrong i don’t think that like i think that yudkowsky was just wrong that

[25:57–26:02] there’s no fire alarm before agi we just had a fire alarm before agi it was a hugging face incident

[26:02–26:07] like there will be lots of other fire alarms i think that he was wrong that we have like one

[26:07–26:13] critical try to get alignment right and if we don’t then it kills everyone like a lot of these

[26:13–26:20] assumptions only make sense in like an extremely foony world that we do not appear to be in

[26:20–26:31] In reality, the world that we live in now is we have a lot of different actors with a lot of sort of AGI-ish things that are sort of mostly aligned.

[26:32–26:38] Obviously, the situation can still go quite wrong from here, but it looks just much better than what Yukowsky imagined 20 years ago.

[26:40–26:42] Yeah, I think that makes sense.

[26:42–26:49] And I think I want to go to a 415 session, so I think we can stop here.

[26:49–26:52] but Theo, thanks so much for, for doing the, the interview.

[26:53–26:58] Absolutely. yeah. So, go watch, go watch MTS. It’s good.

[26:58–27:02] X.com slash MTS live. Yeah. Thanks for having me on. It’s good to be a guest.

[27:03–27:06] I’m so used to being a host. Yeah. Yeah. All right. all right.

Get full access to The Supercycle at supercycle.blog/subscribe