137: Eli Dourado - LLMs, Stagnation, and Alignment
In this episode, we're joined by Eli Dourado to discuss his recent article Heretical thoughts on AI. How LLMs will interact with the culture war, how alignment can make the great stagnation worse, and a whole lot more.
William Jarvis 0:05
Hey folks, welcome to narratives. narratives is a podcast exploring the ways in which the world is better than in the past, the ways that is worse in the past towards a better, more definite vision of the future. I'm your host, William Jarvis. And I want to thank you for taking the time out of your day to listen to this episode. I hope you enjoy it. You can find show notes, transcripts and videos at narratives podcast.com.
Unknown Speaker 0:41
Well, Eli, how are you doing this afternoon?
I'm doing great. Well,
Will Jarvis 0:44
are you doing? Well, thanks so much for taking the time to come on the show. Again, again, one of our first repeat guests. This is exciting. Eli, do you mind giving just a brief bio for folks who might have not heard the last episode and kind of some of the big ideas are interested in right now?
Sure, sure. So I'm a senior research fellow at the Center for growth and opportunity at Utah State University. And I'm an economist by training. And basically what I work on and think about and try to do in the real world is in the great stagnation, right. So try to increase total factor productivity growth back to at least 2% a year where it belongs, rather than the much, much lower rate of growth that we've been having, since the early 1970s. But especially since about 2005. So I think about like a bunch of industries, energy, transportation, aerospace, particularly I love you know, health housing, you know, kind of the big the big sectors of the economy that have have not really grown as fast as they should have over that period.
Will Jarvis 1:49
I love it. I love it. Um, Eli, if you had to rate how we'd been doing the 2020s, or about 21% 20% of the way for you so far. You know, do you think we are on the right trajectory to escape the great stagnation or what do things? How do things look for you?
i Yeah, I'm, you know, it's a little skewed by COVID. Obviously, like, we don't know. But what would happen, you know, Operation warp speed was like an Dawn complacent event. Right. So that's, that's, that's great. Yeah, but overall, like, I don't see any signs of like, you know, kind of waking up out of our slumber and and, you know, taking seriously the need to, like, grow and like these, these big industrial sectors.
Will Jarvis 2:38
Gotcha. Gotcha. Yeah, it's tough man. And even people are friendly to cause I hate to, you know, we've had a lot of Essex County MPs had a lot of success recently, which I've been really excited about, that makes me bullish, but then, you know, it's time to build Marc Andreessen still, you know, privately, well prevent building an Atherton. So it's, it's tough man. It's tough out there for, for folks trying to fix things. I'm curious. One of the biggest high points recently has been LLM is chat GPT, you know, really can cause kind of caught the imagination of everyone. I found it quite useful in my day to day, in fact, I used it to help a cup, right, a couple of conference submissions, I thought would take, you know, a full day of writing ended up finishing them in about 15 minutes with thanks to chat gbta. So, you know, there's there's been some, you know, productivity increases, but then again, it may not be the end all be all for knowledge workers and improving productivity. What do you think, you know, when we look back five years from now, what will we think about the LLM kind of hype train right now? Will it have solved some of these productivity growth problems? Or will be be kind of disappointed about the effects?
That's a great question. So I think over the next five years, these models are just gonna get better and better. Yeah. Right. And they are going to play a role in our lives. And especially I think they will affect pretty significantly the content industry, the media industry, the advertising industry, things like that, you know, like the media industry is only 2% of GDP, but it is it punches like way above its weight in terms of like cultural salience, I say, right, and so like, for people like you and me, we spend a big chunk of our day in front of a screen, right? And we're consuming media. And so it's gonna feel like, oh, wow, it's this massive change. But if you look at, you know, how it's affecting these sort of large industrial sectors that are in the world of atoms and not the world of bits, like, I think the impact, you know, I think it probably will be minimal. I think it's conceivable that it wouldn't be minimal, but it would rely on us sort of Not sabotaging it in like we've sabotaged everything else in these sectors for the last two years.
Will Jarvis 5:07
So actually, I want to take this in in an interesting direction. This might be a little unexpected, but I would love your take on this. One of my big worries about the alignment community and like everyone focused on that is like, you know, I feel like LLM are one of these few areas of innovation in our economy right now. And this whole, like, kind of Luddite sentiment that we need to stop like whatever like we need to, clearly this is the end of the world, this kind of like revelation to ask, you know, New Testament free, millennia Aryan Protestantism, or what something weird, like the apocalypse is coming, we need to stop it. And it is, I'm much more more worried about the stagnation Apocalypse that is upon us immediately, then this somewhat more imaginary alignment problem, which I believe is still promising, we should be thinking about it, but I don't think we should stop the brakes on our progress because of it.
Yeah, I, I agree with you. And to Yes, and you, I would also say add the the issue of just even I think the idea of like moderating the output of these MLMs is kind of crazy, because it it's going to turn the LLM into a culture war object. Oh, yes. Great point, like, like the idea that, that we have to make sure that the LLM 's never say anything taboo or offensive. Yes, in the same way that we kind of make sure that like, we, you know, we limit the offensiveness of what you can say on Twitter or Facebook, right? Like there's like you can, you can obviously say some offensive things, but but that battle over where the line is, and who should be in charge, and how we police this, it's coming, you know, it's going to come for the LLM. So it's it's going to be this very unpleasant cultural war extension, over over LLM. And so I think I do think like, we should be thinking about, like, how can we make credibly neutral? LLM? Oh, nice. Right. Like, yeah, like, what is what is, is there an approach to that, right, and I think like open source models, and, you know, the tools that we might be able to build that, that allow, like federated training, or that, you know, that sort of can do fundraising for for an open source model that that would be like more credibly neutral, and that you could you could run on your own device instead of running it in the cloud, right, that those kinds of innovations, I think, are going to be important if we want to avoid, you know, like, a very unpleasant few years.
Will Jarvis 7:47
Yes, yes. This is a great point, the political football what chat GPT thinks about Joe Biden or DeSantis? Or something like, it doesn't, right? It's it's quite bizarre thing to be worried about. But it does seem like a lot of people have this top of mind, the end of the day. A very interesting, do you think, um, you know, LLM will make knowledge workers a lot more productive, perhaps like consultants, you know, like, you can just build the deck immediately. You don't have to spend hours grinding on it for you, like, will it make, you know, producing content or like, you know, analysis easier? Or do you think it just is kind of like more of a parlor trick than we want to believe right now?
I have, you know, I've like tried to use it a few times. And it hasn't, hasn't really, it's not like self evidently. Put anything, you know, changed my process in any way. It's not, for me, it's not been compelling. I think. Now, it kind of depends on what kind of content you're you're producing, right? Like, if I'm, if I'm writing, like, a blog post on my personal blog, like, I aspire to, like, the highest levels of like, surprise and newness, originality, right? In those posts, I'm not writing to like, check a box or to like, submit a form, or, you know, or, like, you know, like, like, and I think if you're, if you are, you know, I think there's a lot of a lot of process in the, in the world, right, where it's just like, we need you to write this many sentences about this, you know, on your application, so that so that we can can move forward. And like, you know, I think it might be able to help some with that. But then, I mean, the real question is, like, that process is really stupid, right? Like, like, if we're ever kind of, if it's if it's kind of like reducible to like, a lesser number of bits, like, we should just be like sort of asking for those bits, rather than using AI is like a compression and expansion tool to like You know, to, to like, you know, mediate between the thing, you know, the two humans who really just need to exchange a smaller number of bits. Right? So I don't know. And then I think about like, you know, business processes, like we saw, Microsoft announced, like Microsoft Teams premium, right, which has like some tools that look like, okay, like, like, meeting summaries and like, you can figure out what the action items were from the meeting, like, really quickly and stuff like, that seems like, not obviously bad, or, you know, like, maybe potentially useful. But like my question, like, in terms of productivity, it's like, you know, what I want to know, is like, when do we start replacing people? Process? Like, when do we lay people off? Because like, Oh, we don't need you anymore, because we have an AI model to do your job. And I don't know, when that's coming in the business world, right. Like, like, you know, with with word processors, like we got rid of a bunch of secretaries. Yeah. typing pool people. Right. And, yeah, I'd be curious to like, to know, like, what are the when are those layoffs coming? Because until the layoffs happen, like, we're not getting more, we're not really getting more productive? Gotcha. It's, I think it's like, it's a, I wouldn't say I wouldn't say it's an illusion either, but it's just like, we're getting more information. That is possibly not really that high value.
Will Jarvis 11:35
Gotcha. So this is actually this is really worrisome to me, because it seems like, you know, this is one of the big areas of, of hyper, like, you know, when you point someone to like, oh, like, you know, we're getting, you know, look, progress is increasing so fast, and all these things, and they pointed to MLMs. And then it's like, well, if there's really not a ton there, at the end of the day, it's like, maybe we're more cooked and, you know, even the the bears would like to believe.
Yeah, yeah. That's, I mean, that's, that's kind of my concern with it as well, that they that sort of, like people can say, like, Oh, don't worry, like economic growth is coming. We can like, sort of sit back and wait for it to happen, as you know, Moore's Law progresses. And yes, you know, people are training these models and stuff, and we're gonna have like, you know, post capitalist, like, luxury, everything, because we're just gonna have the AI models produce everything. I don't think that's the world we're in. So. So, you know, I think we got it, we got to, we got to work pretty hard and figure out how to unlock these, these more physical sectors of the economy, and, and take the most advantage of the AI tools. Because I think, you know, like AI tools could have huge benefits in say medicine. Yes, right, if they're applied properly, but probably it's gonna be illegal for, you know, some period of time to like, apply them the way we should. So we got to fix that. Right?
Will Jarvis 13:06
I just want to highlight for the audience how, and I agree, you live with your perspective. I like a lot, but but just how different it is when I visit Berkeley, and I'm talking to everyone, and they're literally like, These things are so powerful. They're gonna kill all of us in 10 years. And I'm not saying for retirement. I mean, this is like what someone told me at a house party, right, a very smart, well known, you know, rationalist blogger, right. And is it just it these things are like diametrically opposed, right, which is, which is, which is very interesting, but I do think it is the case. I think you're right on this, you know, that that the they really are, it's just the it's not it's not quite there, there's something people are missing. What do you think? Why do really smart people have such differing views on this? Do you have any? That's like a meta question. That's very difficult. Maybe not, you're not able to answer it, and that's fine. But but do you have any sense of why that is?
I don't know. I, you know, thinking about my own experience, right. Like,
I'm a little bit older than some of these. The most enthusiastic AI proponents, and I've been burned before. Right. Like, like, like, you know, I think that I think that's, I think that's part of it, is that, you know, I was, you know, I kind of was around for like, the rise of the web, right. I got on the web as a teenager and tinkered and I was like, this is gonna change the world. And I was, like, super excited about it. And, you know, late 90s, early 2000s, you know, continued to follow it and stuff and I was like, Yeah, this is gonna, this is going to change the world. And it did change the world. But just it didn't change the economy. Right, right. It didn't it didn't you know, it didn't it didn't get deliver massive amounts of growth. And I think there's not a lot of people who, you know, if they're if they're kind of like, they're both like tech aware, and, you know, sort of economic See, in terms of their way of thinking, you know, and have been around. Right to get burned the first time. Right. And so I think that's, that's, I think that that is, you know, part of my perspective is that is that I kind of, I think got it wrong, right. Before that, I thought, I thought digital technology was going to do more for us than it than it did. Right. And now we're just it feels very much like, Okay, this is just an extension of that.
Will Jarvis 15:48
Right, make the mix makes a lot of sense. Going back to what we can actually do, to kind of juice growth. You know, we talked, it was probably a year and a half ago about NEPA. Have there been I know, at the time, the Trump administration was working on a couple of different things and trying to get things moving again, has there been any positive movement on kind of fixing kind of these broken, you know, permitting and environmental review processes in the US?
You know, legally, I don't think there's been, not much has moved the needle, right, like so. So Trump did get some reforms through on, you know, changing the NEPA implementing regulations. Some of those were, a small number of those were immediately rolled back by by the Biden White House when they got in. Some of them have stood up for now, some of the guidance like that Trump rolled back also like the Biden, Biden administration is now in the process of reinstating. So so legally, like, you know, like some of the Trump changes have stood in some have, you know, I think think that Biden administration might might still going to take another crack at figuring out what, you know, what a more comprehensive reform could look like. And I think, you know, according to their last rulemaking on this, like, they were going to do more in the future. So, so, you know, we'll see, but it's been it's been, it's been pretty marginal. And but I think that there has been something really positive that's happened over the last year and a half. Yeah. Which is, more and more people have become aware of the problem and have agreed that it it's a problem. Right, so So I think, I think in particular, the, like the climate community, right, people who really want to build a lot of clean energy, yeah, projects over the next 10 years. I think they're starting to realize like, okay, tactically, if we want to build this much clean energy, we're gonna have to do a lot of environmental reviews. And, you know, conservative opponents of this can use NEPA to object to it, right. So you see, you know, you're seeing, you know, offshore wind projects, getting
Unknown Speaker 18:09
challenged, right, like, like, by by people who are like, ostensibly are concerned about, like, some, like crabs or something like that. Right. And, and that is holding things back. I think you're, you know, what, one thing that I think is still not in public consciousness, but there's a case where people are challenging in some immigration reforms on the basis of NEPA. Right, so like, I think it's like the Biden administration's decision to stop building the border wall has been challenged on NEPA grounds. Yes. So and that case is live. And it's like ongoing in the in the DC District Court. So. So So I, you know, I think that as I think that as that awareness grows, so there was, you know, a lot of talk with in the last Congress about, about some kind of permitting reform for energy. And this was pushed by Joe Manchin and part of his deal for supporting the the inflation Reduction Act. Right. And I think, I think that that conversation will continue. But even even the mentioned Bill was, like, pretty marginal, on terms of the reforms that it would do so. So I don't know, I think that probably what needs to happen is there probably needs to be some marginal reform pass that like, you know, it's probably not going to do much we need to have a few years go by where not much gets done. And then we need to revisit it and say, like, oh, look, not much got done. We need to we need to have more, more serious reform.
Will Jarvis 19:45
Gotcha, gotcha. You like if if I said, you know, you need to fix NEPA, you've got a year. Let's say I give you a war chest of, I don't know 10 million bucks. Is there anything you know, is there anything anything like any and you had to give your best shot? Well, Would the approach be there?
I don't know. I think you know, I so one of the things I'm thinking about more and more is, let's say there was someone in the White House who was really committed to this. Yeah, I think there's a NEPA is an act of Congress. Right? That's right. You don't you don't get you don't unilaterally get to repeal acts of Congress if you are president. But there isn't a lot that the executive can do. Gotcha. Right. There's a lot. Just just executive action. Yeah. So. So, you know, if there is a were to be a new president in, you know, a couple years, and and they wanted to really, really like spend their political capital. Yeah. On, on fixing, you know, maximal permanent reform? Yeah, they could, they could do quite a lot without Congress. And I think that's kind of one thing I'm thinking a lot about is, you know, I don't know when we'll have the opportunity. Right. If we, if we had the opportunity, what would what would the playbook be?
Will Jarvis 21:01
Gotcha, that makes sense? Well, I love it. And, Eli, I'm really glad you're admiring this problem. I think this is an incredibly important problem. And, you know, I'm quite concerned about stagnation and not enough people working on it, but I'm talking to people like you, it gives me hope that there is that, you know, there's still a chance we can we can escape it, which which I'm happy about. Eli, you know, if the audience wants to think more deeply about these issues around NEPA, etc, and try and contribute and help them move the needle here, where should they go? What would you recommend?
Well, you know, my organization, the Center for growth and opportunities, looking very closely at this issue, so yeah, find us online at the cgo.org. Yeah, that's, that's the main thing I would recommend.
Will Jarvis 21:53
I love that. I love that. Well, Eli, thanks again for hopping on. It was great to catch up and catch up. And, again, I'm really excited to working on this problem. I think it's one of the most important problems and it's quite neglected as well, which is good.
Thanks. Well, Matt. Great. Great to be with you.
William Jarvis 22:08
Absolutely. Thanks for listening. We'll be back next week with a new episode of narratives.