Latest / Politix / The Stunt For Red October

Transcript

The Stunt For Red October · Politix

• 0:00
  SPEAKER_02: Jeffries had indicated that this is going to happen in some form.
• 0:05
  SPEAKER_02: And, you know, we're waiting.
• 0:07
  SPEAKER_02: I mean, it sounds stupid to characterize like Democrats release a policy agenda as
• 0:13
  SPEAKER_02: an October surprise.
• 0:15
  SPEAKER_01: Hey,
• 0:21
  SPEAKER_01: everyone, you're listening to a free preview of the Politics Podcast.
• 0:25
  SPEAKER_01: We recorded last week's episode on AI risk before the tech industry got rocked
• 0:30
  SPEAKER_01: by insider warnings about the dangers of runaway AI development,
• 0:34
  SPEAKER_01: and before Dario Amadei,
• 0:36
  SPEAKER_01: the CEO of Anthropic, proposed a joint phased plan to slow development.
• 0:41
  SPEAKER_01: So we'll circle back to discuss where things stand as we record this week,
• 0:46
  SPEAKER_01: where they seem to be headed,
• 0:47
  SPEAKER_01: and whether it's satisfactory progress given the scale and possible imminence of
• 0:52
  SPEAKER_01: the damage AI technology can do.
• 0:56
  SPEAKER_01: Then, if AI isn't this election's October surprise, what will be?
• 1:00
  SPEAKER_01: Donald Trump is a relentless crook and the night is dark and full of terrors.
• 1:05
  SPEAKER_01: Polls look extremely good for Democrats at the moment,
• 1:07
  SPEAKER_01: but we'll scan the horizon for things that might upend the midterms.
• 1:12
  SPEAKER_01: If you're watching on YouTube, hit like, subscribe to the feed.
• 1:16
  SPEAKER_01: Leave some comments, increase our engagement, help us find new viewers and listeners.
• 1:21
  SPEAKER_01: If you're an old school podcast listener and want to hear the whole conversation,
• 1:25
  SPEAKER_01: upgrade to a paid subscription at politics.fm.
• 1:32
  SPEAKER_01: Hello and welcome to the Politics Podcast.
• 1:33
  SPEAKER_01: I'm Brian Boyler.
• 1:35
  SPEAKER_01: I'm Matthew Glacius.
• 1:37
  SPEAKER_01: So...
• 1:38
  SPEAKER_01: When we finished recording last week, the episode was in the can.
• 1:42
  SPEAKER_01: I went about my business like usual.
• 1:44
  SPEAKER_01: I drafted the landing page.
• 1:47
  SPEAKER_01: We published.
• 1:50
  SPEAKER_01: But I think that night before the episode went live was when Jacob Coxon,
• 1:56
  SPEAKER_01: the AI researcher from Anthropic,
• 1:59
  SPEAKER_01: resigned with a big dire warning about how dangerous
• 2:03
  SPEAKER_01: the technology he's building is becoming and how slow the progress in
• 2:07
  SPEAKER_01: the industry is on,
• 2:09
  SPEAKER_01: Sort of coding in fail-safes to prevent AI agents from doing bad things.
• 2:16
  SPEAKER_01: And that has kicked off a whole new set of debates and discourse and recriminations
• 2:22
  SPEAKER_01: about what should happen to the AI industry.
• 2:25
  SPEAKER_01: It was such perfect timing that I texted Matt to be like,
• 2:29
  SPEAKER_01: were you kind of like keyed in by the By the people in Silicon Valley
• 2:36
  SPEAKER_01: that this was coming.
• 2:37
  SPEAKER_01: Because it was like, this was Matt's idea to have last week's episode be about AI.
• 2:42
  SPEAKER_01: And he insists that he was not part of any op to get people scared of AI.
• 2:48
  SPEAKER_01: There's no op.
• 2:49
  SPEAKER_01: I have my doubts.
• 2:50
  SPEAKER_02: I will tell the conspiracy theorists that while I was gone for the summer,
• 2:53
  SPEAKER_02: all these AI safety orgs moved into the co-working space where Brian and I work.
• 3:01
  SPEAKER_02: So...
• 3:02
  SPEAKER_02: Something happened.
• 3:04
  SPEAKER_02: They're around, you know, we talk to them.
• 3:07
  SPEAKER_02: I try to stay.
• 3:07
  SPEAKER_01: Do you?
• 3:08
  SPEAKER_01: Well, I mean,
• 3:08
  SPEAKER_01: that is interesting,
• 3:09
  SPEAKER_01: but I kind of thought that the idea was like,
• 3:12
  SPEAKER_01: there's a lot of unused labor in Washington,
• 3:15
  SPEAKER_01: thanks to Elon Musk and a lot of people who have been worried about AI for
• 3:19
  SPEAKER_01: a long time.
• 3:20
  SPEAKER_01: And some of them have a lot of money.
• 3:21
  SPEAKER_01: And so things came together and that it might not necessarily have had Anything
• 3:27
  SPEAKER_01: to do with any like insider knowledge about how weird things were getting.
• 3:31
  SPEAKER_01: And then separately you have,
• 3:33
  SPEAKER_01: we'll talk about it probably,
• 3:34
  SPEAKER_01: but the hugging bear attack that really seems to have got people to realize,
• 3:38
  SPEAKER_01: oh, like the things we theorized about really could happen in the real world.
• 3:42
  SPEAKER_01: And that's why you had the-
• 3:48
  SPEAKER_02: What's an op, what's not an op, et cetera.
• 3:50
  SPEAKER_01: I don't think there's any op, to be honest, to be clear.
• 3:53
  SPEAKER_02: But I mean,
• 3:55
  SPEAKER_02: the thing that is most op-like is that if you went back not that long ago,
• 4:02
  SPEAKER_02: a couple of years ago,
• 4:03
  SPEAKER_02: I think the mainstream view among people concerned about AI safety is Was
• 4:10
  SPEAKER_02: that making political progress on safety oriented AI regulation was
• 4:17
  SPEAKER_01: very unlikely.
• 4:19
  SPEAKER_02: And so they wanted to invest their efforts in technical safety work.
• 4:24
  SPEAKER_02: That there are a number of safety-minded executives.
• 4:28
  SPEAKER_02: People disagree about how sincere anybody is.
• 4:32
  SPEAKER_02: But there are these alignment teams at OpenAI and Anthropic and to a lesser extent
• 4:37
  SPEAKER_02: at Google and a lesser than that extent at Meta.
• 4:41
  SPEAKER_02: And so the main thing that people interested in AI safety were going
• 4:45
  SPEAKER_02: to do is encourage technically-minded people to work at AI companies and work
• 4:51
  SPEAKER_02: on addressing things These different problems about monitoring,
• 4:56
  SPEAKER_02: about interpretability, et cetera, et cetera, et cetera.
• 4:59
  SPEAKER_02: And there has been a shift, a vibe shift to the view that this is not working, right?
• 5:08
  SPEAKER_02: That the pace of development of the core models,
• 5:13
  SPEAKER_02: that they are growing more capable faster than we are being able to And that is a,
• 5:19
  SPEAKER_02: you know, um...
• 5:32
  SPEAKER_02: The public first super PAC,
• 5:34
  SPEAKER_02: the political campaigns of Alex Boris,
• 5:37
  SPEAKER_02: who lost his primary,
• 5:38
  SPEAKER_02: Manny Routinel,
• 5:39
  SPEAKER_02: who won his primary,
• 5:40
  SPEAKER_02: Scott Wiener,
• 5:41
  SPEAKER_02: who is still TBD in the election,
• 5:44
  SPEAKER_02: and to a lot of AI organizations that had existed in a rudimentary form in
• 5:49
  SPEAKER_02: the Bay Area, opening DC offices, new DC institutions springing up.
• 5:55
  SPEAKER_02: So that had been happening for the past several months.
• 5:58
  SPEAKER_02: And then after Hugging Face...
• 6:02
  SPEAKER_02: We're seeing a preference cascade,
• 6:05
  SPEAKER_02: you know,
• 6:05
  SPEAKER_02: just like more and more people talking about this topic,
• 6:09
  SPEAKER_02: saying that something should be done.
• 6:10
  SPEAKER_02: You know,
• 6:10
  SPEAKER_02: this reached its apogee over the weekend when you had Jaira Amadei,
• 6:15
  SPEAKER_02: Sam Altman, and Elon Musk, who all hate each other.
• 6:20
  SPEAKER_02: And at one time, we're all collaborating together together.
• 6:24
  SPEAKER_02: On the idea that they needed to make AI that was safe for humanity.
• 6:27
  SPEAKER_02: And then they had a series of falling outs with each other and they became rivals
• 6:31
  SPEAKER_02: to say,
• 6:32
  SPEAKER_02: we think that there needs to be a cooperative agreement among
• 6:36
  SPEAKER_02: the leading AI companies to do what they call pacing the frontier,
• 6:42
  SPEAKER_02: which is to say just slowing down the rate at which this goes forward.
• 6:46
  SPEAKER_02: And it looked for a day or so to me like this might happen.
• 6:53
  SPEAKER_02: These guys are ferocious business rivals.
• 6:58
  SPEAKER_02: Elon is obviously known to be a conservative Republican.
• 7:03
  SPEAKER_02: Amode is involved in some broader effective altruism circles that are more on the left.
• 7:11
  SPEAKER_02: The companies were calling to regulate themselves in a way
• 7:16
  SPEAKER_02: that it appeared was not strongly contrary to the interests
• 7:20
  SPEAKER_02: of any other major business undertaking.
• 7:25
  SPEAKER_02: Antitrust people were raising a concern about this that we should talk about.
• 7:31
  SPEAKER_02: But like left-wing antitrust theory does not have a ton of information.
• 7:36
  SPEAKER_02: Yeah.
• 7:57
  SPEAKER_02: But then Trump on Monday did a like phone a friend with Justin Huang of Nvidia
• 8:06
  SPEAKER_02: and said he thinks all this doom,
• 8:09
  SPEAKER_02: all this AI risk stuff is dumb,
• 8:11
  SPEAKER_02: that all All you really need to keep AI safe is a strong
• 8:15
  SPEAKER_02: and powerful president like Donald Trump.
• 8:18
  SPEAKER_02: The Department of Defense,
• 8:20
  SPEAKER_02: which has named itself the Department of War,
• 8:23
  SPEAKER_02: put out a preposterous social media graphic about how, like,
• 8:27
  SPEAKER_02: they will conquer effective altruism and we need to eliminate it in all
• 8:32
  SPEAKER_02: of its forms.
• 8:33
  SPEAKER_02: And this strong effort that had been done over the weekend to have coxswain Go
• 8:41
  SPEAKER_02: on Fox News,
• 8:42
  SPEAKER_02: to have Nate Soros go on Tucker Carlson's show,
• 8:48
  SPEAKER_02: to do interviews about this with right-leaning media,
• 8:51
  SPEAKER_02: because there was an understanding that Republicans are both less likely to want
• 8:55
  SPEAKER_02: to regulate AI,
• 8:56
  SPEAKER_02: and also that Republicans currently are the only people who have the authority
• 9:00
  SPEAKER_02: to regulate AI.
• 9:01
  SPEAKER_02: This big push to kind of incept AI safety on the right failed, and now we're left with...
• 9:08
  SPEAKER_02: I don't know what, you know, politics, right?
• 9:11
  SPEAKER_02: Like, like normal politics.
• 9:12
  SPEAKER_01: So I have a couple of questions about, um, I think this is a good summary.
• 9:16
  SPEAKER_01: Uh, I have a couple of questions about some of like the, uh,
• 9:20
  SPEAKER_01: the pre the antecedent activities that you said at the beginning,
• 9:23
  SPEAKER_01: but I don't, I,
• 9:24
  SPEAKER_01: and I agree that Donald Trump definitely just revealed himself to be like a,
• 9:28
  SPEAKER_01: a sabotage agent and all this.
• 9:30
  SPEAKER_01: Like, even if,
• 9:31
  SPEAKER_01: even if everyone at the frontier got together and started having conversations,
• 9:35
  SPEAKER_01: he would try to sabotage that.
• 9:36
  SPEAKER_01: Right.
• 9:37
  SPEAKER_01: Um, But I'm not as sanguine as you are about how far along things got until Monday, right?
• 9:44
  SPEAKER_01: Because Dario Amadei's statement listed a number of things that could be done
• 9:50
  SPEAKER_01: to slow the frontier,
• 9:52
  SPEAKER_01: as they say,
• 9:53
  SPEAKER_01: and that Anthropic would voluntarily on its own do the first three, no matter what.
• 9:59
  SPEAKER_01: And I think Altman agreed to do at least the first one.
• 10:03
  SPEAKER_01: Yeah.
• 10:04
  SPEAKER_02: Yeah.
• 10:04
  SPEAKER_01: So already you have like not quite a meeting of the minds.
• 10:08
  SPEAKER_01: And then all Elon said was Dario is right, but he made no commitment to anything.
• 10:14
  SPEAKER_01: And you heard nothing from the other labs.
• 10:16
  SPEAKER_01: So you were far off from this kind of
• 10:22
  SPEAKER_01: moment like when Einstein – Writes a letter
• 10:26
  SPEAKER_01: to FDR saying, hey,
• 10:27
  SPEAKER_01: we need to do a government nuclear program because otherwise everything's going
• 10:34
  SPEAKER_01: to turn out bad.
• 10:38
  SPEAKER_01: And so that's part of the reason why I'm like...
• 10:43
  SPEAKER_01: I think if everyone was on board and then Donald Trump tried to stop them,
• 10:47
  SPEAKER_01: you could have like a pretty pairwise fight about it.
• 10:50
  SPEAKER_01: But I think now anyone who thinks that they can still win the race without
• 10:55
  SPEAKER_01: and like will take that,
• 10:57
  SPEAKER_01: you know, 90 percent chance that they don't wreck all humanity.
• 10:59
  SPEAKER_01: They have a wide open door to run through.
• 11:01
  SPEAKER_02: Yeah.
• 11:02
  SPEAKER_01: So that's a problem.
• 11:03
  SPEAKER_01: But what I wanted to ask was a couple of things.
• 11:06
  SPEAKER_01: One is.
• 11:09
  SPEAKER_01: I mean, do you really think that these guys believed that the safety stuff on
• 11:15
  SPEAKER_01: the technical side could have kept pace with frontier development,
• 11:21
  SPEAKER_01: given that all of the incentives are to build capability and none of them are
• 11:26
  SPEAKER_01: to build safety?
• 11:28
  SPEAKER_02: Well, that I don't really know because the people who I know, I mean, the...
• 11:34
  SPEAKER_02: The people who talk to me about AI safety have always been people who are interested
• 11:40
  SPEAKER_02: in politics and policy, and that's why they want to talk to me.
• 11:45
  SPEAKER_02: I don't know anything about technical artificial intelligence,
• 11:48
  SPEAKER_02: and I know a fair amount about American politics and public policy.
• 11:51
  SPEAKER_02: So I have always heard from people who are like,
• 11:55
  SPEAKER_02: this technical safety approach is not going to work.
• 11:58
  SPEAKER_02: And like, we need to invest in politics.
• 12:00
  SPEAKER_02: All I can say is that they were losing the internal argument about like,
• 12:07
  SPEAKER_02: what should the movement prioritize until relatively recently,
• 12:11
  SPEAKER_02: at which point they started getting more money.
• 12:17
  SPEAKER_01: That makes sense to me.
• 12:18
  SPEAKER_01: I feel like the people sort of higher up in the executive realm to the extent
• 12:23
  SPEAKER_01: that they were saying, hey, we'll just do these things in tandem.
• 12:26
  SPEAKER_01: We're Probably smart enough to realize that that was bullshit because the whole idea
• 12:33
  SPEAKER_01: of a cash incentivized race is that if you take resources away from winning
• 12:38
  SPEAKER_01: that race to put them into slowing your own self down, you're going to lose.
• 12:41
  SPEAKER_01: So it just doesn't work as like a theory, right?
• 12:47
  SPEAKER_01: And so I don't know,
• 12:50
  SPEAKER_01: like it is to me another reason to doubt that the People who maybe are trying
• 12:59
  SPEAKER_01: to forge a meeting of the minds over AI safety really mean it.
• 13:04
  SPEAKER_01: But the other part that like,
• 13:06
  SPEAKER_01: and maybe you're the wrong person to ask
• 13:08
  SPEAKER_01: because I also don't have AI technical expertise,
• 13:12
  SPEAKER_01: but you alluded to this concept called alignment.
• 13:15
  SPEAKER_01: And it is their jargon for the idea that AI agents can be essentially programmed
• 13:24
  SPEAKER_01: to be pro-social.
• 13:26
  SPEAKER_01: When you give them a task or a goal or a problem to solve,
• 13:31
  SPEAKER_01: they don't go about doing it in ways that cause harm to society, right?
• 13:38
  SPEAKER_00: And again, don't really know what I'm talking about here.
• 13:45
  SPEAKER_00: But it seems like if you're talking about building agents that are powerful enough to hack...
• 13:52
  SPEAKER_01: Big entities or make viruses or whatever,
• 13:57
  SPEAKER_01: that you need that technical thing to be done perfectly.
• 14:02
  SPEAKER_01: The first ever perfect unhackable system, safety system ever in the world.
• 14:09
  SPEAKER_01: And if it fails once, that might be game up.
• 14:14
  SPEAKER_01: Yeah, I mean, that's the... And that does not sound like a real idea to me.
• 14:17
  SPEAKER_01: That sounds like... That is the...
• 14:21
  SPEAKER_02: You know, there are multiple strands of thought in this space.
• 14:27
  SPEAKER_02: And the argument you're making there,
• 14:29
  SPEAKER_02: this is essentially the Eliezer Yukowski, Nick Soar's argument.
• 14:34
  SPEAKER_02: Their book is called, If Anyone Builds It, Everyone Dies.
• 14:37
  SPEAKER_02: Everyone Dies, yeah.
• 14:38
  SPEAKER_02: So this is a,
• 14:40
  SPEAKER_02: Eliezer was probably like way before any of this was happening,
• 14:47
  SPEAKER_02: was like the first thing.
• 14:50
  SPEAKER_02: Technical alignment bull.
• 14:52
  SPEAKER_02: And then he talked himself out of that,
• 14:55
  SPEAKER_02: you know, again,
• 14:55
  SPEAKER_02: before any of this happened with,
• 14:58
  SPEAKER_02: I would say essentially the argument that you just made there,
• 15:01
  SPEAKER_02: that the problem here is that this is a one-shot question
• 15:07
  SPEAKER_02: and you don't have any opportunity to do the work that almost all science
• 15:12
  SPEAKER_02: and engineering proceeds by trial and error, right?
• 15:16
  SPEAKER_02: Like you don't,
• 15:18
  SPEAKER_02: The Wright brothers didn't sit at their desk and design a 747 and then build it.
• 15:28
  SPEAKER_02: You have to do stuff and it typically doesn't work.
• 15:32
  SPEAKER_02: And the problem on this view is that you can't test alignment, really.
• 15:40
  SPEAKER_02: You only get one shot, right?
• 15:43
  SPEAKER_02: And yeah, so that has always been the case for just don't do it, right?
• 15:48
  SPEAKER_02: And I think a question that everybody should ask about the
• 15:57
  SPEAKER_02: sort
• 15:57
  SPEAKER_02: of moderate safetyist view is like,
• 16:03
  SPEAKER_02: To what extent is that the real view of the people involved or are they just trying
• 16:10
  SPEAKER_02: to navigate these various political economy and diplomatic and other sorts
• 16:17
  SPEAKER_02: of things, concerns, right?
• 16:19
  SPEAKER_02: Because it is true that if,
• 16:24
  SPEAKER_02: you know,
• 16:25
  SPEAKER_02: the 20% of like most safety concerned people all just went and quit tomorrow,
• 16:32
  SPEAKER_02: That wouldn't like achieve anything, right?
• 16:36
  SPEAKER_02: Or if you just said like,
• 16:37
  SPEAKER_02: there should be a law that bans AI, like there wouldn't be a law like that.
• 16:44
  SPEAKER_02: The way regulation gets done is usually quite a bit more stepwise than that
• 16:50
  SPEAKER_02: and involves like creating agencies and blah, blah, blah, blah, blah.
• 16:53
  SPEAKER_02: And so I just don't know, you know?
• 16:57
  SPEAKER_01: I mean, I think you can see like,
• 17:02
  SPEAKER_01: You can draw the clear inference that the people leading the development at
• 17:07
  SPEAKER_01: the executive level don't believe that if anyone builds it, everyone dies.
• 17:12
  SPEAKER_01: Because if they did,
• 17:14
  SPEAKER_01: they'd all stop because a trillion dollars doesn't mean anything if the day
• 17:18
  SPEAKER_01: after you earn it, the robot kills you, right?
• 17:21
  SPEAKER_01: Sure, yes.
• 17:24
  SPEAKER_01: But then this is actually the one thing That like, I am,
• 17:30
  SPEAKER_01: if you can't tell,
• 17:31
  SPEAKER_01: like quickly radicalizing myself into a like super Bernie Sanders position and
• 17:36
  SPEAKER_01: that you should essentially try to
• 17:39
  SPEAKER_00: ban AI.
• 17:43
  SPEAKER_00: But it like,
• 17:46
  SPEAKER_00: as like,
• 17:48
  SPEAKER_00: you know,
• 17:48
  SPEAKER_00: as crazy as I think some of these guys are,
• 17:53
  SPEAKER_01: if they had like,
• 17:58
  SPEAKER_01: Genuinely,
• 17:58
  SPEAKER_01: like 10% concern that this would be out of control and
• 18:02
  SPEAKER_01: then Earth would not be hospitable to humans anymore.
• 18:06
  SPEAKER_01: I think they'd stop just out of sheer selfish self-preservation, right?
• 18:12
  SPEAKER_01: At the same time,
• 18:14
  SPEAKER_01: nothing motivates positive thinking like the lure of a trillion dollars, I guess.
• 18:20
  SPEAKER_01: And so, I mean, this is just one aspect of...
• 18:25
  SPEAKER_01: This problem that we're all observing from the outside that makes it really hard
• 18:29
  SPEAKER_01: to know what's what.
• 18:31
  SPEAKER_01: And given that,
• 18:33
  SPEAKER_01: and then also given who Trump is and how he operates and how much power he has
• 18:37
  SPEAKER_01: and how much impunity Republicans have given to him and how much they've shown
• 18:42
  SPEAKER_01: that they will let him ride,
• 18:45
  SPEAKER_01: you know, really to the edge of Armageddon, that like the...
• 18:52
  SPEAKER_01: Intelligent solutions that we might want to play around with by constructing
• 18:56
  SPEAKER_01: a regulatory agency, by moving quickly, blah, blah, blah, are not really available.
• 19:00
  SPEAKER_01: And what you need is two-thirds majorities in the House and Senate to say no.
• 19:06
  SPEAKER_01: Not now.
• 19:09
  SPEAKER_00: Yeah, I mean...
• 19:12
  SPEAKER_02: You know,
• 19:12
  SPEAKER_02: I just think the question is like,
• 19:14
  SPEAKER_02: what can feasibly be done and on what timescale, right?
• 19:20
  SPEAKER_02: Something like an executive branch,
• 19:23
  SPEAKER_02: you know,
• 19:25
  SPEAKER_02: waiver to like say that,
• 19:28
  SPEAKER_02: you know,
• 19:29
  SPEAKER_02: the three leading companies can collaborate on some kind of slowdown so
• 19:34
  SPEAKER_02: that we can buy some time so that we can- Do they need that?
• 19:37
  SPEAKER_01: Probably.
• 19:38
  SPEAKER_02: Probably.
• 19:39
  SPEAKER_02: Yes.
• 19:40
  SPEAKER_02: I mean, the reason,
• 19:41
  SPEAKER_02: I think that one reason why the neobrandeisian people,
• 19:47
  SPEAKER_02: you know,
• 19:47
  SPEAKER_02: led as always by Matt Stoller,
• 19:49
  SPEAKER_02: but backed up more judiciously by Lena Kahn and others,
• 19:53
  SPEAKER_02: were the people on the left who were most vocally skeptical of this safety push,
• 19:59
  SPEAKER_02: is that they see it as,
• 20:03
  SPEAKER_02: you know,
• 20:04
  SPEAKER_02: The line that they're giving is that there is already an antitrust safe harbor
• 20:09
  SPEAKER_02: for product safety collaboration, right?
• 20:12
  SPEAKER_02: So if like Delta and American want to like share information
• 20:19
  SPEAKER_02: about Aviation is probably actually not a good example of this because it's
• 20:24
  SPEAKER_02: so regulated on the other level.
• 20:25
  SPEAKER_02: But, you know, whatever.
• 20:26
  SPEAKER_02: If you have multiple industries that are, like,
• 20:28
  SPEAKER_02: using a similar input,
• 20:30
  SPEAKER_02: and then there's, like, a question about, like, how do you make it not spill?
• 20:35
  SPEAKER_02: You know, you could do collaborative work on that.
• 20:37
  SPEAKER_02: There's an antitrust safe harbor.
• 20:39
  SPEAKER_02: What you cannot do, right, is collectively decide, like...
• 20:45
  SPEAKER_02: Oh, you know, new airplanes like might be risky.
• 20:51
  SPEAKER_02: So all the airlines are going to decide we're not going to buy any new airplanes, right?
• 20:56
  SPEAKER_02: Because the new airplanes,
• 20:57
  SPEAKER_02: they're so dangerous because you would do that as an airline industry in bad faith
• 21:04
  SPEAKER_02: to like actually avoid competing with each other, right?
• 21:07
  SPEAKER_02: If everyone agrees that we're not going to buy any new airplanes for like safety, right?
• 21:12
  SPEAKER_02: Then as demand for air travel increases,
• 21:14
  SPEAKER_02: we can just raise prices forever rather than investing in capacity.
• 21:19
  SPEAKER_02: So the antitrust concern would be that all this safety stuff from Musk,
• 21:26
  SPEAKER_02: from Amadei, from Altman is like a ruse.
• 21:30
  SPEAKER_02: And that actually these companies just want to save money.
• 21:35
  SPEAKER_02: By not investing anymore,
• 21:37
  SPEAKER_02: but they don't want to lose competitive advantage to each other
• 21:40
  SPEAKER_02: by not investing anymore.
• 21:41
  SPEAKER_02: So they're going to like create this fake safety cartel.
• 21:44
  SPEAKER_01: Okay, but- I don't think that that's true,
• 21:47
  SPEAKER_01: but like that is
• 21:48
  SPEAKER_02: a not like,
• 21:50
  SPEAKER_02: I think that's a correct legal analysis of the situation.
• 21:54
  SPEAKER_01: So I read Lena Kahn's Long post about this, and I want to talk about it a second.
• 22:01
  SPEAKER_01: But before I do, because I think that there's something important in there.
• 22:08
  SPEAKER_01: I don't understand what the...
• 22:11
  SPEAKER_01: Let's pretend that Boeing 737 Maxes weren't falling out of the sky.
• 22:15
  SPEAKER_01: There might actually be a reason for that kind of collusion in the airline industry.
• 22:20
  SPEAKER_01: It's totally straightforward to me how you could imagine them colluding in that way
• 22:24
  SPEAKER_01: in order to raise prices and not spend money.
• 22:29
  SPEAKER_01: Who's using, these are proprietary models that they are using at a loss, right?
• 22:35
  SPEAKER_01: Like that's what they're talking about slowing down.
• 22:38
  SPEAKER_02: But it's not a loss.
• 22:39
  SPEAKER_02: I mean, this is, you know, again,
• 22:42
  SPEAKER_01: I just like, I
• 22:42
  SPEAKER_01: want to flag for people.
• 22:43
  SPEAKER_01: Like I do not agree with this.
• 22:45
  SPEAKER_01: I don't see how that makes Claude more expensive for me.
• 22:47
  SPEAKER_02: Okay.
• 22:48
  SPEAKER_02: So right now, Anthropic, OpenAI, X, Google, especially, they are selling to you at a profit.
• 23:00
  SPEAKER_02: Like yesterday's model.
• 23:04
  SPEAKER_02: But they are then spending a shitload of money on building tomorrow's model.
• 23:09
  SPEAKER_02: So the company would just have more cash on hand if it didn't train a new model.
• 23:15
  SPEAKER_01: Yep, I'm with you.
• 23:16
  SPEAKER_02: Like a lot more cash, right?
• 23:18
  SPEAKER_02: I'm with you.
• 23:19
  SPEAKER_02: But they would be afraid, right?
• 23:22
  SPEAKER_02: If OpenAI just decided, well,
• 23:25
  SPEAKER_02: we're going to accumulate more cash instead of training a new model,
• 23:28
  SPEAKER_02: they'd be like, oh,
• 23:30
  SPEAKER_02: but Anthropic's going to come out with a new model and that's going to undercut
• 23:35
  SPEAKER_02: the value of ChatGPT and we'll be fucked.
• 23:38
  SPEAKER_02: But if we can agree with Anthropic that neither of us are going
• 23:42
  SPEAKER_02: to train any new models, then we can both just build up enormous cash reservoirs.
• 23:47
  SPEAKER_01: Okay, but I still don't get what the harm to the consumer is if we're stuck
• 23:52
  SPEAKER_01: with Opus whatever and ChatGPT.
• 23:55
  SPEAKER_02: Yes, I agree with you.
• 23:56
  SPEAKER_02: I think that this is a trivial consumer harm.
• 23:58
  SPEAKER_02: Now, I would say we do have to like officially write down in our notebook, like,
• 24:06
  SPEAKER_02: or perhaps our AI assistant can remind us, like, check back on this in 18 months.
• 24:10
  SPEAKER_02: Yeah.
• 24:13
  SPEAKER_02: You know, you know what I mean?
• 24:13
  SPEAKER_02: Like, well, like, well, yeah, you know, anything.
• 24:15
  SPEAKER_02: Monitor the situation.
• 24:16
  SPEAKER_02: Okay.
• 24:17
  SPEAKER_02: Anything,
• 24:17
  SPEAKER_02: anything you do like that on an age you just front,
• 24:19
  SPEAKER_02: like you don't want to make an irrevocable commitment to,
• 24:22
  SPEAKER_02: but I think that this is a real case of like,
• 24:27
  SPEAKER_02: When all you have is a hammer, everything looks like a nail.
• 24:30
  SPEAKER_02: And the antitrust concern here, I think, is really trivial.
• 24:34
  SPEAKER_02: And people who've just dedicated their lives to antitrust are making too big of a deal.
• 24:38
  SPEAKER_01: Now, here's my other question.
• 24:39
  SPEAKER_01: Did the antitrust people come out against the idea of a regulatory apparatus along
• 24:46
  SPEAKER_01: the lines of what you might wish to see in the climate change realm or whatever?
• 24:50
  SPEAKER_01: Or were they talking about the voluntary...
• 24:54
  SPEAKER_01: Agreement among frontier executives to slow down.
• 24:58
  SPEAKER_02: I think that their basic worldview is that the Chinese open weight models are going
• 25:08
  SPEAKER_02: to catch up with the frontier labs very quickly.
• 25:12
  SPEAKER_02: And that...
• 25:14
  SPEAKER_02: AI will be commoditized and that the huge stock valuations
• 25:20
  SPEAKER_02: that are currently associated with Anthropic and OpenAI are going to evaporate.
• 25:25
  SPEAKER_02: And like,
• 25:26
  SPEAKER_02: not only do they believe that,
• 25:28
  SPEAKER_02: but they believe that the executives at the companies believe that.
• 25:34
  SPEAKER_02: And so they see everything through a lens of these are sweaty,
• 25:39
  SPEAKER_02: desperate efforts by people at frontier companies
• 25:43
  SPEAKER_02: to prop up their own stock valuations by staving off commodification at the hands
• 25:50
  SPEAKER_02: of this Chinese competition.
• 25:52
  SPEAKER_02: I don't think that that's correct,
• 25:54
  SPEAKER_02: but it's like if you are inside that worldview, then you view everything.
• 25:59
  SPEAKER_02: It's like a...
• 26:00
  SPEAKER_01: I don't even understand the theory.
• 26:02
  SPEAKER_02: It's like one of these like, you know that like rabbit duck optical illusion?
• 26:07
  SPEAKER_01: Uh-huh, uh-huh.
• 26:08
  SPEAKER_02: Right?
• 26:08
  SPEAKER_02: So it's like from their view,
• 26:14
  SPEAKER_02: like open AI and anthropic executives are sitting on a pile of fool's gold
• 26:20
  SPEAKER_02: and they are trying to deploy this like fake fear of existential risk and In order
• 26:27
  SPEAKER_02: to save their companies by squelching competition with each other and then creating
• 26:33
  SPEAKER_02: a downstream regulatory pretext to crack down on the...
• 26:39
  SPEAKER_02: The open weight companies.
• 26:41
  SPEAKER_02: And you hear this really clearly because, I mean, we've been talking about the antitrust left.
• 26:45
  SPEAKER_02: Okay.
• 26:45
  SPEAKER_02: But this is also what Marc Andreessen is saying.
• 26:48
  SPEAKER_02: Yeah, the bad guys are clear to me, right?
• 26:50
  SPEAKER_02: Like Andreessen and
• 26:51
  SPEAKER_01: David Sachs.
• 26:52
  SPEAKER_01: Like, I can smell their shit from here to Silicon Valley.
• 26:55
  SPEAKER_02: But they have reached a meeting of the minds because,
• 26:57
  SPEAKER_02: I mean, in some ways- Not really, I don't think.
• 26:59
  SPEAKER_02: I mean, in some ways it's clearer in business terms, right?
• 27:02
  SPEAKER_02: Because, like, Marc Andreessen, what he does professionally is he invests in tech startups.
• 27:10
  SPEAKER_02: He does not own a big piece of open AI or Anthropic.
• 27:15
  SPEAKER_02: Years and years ago,
• 27:17
  SPEAKER_02: when Sam Altman and Dario Amodei were both working at OpenAI,
• 27:21
  SPEAKER_02: and people have crazy startup pitches,
• 27:25
  SPEAKER_02: and their pitch was these technical papers that Google researchers published,
• 27:31
  SPEAKER_02: and they're just sitting around, not being really worked on, are true and correct.
• 27:37
  SPEAKER_02: And if you throw a bunch of money at building...
• 27:43
  SPEAKER_02: More physical resources into training large language models.
• 27:47
  SPEAKER_02: You could build really, really powerful technology on the basis of that.
• 27:51
  SPEAKER_02: Some people believe that.
• 27:53
  SPEAKER_02: Most of the people who believe that are AI safety people.
• 27:58
  SPEAKER_02: And most mainstream venture capitalists thought that was dumb.
• 28:01
  SPEAKER_02: And they invested their money in the metaverse.
• 28:04
  SPEAKER_02: They invested their money in NFTs and crypto.
• 28:08
  SPEAKER_02: And now they've invested their money in AI companies,
• 28:13
  SPEAKER_02: which is to say companies whose business model is to purchase AI and use it
• 28:22
  SPEAKER_02: to sell products to the world, right?
• 28:23
  SPEAKER_02: So like I have a service called Town.
• 28:28
  SPEAKER_02: Which is like,
• 28:29
  SPEAKER_02: my townie reads my email and reads my calendar and gives me a daily briefing
• 28:35
  SPEAKER_02: and tells me stuff, blah, blah, blah.
• 28:37
  SPEAKER_02: But town is not developing a frontier AI model, right?
• 28:43
  SPEAKER_02: I don't know what they're doing, but they're licensing something from somebody.
• 28:47
  SPEAKER_02: And all of those investments, right...
• 28:50
  SPEAKER_02: That all goes to zero if recursive self-improvement happens,
• 28:56
  SPEAKER_02: but it becomes very,
• 28:57
  SPEAKER_02: very profitable if AI,
• 29:01
  SPEAKER_02: if intelligence is commoditized and product development is what counts.
• 29:05
  SPEAKER_02: And so in a lot of ways,
• 29:06
  SPEAKER_02: I just think like the Andreessen side of this argument is a lot clearer
• 29:11
  SPEAKER_02: because he's just like,
• 29:12
  SPEAKER_02: he's literally an investor and he has both made an intellectual bet and now he wants
• 29:18
  SPEAKER_02: a political bet.
• 29:19
  SPEAKER_01: Right,
• 29:20
  SPEAKER_01: and this is why I want to be careful about like alleging a horseshoe type alignment
• 29:27
  SPEAKER_01: between these VCs and the antitrust people.
• 29:30
  SPEAKER_01: Because like I saw David Sachs,
• 29:34
  SPEAKER_01: who's like dirty in all this,
• 29:38
  SPEAKER_01: retweet what Lena Kahn wrote on Twitter or on X about the tools that already exist
• 29:46
  SPEAKER_01: for To rein in AI, right?
• 29:48
  SPEAKER_01: Or to at least discourage, disincentivize reckless AI development.
• 29:57
  SPEAKER_01: And I agree that that might not be enough,
• 30:01
  SPEAKER_01: but there was nothing in Lena's post as far as I read it
• 30:04
  SPEAKER_01: that said what exists is good enough and we shouldn't let these companies...
• 30:11
  SPEAKER_01: Put forward ideas about how to slow the frontier
• 30:14
  SPEAKER_01: or we shouldn't entertain regulatory ideas from Congress.
• 30:19
  SPEAKER_01: What I took her to be saying is one thing that might put your mind at ease is
• 30:25
  SPEAKER_01: that there's a lot of space between the Hugging Bear attack and Skynet.
• 30:30
  SPEAKER_01: And if one of these companies has a model or a swarm that breaks containment
• 30:36
  SPEAKER_01: and they hack Citibank, they're going to be in big trouble.
• 30:40
  SPEAKER_02: And they're probably not
• 30:41
  SPEAKER_01: going to let that happen.
• 30:42
  SPEAKER_01: But that means that we need to keep those tools animated.
• 30:46
  SPEAKER_01: And I fucking agree with that 100%.
• 30:49
  SPEAKER_01: But if there's actually a large contingent
• 30:55
  SPEAKER_01: of these antitrust activists getting members of Congress on board with the idea
• 31:00
  SPEAKER_01: that they don't need to regulate or that they should try
• 31:02
  SPEAKER_01: to stop this quote-unquote collusion,
• 31:05
  SPEAKER_01: That I'd have a problem with, but I haven't seen that, and I'm not sure it's real.
• 31:09
  SPEAKER_01: I mean, listen,
• 31:11
  SPEAKER_02: so I agree.
• 31:11
  SPEAKER_02: It's not a true horseshoe.
• 31:14
  SPEAKER_02: I think the antitrust people, as usual, are basically just useful idiots for the right.
• 31:23
  SPEAKER_02: But...
• 31:24
  SPEAKER_02: There is a horseshoe on this specific question of should we allow
• 31:31
  SPEAKER_02: or mandate slowing down a frontier progress?
• 31:35
  SPEAKER_02: And that's because there is an underlying horseshoe, not about politics, but about the likely...
• 31:44
  SPEAKER_02: Trajectory of the underlying technology.
• 31:47
  SPEAKER_02: I mean, again,
• 31:48
  SPEAKER_02: I hesitate to say recommend,
• 31:50
  SPEAKER_02: but I think people should read Matt Stoller's article, Stop Panicking About AI.
• 31:56
  SPEAKER_02: Because, you know, he lays it out.
• 31:58
  SPEAKER_02: Obviously, Matt Stoller is not a libertarian like Marc Andreessen.
• 32:02
  SPEAKER_02: Matt Stoller loves regulation.
• 32:04
  SPEAKER_02: Marc Andreessen hates it, right?
• 32:06
  SPEAKER_02: The animating terror of the tech right worldview is
• 32:11
  SPEAKER_02: that like nuclear fission is It was like
• 32:14
  SPEAKER_02: a basically safe civilian power generating technology that was killed off based
• 32:20
  SPEAKER_02: on like diffuse ambient fears.
• 32:24
  SPEAKER_02: That is not at all Matt Stoller's worldview.
• 32:27
  SPEAKER_02: He thinks it's incredible when regulation kills off new areas of product technology
• 32:32
  SPEAKER_02: and innovation,
• 32:33
  SPEAKER_02: but they are in total alignment on the idea that like existential risk
• 32:39
  SPEAKER_02: from AI is like a plot of From anthropic and open AI to engage in some kind
• 32:48
  SPEAKER_02: of regulatory capture.
• 32:50
  SPEAKER_02: And I think that that is wrong.
• 32:52
  SPEAKER_01: So I just think that it sounds to me like they can come on side.
• 33:00
  SPEAKER_01: Look, like if the three guys that we talked about at the beginning, Altman, Musk,
• 33:05
  SPEAKER_01: and Amadei released a joint working paper,
• 33:08
  SPEAKER_01: working statement that said, here's our principles.
• 33:11
  SPEAKER_01: And they were clearly just like,
• 33:14
  SPEAKER_01: We're going to collude to stop everyone's progress and be incumbent.
• 33:20
  SPEAKER_01: Then the antitrust people might be vindicated.
• 33:23
  SPEAKER_01: We would probably come around to say that's bullshit.
• 33:26
  SPEAKER_01: If they put forward ideas that were like,
• 33:27
  SPEAKER_01: this might actually allow development of less dangerous technology to continue,
• 33:34
  SPEAKER_01: and it's just a buffer to try to figure out whether alignment is possible,
• 33:38
  SPEAKER_01: whatever, then...
• 33:40
  SPEAKER_01: Then they, I imagine, support it.
• 33:42
  SPEAKER_01: And you might get Congress involved, right?
• 33:44
  SPEAKER_01: If needed to dragoon in the other companies.
• 33:48
  SPEAKER_02: So I guess like, here's the main thing I would say.
• 33:50
  SPEAKER_02: You know,
• 33:51
  SPEAKER_02: like the other,
• 33:52
  SPEAKER_02: other thing,
• 33:52
  SPEAKER_02: it was like,
• 33:53
  SPEAKER_02: I saw like Leah Stokes was like,
• 33:54
  SPEAKER_02: why are we talking about existential threat from AI when we should be talking about
• 33:58
  SPEAKER_02: the existential risk of climate change?
• 34:01
  SPEAKER_02: And it's like,
• 34:02
  SPEAKER_02: I have a lot of disagreements with Leah Stokes,
• 34:05
  SPEAKER_02: but I think that she and I could probably reach agreement on the fact
• 34:08
  SPEAKER_02: that Donald Trump,
• 34:09
  SPEAKER_02: is not going to promulgate tough new regulations that address the problem
• 34:15
  SPEAKER_02: of climate change.
• 34:16
  SPEAKER_01: No, he just, he's doing the opposite.
• 34:18
  SPEAKER_01: Yes, yes, yes.
• 34:20
  SPEAKER_01: Just this week,
• 34:20
  SPEAKER_01: he's like,
• 34:21
  SPEAKER_01: the Clean Air Act doesn't
• 34:22
  SPEAKER_02: cover- Regardless of what you say or I say,
• 34:26
  SPEAKER_02: regardless of what Hakeem Jeffries says, right?
• 34:29
  SPEAKER_02: Like, there is no way that that's going to happen.
• 34:32
  SPEAKER_02: I hold out some hope that if Democrats take Congress in the midterms,
• 34:36
  SPEAKER_02: we can slow the rate at which Donald Trump dismantles climate regulation.
• 34:41
  SPEAKER_02: But it's like not planetary to think that over the next two and a half years,
• 34:46
  SPEAKER_02: the executive branch of the federal government is going to address climate change in
• 34:50
  SPEAKER_02: a meaningful way.
• 34:53
  SPEAKER_02: It is...
• 34:54
  SPEAKER_02: Really problematic if for the next two and a half years,
• 34:57
  SPEAKER_02: we don't do anything to address the pace of frontier AI development.
• 35:01
  SPEAKER_02: And this is just like not a ripe topic.
• 35:05
  SPEAKER_02: You know, like Steve Bannon thinks we should address the pace of frontier AI development.
• 35:09
  SPEAKER_02: Tucker Carlson thinks we should address the pace of frontier AI development.
• 35:12
  SPEAKER_02: Elon Musk thinks we should address the pace of frontier AI development.
• 35:16
  SPEAKER_02: But yes,
• 35:16
  SPEAKER_02: but I mean,
• 35:17
  SPEAKER_02: you can imagine a world in which like pressure is brought to bear on Trump
• 35:22
  SPEAKER_02: and he does something about this.
• 35:24
  SPEAKER_02: He's not going to do what I want him to do.
• 35:27
  SPEAKER_02: He's not going to address the labor market impacts in a satisfactory way.
• 35:31
  SPEAKER_02: There's a million things he's not going to do,
• 35:34
  SPEAKER_02: but like maybe – We could get this executive branch that we are stuck with for
• 35:42
  SPEAKER_02: while to help pump the brakes a little bit, especially since...
• 35:49
  SPEAKER_02: There is so much regulatory backlash at the state level and other things going on
• 35:55
  SPEAKER_02: that would be complementary to relatively minor federal action.
• 36:01
  SPEAKER_02: And just like that literally is the reason why a lot of people are talking right now
• 36:07
  SPEAKER_02: about trying to get something small but like conceivable done in a finite way.
• 36:15
  SPEAKER_02: I think two weeks ago,
• 36:16
  SPEAKER_02: three weeks ago,
• 36:16
  SPEAKER_02: I forget if it was two or three,
• 36:18
  SPEAKER_02: but like I was talking to somebody who works on this and he was so frustrated.
• 36:36
  SPEAKER_02: Because it was like,
• 36:38
  SPEAKER_02: anytime we tried to talk to them,
• 36:39
  SPEAKER_02: it's like word from Hakeem and Schumer is like, do not talk about this.
• 36:43
  SPEAKER_02: Do not talk about AI.
• 36:44
  SPEAKER_02: Do not talk about AI.
• 36:45
  SPEAKER_02: Our number one priority is just to prevent a leading
• 36:50
  SPEAKER_02: the future campaign spending fire hose from being rained down
• 36:53
  SPEAKER_02: on our frontline members.
• 36:55
  SPEAKER_02: And that has changed, right?
• 36:57
  SPEAKER_02: So if you are mostly interested in just the kind of Mesa pace at
• 37:05
  SPEAKER_02: which politics happens,
• 37:07
  SPEAKER_02: like the change that would, the feasible change has occurred, right?
• 37:12
  SPEAKER_02: And like now democratic leaders are saying, sure, talk about AI.
• 37:16
  SPEAKER_02: Like people want to hear AI.
• 37:18
  SPEAKER_02: Democrats should try to capture the leading position on AI regulation.
• 37:21
  SPEAKER_02: There's going to be a midterm election soon.
• 37:25
  SPEAKER_02: Something will happen after that.
• 37:26
  SPEAKER_02: There will be a policy agenda for 2029, et cetera, et cetera, et cetera.
• 37:31
  SPEAKER_02: It's just none of that,
• 37:32
  SPEAKER_02: like formulating a really good 2029 policy agenda doesn't help you
• 37:39
  SPEAKER_02: if recursive self-improvement comes in the spring of 2028.
• 37:42
  SPEAKER_02: Yes.
• 37:44
  SPEAKER_01: And so there's two things that I want to follow on.
• 37:50
  SPEAKER_01: And one is probably better for the other side of the paywall.
• 37:53
  SPEAKER_01: Yeah.
• 37:54
  SPEAKER_01: And that one pertains to this risk.
• 38:00
  SPEAKER_01: Does Hakeem Jeffries think that if Democrats enter the fetal position
• 38:08
  SPEAKER_01: that people like Andreessen aren't going to notice that between
• 38:13
  SPEAKER_01: the Trump-led Republican Party and the Jeffries-led Democratic Party,
• 38:17
  SPEAKER_01: Democrats are likelier to do more AI regulation than Trump wants,
• 38:22
  SPEAKER_01: and thus he needs to open the spigot anyway.
• 38:24
  SPEAKER_02: Well, I mean,
• 38:25
  SPEAKER_02: this is Chris Lehane's job,
• 38:29
  SPEAKER_02: and I didn't answer a call from him yesterday,
• 38:32
  SPEAKER_02: so I don't know what he would say in response to that.
• 38:36
  SPEAKER_01: That just strikes me as, like, not very smart thinking.
• 38:38
  SPEAKER_01: Especially, well, look, like, as you say, it changed.
• 38:41
  SPEAKER_01: And if Donald Trump's out there being like, yeah,
• 38:43
  SPEAKER_01: let AI reign,
• 38:45
  SPEAKER_01: you know,
• 38:46
  SPEAKER_01: Then maybe Democrats will actually take the position
• 38:50
  SPEAKER_01: that wins them political support from the public, which, yes,
• 38:54
  SPEAKER_01: does risk more AI spending against them,
• 38:58
  SPEAKER_01: but at least they're getting the upside of it.
• 39:00
  SPEAKER_02: Well, I would just also say in like super narrow terms, right,
• 39:02
  SPEAKER_02: like politics, right,
• 39:03
  SPEAKER_02: the legitimately the best way to play this for Democrats is to lay low
• 39:12
  SPEAKER_02: until mid-September.
• 39:16
  SPEAKER_02: Because just like it takes time for money to move.
• 39:19
  SPEAKER_02: And right now,
• 39:20
  SPEAKER_02: this has become a high salience issue that there's lots of free media available on.
• 39:25
  SPEAKER_02: So coming to the topic late in the game as a kind of September surprise.
• 39:29
  SPEAKER_01: Okay, yeah, this is why I wanted to take it over there.
• 39:34
  SPEAKER_01: The other,
• 39:34
  SPEAKER_01: you know,
• 39:37
  SPEAKER_01: when I see Elon Musk,
• 39:39
  SPEAKER_01: who I don't trust at all,
• 39:41
  SPEAKER_01: and I think for good reason, say, Dario is right about pacing the frontier.
• 39:47
  SPEAKER_01: And then Donald Trump's saying there will be no pacing.
• 39:49
  SPEAKER_01: We don't need to pace.
• 39:51
  SPEAKER_01: And yet,
• 39:52
  SPEAKER_01: you know,
• 39:52
  SPEAKER_01: these other right wing figures are trying to say,
• 39:56
  SPEAKER_01: hey, actually, maybe we should take this seriously.
• 39:59
  SPEAKER_01: Donald Trump's a little bit mercurial.
• 40:02
  SPEAKER_01: I know, but so is Donald Trump's mercurial and Elon is conniving.
• 40:08
  SPEAKER_01: And it's like, yeah, he seems like he might want to slip through the middle.
• 40:14
  SPEAKER_01: And and like,
• 40:15
  SPEAKER_01: you know,
• 40:16
  SPEAKER_01: make Mecca Hitler the incumbent, you know, and that's not an acceptable outcome.
• 40:24
  SPEAKER_01: And like Donald Trump would allow that to happen.
• 40:26
  SPEAKER_01: And this is why I feel like all of the intermediate ideas to regulate this or
• 40:33
  SPEAKER_01: to slow it don't work.
• 40:35
  SPEAKER_01: Like you can't wish away the Donald Trump problem.
• 40:38
  SPEAKER_01: And so you need to.
• 40:41
  SPEAKER_01: Get public support for pausing AI development up into the 70s.
• 40:46
  SPEAKER_01: It's probably close to there right now, I gather.
• 40:50
  SPEAKER_01: And then you need to make it a big, big, big legislative push.
• 40:54
  SPEAKER_01: And then, I mean, look,
• 40:56
  SPEAKER_01: like on the other side of that,
• 40:59
  SPEAKER_01: if you succeed,
• 41:01
  SPEAKER_01: what you have to do is like say,
• 41:03
  SPEAKER_01: if anything bad happens, it's China's fault and kind of hope for the best.
• 41:07
  SPEAKER_01: If Republicans want to block a bill like that,
• 41:10
  SPEAKER_01: You can say if anything bad happens,
• 41:11
  SPEAKER_01: it's Republicans' fault and then hope for the best.
• 41:14
  SPEAKER_01: But at least you tried.
• 41:16
  SPEAKER_01: Like, I don't...
• 41:17
  SPEAKER_00: Yeah, I mean, you know, I mean, several things.
• 41:21
  SPEAKER_02: I mean, there is a reason why Dustin Moskovitz was...
• 41:28
  SPEAKER_02: The largest Democratic donor in the 2024 election cycle.
• 41:33
  SPEAKER_02: You know, like it is possible that we are just, fuck now.
• 41:40
  SPEAKER_02: Justin Moskowitz, one of the main EA financiers in the universe.
• 41:45
  SPEAKER_01: Former Facebook billionaire.
• 41:47
  SPEAKER_02: Yes.
• 41:48
  SPEAKER_02: Well, although now, yeah.
• 41:53
  SPEAKER_02: And this was often covered at the time as like Kamala was like getting close
• 41:59
  SPEAKER_02: to big tech or something.
• 42:01
  SPEAKER_02: And that's why Dustin was contributing to her campaign,
• 42:05
  SPEAKER_02: which is like a total misread of what was going on there.
• 42:08
  SPEAKER_02: He was trying to avert this outcome, which we are now stuck with.
• 42:13
  SPEAKER_02: You know,
• 42:15
  SPEAKER_02: everybody,
• 42:16
  SPEAKER_02: I think,
• 42:16
  SPEAKER_02: understands that The reason why the people who don't want AI
• 42:22
  SPEAKER_02: to be regulated keep referring to the people who do want it to be regulated
• 42:26
  SPEAKER_02: as doomers is that I think everybody understands
• 42:30
  SPEAKER_02: that dooming is like actually not constructive political action.
• 42:34
  SPEAKER_02: And then no matter how bad things are and no matter how limited the realm
• 42:39
  SPEAKER_02: of success is, the only thing you can do is like keep your head up.
• 42:44
  SPEAKER_02: Try to understand the landscape as best you can and make the plays that make
• 42:48
  SPEAKER_02: the most sense, given the options that are available and the situation that is here.
• 42:54
  SPEAKER_02: There are many things that I wish many people had done differently in the past
• 42:57
  SPEAKER_02: that we would have had different outcomes and be in a different state now,
• 43:01
  SPEAKER_02: but we are where we are, right?
• 43:02
  SPEAKER_02: Right.
• 43:05
  SPEAKER_02: I hope that the technical pace of AI development just hits some roadblocks.
• 43:11
  SPEAKER_02: This is something that people have raised many times in the past as something
• 43:14
  SPEAKER_02: that might happen,
• 43:16
  SPEAKER_02: that scaling might not hold,
• 43:18
  SPEAKER_02: that there may just be some problems,
• 43:20
  SPEAKER_02: problems with the fundamentals that would buy us some more time.
• 43:23
  SPEAKER_02: And I think that would be great.
• 43:25
  SPEAKER_02: I...
• 43:26
  SPEAKER_02: You know, don't mind a little data center backlash.
• 43:31
  SPEAKER_02: I really wish people would lean on Trump to be tougher on chip exports.
• 43:36
  SPEAKER_02: I think like this is like really because it's totally plausible to imagine
• 43:40
  SPEAKER_02: a Republican administration deciding to help a bunch of American companies
• 43:46
  SPEAKER_02: by not helping one American company sell powerful technology to China.
• 43:51
  SPEAKER_02: Yeah.
• 43:52
  SPEAKER_02: Right?
• 43:52
  SPEAKER_02: Like Trump is supposed to be a China hawk.
• 43:54
  SPEAKER_02: He's getting played like a fiddle by Jensen on this topic.
• 43:58
  SPEAKER_02: It's totally inconsistent with like everything he claims to believe and with half
• 44:03
  SPEAKER_02: the shit that the anti-regulation people are saying.
• 44:06
  SPEAKER_02: And, you know, we've got to do what we can do.
• 44:08
  SPEAKER_02: Right.
• 44:10
  SPEAKER_02: And then Congress is going to have to do stuff in the new year.
• 44:15
  SPEAKER_02: And I mean, we should maybe talk about the midterms.
• 44:18
  SPEAKER_02: Yeah, we should.
• 44:19
  SPEAKER_01: I just want to see how aligned you and I are.
• 44:24
  SPEAKER_01: Obviously,
• 44:25
  SPEAKER_01: nothing like the Bernie Sanders bill is going to pass between now and
• 44:27
  SPEAKER_01: the next Congress.
• 44:28
  SPEAKER_01: So my radicalization in that direction, I understand, is like perspective.
• 44:34
  SPEAKER_01: And I'd be willing to climb down,
• 44:36
  SPEAKER_01: perhaps, if some of these things that you're imagining happen.
• 44:40
  SPEAKER_01: But if they don't, does that leave you aligned with me?
• 44:44
  SPEAKER_01: That you just need to, we need to throw bodies in the way of this.
• 44:48
  SPEAKER_02: Well, you have to do a bunch of different things because you have
• 44:50
  SPEAKER_02: to have diplomatic agreements, et cetera, et cetera, et cetera.
• 44:53
  SPEAKER_01: I don't even know that you do.
• 44:55
  SPEAKER_01: It would be best,
• 44:57
  SPEAKER_01: I agree,
• 44:57
  SPEAKER_01: but I think that you say, hey, we,
• 45:01
  SPEAKER_01: the United States of America,
• 45:02
  SPEAKER_01: for the good of the world,
• 45:03
  SPEAKER_01: are stopping now because we think that this imperils the world.
• 45:07
  SPEAKER_01: A, it would be the first thing Donald Trump ever did to win favor with the world.
• 45:11
  SPEAKER_01: And B, it would put immense pressure on China to be like, okay, we are too.
• 45:15
  SPEAKER_01: The thing I would say about Bernie...
• 45:18
  SPEAKER_01: And his bill is that- I mean,
• 45:20
  SPEAKER_01: it's like the specifics are a little confused, I think, but the thrust is right.
• 45:25
  SPEAKER_02: But what I think Bernie should do is like,
• 45:30
  SPEAKER_02: he needs to write the best bill,
• 45:33
  SPEAKER_02: the bill that he likes best, that he can get a Republican co-sponsor for.
• 45:43
  SPEAKER_02: Now, at some point,
• 45:46
  SPEAKER_02: I think that Bernie Sanders and Greg Kassar being like this bill that we have,
• 45:52
  SPEAKER_02: like this is the gold standard,
• 45:54
  SPEAKER_02: is not ultimately going to be a constructive intervention in the discourse.
• 46:01
  SPEAKER_01: I mean, but that's not really how legislation under this kind of duress situation works.
• 46:08
  SPEAKER_02: Legislation happens in all kinds of different ways.
• 46:11
  SPEAKER_02: I think progressives are unduly fixated on the model of legislation.
• 46:17
  SPEAKER_02: Where you do your maximalist.
• 46:19
  SPEAKER_02: Where you do the Overton window and
• 46:21
  SPEAKER_01: you drag it to 50.
• 46:22
  SPEAKER_01: But here's the reason.
• 46:24
  SPEAKER_01: Republicans control Congress,
• 46:25
  SPEAKER_01: so what he has done is stake out the popular position for the left.
• 46:30
  SPEAKER_01: Sure, it's fine.
• 46:31
  SPEAKER_01: And I think that there is a non-zero chance that public demand
• 46:37
  SPEAKER_01: for legislative action gets extremely high.
• 46:41
  SPEAKER_01: And if there is one bill that has had any thought put into it,
• 46:45
  SPEAKER_01: then people are going to be like,
• 46:47
  SPEAKER_01: either we need to pass that or we need to come up with something
• 46:51
  SPEAKER_01: that incorporates some ideas from that,
• 46:55
  SPEAKER_01: but that doesn't lose people on the other side and gets something passed.
• 46:58
  SPEAKER_01: And that it can serve in an unusual way.
• 47:01
  SPEAKER_01: The like Overton window shifting objective that
• 47:05
  SPEAKER_01: so much progressive legislation is meant to do, but doesn't actually do in practice.
• 47:09
  SPEAKER_01: And I am glad that bill exists because in the same way that like after
• 47:14
  SPEAKER_01: the financial crisis, there was like, we need to nationalize the banks.
• 47:16
  SPEAKER_01: And then we need to do like the blunt instrument, like break them up, blah, blah, blah.
• 47:22
  SPEAKER_01: Those ideas didn't get put into law.
• 47:24
  SPEAKER_01: And I think maybe it would have been better if some of them had.
• 47:27
  SPEAKER_01: But the fact that they were out there and that they were easy to understand and
• 47:33
  SPEAKER_01: that there was a lot of popular- Okay, let me put this another way.
• 47:37
  SPEAKER_01: It was good.
• 47:37
  SPEAKER_01: It was to the good.
• 47:38
  SPEAKER_02: I am also glad that the bill exists, but
• 47:46
  SPEAKER_00: there's this disagreement on the left.
• 47:48
  SPEAKER_02: We were talking about the antitrust people and the Neil Brandeis just before.
• 47:51
  SPEAKER_02: And they, Bernie and them, clearly disagree.
• 47:58
  SPEAKER_02: Matt Stoller is saying we shouldn't panic about existential risk from AI,
• 48:03
  SPEAKER_02: and Bernie Sanders is saying we in fact should panic about existential risk from AI.
• 48:09
  SPEAKER_02: Bernie is right about this,
• 48:12
  SPEAKER_02: and I think it will be great if Bernie can mobilize other left-wing people to argue
• 48:20
  SPEAKER_02: with The Stolarite left about this, right?
• 48:24
  SPEAKER_02: Like the intellectual question,
• 48:27
  SPEAKER_02: should we be worried about existential risk from superintelligence is really,
• 48:31
  SPEAKER_02: really important.
• 48:32
  SPEAKER_02: It's a really important predecessor to any kind of implementation.
• 48:37
  SPEAKER_02: Bernie Sanders has a lot of credibility.
• 48:39
  SPEAKER_02: He is very famous.
• 48:41
  SPEAKER_02: He's very well-known.
• 48:42
  SPEAKER_02: If he engages on that, that will be great.
• 48:45
  SPEAKER_02: What I fear is that the comfort zone for Bernie World is going to be to look
• 48:51
  SPEAKER_02: at people differently.
• 48:52
  SPEAKER_02: Who agree with him that we should take AI risk seriously,
• 48:56
  SPEAKER_02: but who are trying to work in a more measured or more bipartisan way and just snipe
• 49:01
  SPEAKER_02: at them from the left rather than engage in the substantive question here,
• 49:08
  SPEAKER_02: which is like, who is correct?
• 49:10
  SPEAKER_02: Should we be worried about superintelligence?
• 49:12
  SPEAKER_02: I think it's good, Bill.
• 49:13
  SPEAKER_02: It's good to have it out there.
• 49:15
  SPEAKER_02: But we have to follow up, not in just like, now we're going to all...
• 49:23
  SPEAKER_02: Like agree to hate on Scott Weiner.
• 49:27
  SPEAKER_01: I agree with this insofar as the don't panic about AI,
• 49:32
  SPEAKER_01: just do antitrust people end up getting truck with people in Congress
• 49:39
  SPEAKER_01: who have real stature.
• 49:41
  SPEAKER_01: But if they're just cranks on the margins of the debate, I don't think it.
• 49:45
  SPEAKER_02: But they're not cranks.
• 49:46
  SPEAKER_02: They ran the Biden administration.
• 49:49
  SPEAKER_02: Yeah.
• 49:51
  SPEAKER_01: I mean, like if Lena Kahn retweets the Stoller piece and then Elizabeth Warren says,
• 49:59
  SPEAKER_01: yeah, you know what?
• 50:00
  SPEAKER_01: They're right.
• 50:00
  SPEAKER_01: And now we got a now we got an interleft problem.
• 50:04
  SPEAKER_01: And I hope that Bernie takes his superior.
• 50:06
  SPEAKER_01: All right.
• 50:07
  SPEAKER_02: I think that Matt Stoller is like one of the most influential people in Democratic politics.
• 50:15
  SPEAKER_02: And I take, or at least he was in the prior iteration that we had.
• 50:19
  SPEAKER_02: And like,
• 50:20
  SPEAKER_02: I am really excited to hear Bernie Sanders articulate a substantive disagreement
• 50:25
  SPEAKER_02: with him.
• 50:25
  SPEAKER_02: And like,
• 50:26
  SPEAKER_02: I think if we can pivot what,
• 50:28
  SPEAKER_02: like who the left listens to and what their analytic framework is,
• 50:31
  SPEAKER_02: like that's incredibly important.
• 50:32
  SPEAKER_01: All right.
• 50:33
  SPEAKER_01: I'm fine.
• 50:33
  SPEAKER_01: I'm fine with it.
• 50:34
  SPEAKER_01: Good.
• 50:35
  SPEAKER_01: Let's talk about the October surprise.
• 50:37
  SPEAKER_01: No, actually it could be like multiple October surprises colliding
• 50:41
  SPEAKER_01: and like Armageddon all at once.
• 50:45
  SPEAKER_01: So if you want to hear that conversation, go to politics.fm and upgrade your subscription.