Brian Chau, founder of Effort News, joins Infinite Loops to explore how AI could fundamentally reshape investigative journalism - not by writing articles, but by searching, analyzing, and connecting millions of legal, financial, and government records that humans could never examine manually.
We discuss how AI changes the economics of finding needles in haystacks, Effort’s investigations into hidden funding and government spending, whether AI can actually make journalism more rigorous, and what happens when everyone suddenly has the ability to interrogate the information around them. We also explore the darker side of these tools, including AI-powered persuasion, misinformation, cognitive security and the “crisis of nominalism.”
We’ve shared some highlights below, together with links & a full transcript. If you like what you hear/read, please leave a comment or drop us a review on your provider of choice.
Links
Highlights
How Journalists Source Stories
Brian Chau
The way that journalism is sourced is totally shocking to me. And maybe you’ve been around the block enough that you kind of just know this and have known this forever. But almost all of investigative journalism is solicited. In other words, it’s someone coming to you, the journalist, with a story they want you to run. And I’m not saying that it’s never good, it’s never ethical, to run with a story that’s solicited. Of course, sometimes people come to you with a story, and that’s really important news. It’s genuinely important. And it’s totally a good thing that they run that. And it’s been like that since Watergate, right? Watergate’s famously a solicited story. And it’s been like that actually long before that.But that actually warps the entire ecosystem of what you see in front of you. And until very recently, it was physically impossible to have stories that came basically any other way. Because to do something like what we’re doing now, in which, for example, we query every single row of the HHS grant table, and that’s how we find a story. And we do that literally, exhaustively, literally looking at the same criteria for all grants, saying, okay, which grants have spiked in the past five years, which grants have grown a ton, either proportionately or as an absolute value, and what relationship those might have to something that’s newsworthy. And I really don’t blame journalists pre AI, because like you said, if you were to do that manually, that would take you literally a lifetime.
And so, just based on the constraints of practicality, not necessarily due to any particular motivation, it was not possible to do that. And now we can have journalism that we source through these exhaustive investigations into legal and financial records, and we can actually have a more scientific way of approaching these things, where it’s like, okay, here are the data sources that we’ve searched. Here are the prompts that we used. Here are the results that we have. And here are the investigations that we tried to do and that we actually hit a brick wall. We didn’t. We found nothing.
A Conspiracy of Incompetence
Jim O’Shaughnessy
We all love our priors, and that turns us into confirmation bias machines where we literally are blinded to information that conflicts with our priors. So let’s just say that’s the average voter, right, if we keep this in the political realm. So an unkind term would be low information voters, right? And nevertheless you could make, I think, I’m asking you really, I think you could cherry pick real facts, not “I made it the fuck up”, but real facts that are verifiable in, let’s say, one of these many government data sets or private data sets or whatever, and you could turn that into an incredibly persuasive story […]By the way, I’m amused by most conspiracy theorists, because the real conspiracy is there really is no conspiracy except in filing.
Brian Chau
Yeah, it’s a conspiracy of incompetence. My least favorite phrase in the entire English language is “do not attribute to malice what can be attributed to incompetence.” Because if you’ve ever spent a moment in the real world, you would know that the most malicious people are also extremely incompetent. And very often incompetent people are also malicious. It’s not mutually exclusive. And yeah, I think you’re totally right
Edited Transcript
Jim O’Shaughnessy
Well, hello everyone, it’s Jim O’Shaughnessy with yet another Infinite Loops. I am very excited to have Brian Chau, the founder of Effort News, an innovative open-source investigative journalism site, on as a guest today. Brian, I love this idea so much. We have our own on-prem AI installation, hardware, support software, models, that we use, and this was one of the ideas, and you beat us to it. So congratulations.
Brian Chau
Thank you. It’s been really surprising, both the stories that have gone around and people just didn’t investigate at all, and the underlying infra. Because people have been talking about AI and journalism for basically as long as I’ve been around. This is pre ChatGPT, people are talking about AI and journalism, and just no one has done it to any effective degree at this point. I think everyone tries to do AI for writing, right? Everyone tries to make AI write the articles. And I don’t know if you’ve noticed this, but when AI writes the articles, they are just terrible. They are utterly terrible. And no one has really made a dent in using AI to investigate these troves of legal and financial documents, which is what we’re doing, and we’re changing that.
Jim O’Shaughnessy
So for listeners and viewers who are not familiar with your new endeavor, explain exactly for our listeners and readers what you’re doing, why you’re doing it. I know that I’m a huge fan of tech making things possible that were impossible before, right? So the first book I wrote was called Invest Like the Best: How to Use Your Computer to Unlock the Secrets of the Top Money Managers. And it was literally, hey, no human being can go through the 1600 annual reports of every company that get released on a quarterly basis. But it’s trivial for a computer to do that. That was way back in the 90s. Now we’ve got the supercharged AI. It can go through tens of millions of records. But talk about A, what was the motivation, B, how you put it into practice effectively. And another thing I really want you to comment on is something that I love, and that is you published the null sets. I think that’s really, really important.
Brian Chau
Yeah, it’s really shocking even before you get to the null sets. The way that journalism is sourced is totally shocking to me. And maybe you’ve been around the block enough that you kind of just know this and have known this forever. But almost all of investigative journalism is solicited. In other words, it’s someone coming to you, the journalist, with a story they want you to run. And I’m not saying that it’s never good, it’s never ethical, to run with a story that’s solicited. Of course, sometimes people come to you with a story, and that’s really important news. It’s genuinely important. And it’s totally a good thing that they run that. And it’s been like that since Watergate, right? Watergate’s famously a solicited story. And it’s been like that actually long before that.
But that actually warps the entire ecosystem of what you see in front of you. And until very recently, it was physically impossible to have stories that came basically any other way. Because to do something like what we’re doing now, in which, for example, we query every single row of the HHS grant table, and that’s how we find a story. And we do that literally, exhaustively, literally looking at the same criteria for all grants, saying, okay, which grants have spiked in the past five years, which grants have grown a ton, either proportionately or as an absolute value, and what relationship those might have to something that’s newsworthy. And I really don’t blame journalists pre AI, because like you said, if you were to do that manually, that would take you literally a lifetime.
And so, just based on the constraints of practicality, not necessarily due to any particular motivation, it was not possible to do that. And now we can have journalism that we source through these exhaustive investigations into legal and financial records, and we can actually have a more scientific way of approaching these things, where it’s like, okay, here are the data sources that we’ve searched. Here are the prompts that we used. Here are the results that we have. And here are the investigations that we tried to do and that we actually hit a brick wall. We didn’t. We found nothing.
Jim O’Shaughnessy
Yeah. And I love the nothing, because learning via negativa is something that is not natural to we humans, right, for the most part. I was a big Sherlock Holmes fan when I was a kid, and I remember the story.
Brian Chau
Did you want the Sherlock Holmes cold cases?
Jim O’Shaughnessy
Yeah, yeah. And I always loved the story where he knew that the intruder was known to the family because the dog didn’t bark. And that’s something people just have a hard time getting their heads around, right? All of the null cases that sound very enticing, or like a motivated reasoner could say, well, of course there’s a conspiracy, well, of course this is going on. And then null says, yeah, we scoured 100 million data points and we’re not finding anything.
I wonder though, is, you mentioned a traditional journalist will generally look to sources, right, and then basically go try to either confirm or deny what the source is telling them. Here the sources are a cornucopia of documents that have been required for filing for public information, et cetera. What about situations where there are very little reporting requirements, and/or, for example, the size of the offshore empire, I call it the empire of elsewhere because of the way it gets cleared through the City of London but settles in jurisdictions that used to be part of the British Empire but have very lax financial diligence and paperwork requirements. How do you get into things that aren’t in a Freedom of Information, or aren’t one Freedom of Information Act request away?
Brian Chau
I want to acknowledge that there’s a lot of truth in what you’re saying, but also a lot of the time these barriers are more paper barriers than you might think. So we had one story where we uncovered the donors behind this degrowth organization in Europe called Partners for a New Economy, P4NE. And what we did there is that they were actually hiding their money. It’s funny enough, this degrowth organization calls for an end to growth, shifting of economic policies away from prioritizing growth. Who are they funded by? They’re funded by these 19th century European heirs who are hiding their donations through literal Swiss bank accounts. So they had this big, essentially money mixer, legal version of a money mixer, called Swiss Philanthropy Foundation. And what Swiss Philanthropy Foundation does is that it fiscally sponsors 106 different, essentially like NGOs.
And whenever anyone makes a donation to any of the 106, including the degrowth organization we are investigating, on paper that’s a donation to Swiss Philanthropy Foundation. In theory, no one knows who it’s going to. No one knows if it’s going to P4NE or anyone else or any one of the 106 different foundations that they have in there. So how did we crack this case? We ended up using those donation rows to trace back the donor side disclosures. In other words, in a lot of cases, even if P4NE doesn’t have to disclose their donors because technically it’s being done through the Swiss Philanthropy Foundation, a lot of the time the donors have their own disclosures that they need to do, and you can go line by line and see what matches, and then go to the donor rows and then see if they have a record for those same transactions to Swiss Philanthropy Foundation. And that’s how we traced back a lot of the donors to P4NE.
So there’s both cases where you’re right, there’s more of a layer of obscurity. But I think there’s a big juicy middle there, where there’s a lot of obscurity that does enough to obfuscate you from traditional journalists, or from people who are just doing it by hand or just doing it using traditional software, and from some of the tools and the search mechanisms that we’ve unlocked at Effort. And I think that’s true on a national and also an international level.
You mentioned routing offshore. We know where a lot of these accounts are filed. A lot of them are Hong Kong, a lot of them are Dubai. A lot of them are kind of like vaguely Saudi related accounts. I don’t know if I’m saying anything I shouldn’t on this, but this is all pretty well known, and they’re well known operators that will teach you how to do it as well. And I think that there is a lot more to uncover there. Maybe that’s not the ideal, maybe that’s not the pinnacle of 100% transparency, but I think just as a practical effort, we can put together a lot on that front still.
Jim O’Shaughnessy
Yeah, I’m finally writing a fictional novel that I’ve wanted to write. I had the idea for it like 34 years ago and I’m finally writing it. And we’re using a very experimental technique where we have a writer’s room. It’s me generating all of the story and all of the ideas behind it. But then the writer’s room is like, yeah, Jim, this doesn’t make any sense. And how do you ever think about that? It’s the way a lot of TV shows are produced.
Brian Chau
Yeah. I was just about to say, yeah, you’re writing the Seinfeld sci-fi novel.
Jim O’Shaughnessy
We are actually putting AI editors and writers into the writers room too. So I’m a big believer in the centaur model, which is man plus the tool, right? I look at AI as just an incredible tool that empowers us, we humans, to make the story better, et cetera. One of the things I bring that up because, as I mentioned, if you...
Brian Chau
Can I ask, how well is that going? I think a lot about using AI to improve my writing. And I’ve not really gotten anywhere with this. So I’d be curious on how that’s going with AI editors.
Jim O’Shaughnessy
The editing, certainly on copy edits, it is virtually perfect. On fact checking now that hallucinations have gone way down. Again, it’s difficult because we have our own on-prem which has been fine tuned for the things we’re looking for. So we’re not using just a commercial model off the shelf. And so it’s also very good at that.
On the writing, it’s a very good critic of writing, but that’s up to you. By that I mean you’ve got to come up with a prompt where you get rid of all the sycophancy, where you get rid of. And so I came up with one that was like, I want you to embody one of the most erudite, well-read critics, and I want him to hate everything that I’m going to put in front of him. And oh man, did that hurt. And the thing that is really interesting is it caught things that I would not, and my human writing team and editors also missed. Like, oh, we didn’t even think about that. So it’s really good at all of that. On the writing side it’s getting a lot better. But what you want it to do is you just have to constantly reinforce, no, this is not the voice. This is not my voice. Sorry, gone. But I suspect that as it goes along it will improve dramatically.
But the point I was going to bring up was, as now, remember, this is a fictional novel, but I run it through not only a panel of experts that we generate. So if a chapter happens to be about offshore accounts, we spin up expert panelist members who are expert in that particular thing. We spin up experts in the City of London and the Remembrancer, and on the Bank of International Settlement, et cetera. And then the reasoners are the ones that check all the facts. And we use an adversarial system there. We have one group of agents saying this is true, and we have another group of agents saying, no it’s not, and here’s the evidence.
But I ran a fictional chapter through it just recently and it kind of blew me away, because remember, this is fiction, but it’s looking for the facts around the town I’m using, right? And it comes back, one of our villains starts a terrorist campaign in Europe, and one of the things that is part of that is an electrical grid attack, right? And the AI came back with, yes, that actually did happen at that particular time. But what you maybe didn’t know was the family that electrified that town in this particular country, they had their own house on top of a hill that they had their own electrical system built for. So even if the entire town went black, their house would still be lit up. And I’m like, wait a minute. And so I went the traditional route and went and found the paper records and found all of that. I’m like, damn, it was right. So I suspect that in what you’re doing, you can find all sorts of similar items that don’t intuitively appeal to us.
Brian Chau
Yes. And there’s all sorts of things that become notable via the combination of information. The most obvious example is plagiarism, which is now in the news for other reasons. But you have one person write a thesis, that’s no big deal. You have two people write the exact same thesis, now that’s news. And it’s a lot like that for these entire categories of crimes or of newsworthy findings. Take fraud, for example, right? You have one ledger that’s just all of the expenses. Okay, that’s like, maybe you can look at that and say, oh, they’re paying too much for these services, they’re running inefficiently. Okay. But it is another level of newsworthiness if you have two ledgers, one for the people who are getting paid and one for the people who are paying, and they don’t match. Now that’s an entire different category of information.
Jim O’Shaughnessy
Yeah. And one of the things I thought is, the old way of doing things was a reporter would start with a hypothesis, and then he or she would go and try to find evidence in favor of that hypothesis and, ideally, information that was at odds with that hypothesis. But aren’t the prompts you’re using, isn’t this just like your prompt is your hypothesis?
Brian Chau
No, I think that is true at an abstract, philosophical level, and I think that is less true in practice, where there are degrees to this. For example, when we found the Office of Refugee Resettlement grant story, this was our biggest story at like a hundred, or sorry, it had a million and a half views on Twitter. The prompt that generated that was something like, I had at that point already set up the infra to efficiently fetch and create a mirror of the HHS grant table. But it was, search this database, identify any newsworthy findings, transactions, or evidence of wrongdoing. Look carefully for rapid jumps in absolute and relative values of grants, and cross reference that with other investigative reporting or legal findings.
And what we ended up finding was something that was very specific, was related to these particular religious charities, which claim to represent the viewpoints of religious Americans, both Catholic and other Christian and Jewish Americans, but really had the vast majority of their revenue, 81.8% in one case, coming from these federal migration services grants. So I know a lot of people with different perspectives. You can have whatever perspective you want on refugee resettlement, but the organizations that are moral arbiters for this issue should not be organizations that are actively profiting from these federal grants. They can provide those services or they can be the commentators, but that’s a clear conflict of interest.
But the way that we got these stories was with this open ended query. And in many cases the open ended queries are just more effective. They are just better. And it’s a way that the information flows are shaping the direction of journalism, I think for the better. Because if I set out and I’m like, okay, find me fraud in daycare centers ahead of time, and that’s all I say, then I probably would have missed the story. It’s not overtly fraud. It’s technically legal. It’s not in daycare centers, obviously, it’s in a different funding stream. And in this case, actually having a more exhaustive and more expansive prompt at the start leads you to having a higher likelihood of finding a newsworthy story.
And that is something that is inverted from journalism in the past, where in the past it would almost always be the case, just because of the sheer effort involved, the sheer human quantity of time and labor, that the more narrow your hypothesis, the more likely you were to get somewhere quickly. And that is just the opposite of true at this point, at least in our experience. And I think that this will have a permanent reshaping on what stories end up getting published.
Jim O’Shaughnessy
Okay, so let me ask a question. Imagine we gave your harness and your models to a highly ideological person. And it doesn’t matter what extreme they’re on. Okay, so they’re either a communist or they’re a mega free market type, right? And they were able to run a million queries against a government database. Couldn’t they construct an extraordinarily persuasive but misleading story using only actual factual data that they cherry picked from those government databases?
Brian Chau
I think that’s true in theory, but I think that’s a better equilibrium than what we have now, right? The equilibrium that we have now is basically the “I made it the fuck up” equilibrium, right? It is, you know, all these stories about data center and water usage. It’s like, the real number is like 1%. Of what? Of what? Like, or the real amount is like 1% of total water usage. It’s less than 1%. It’s like a fraction of what, like almond farmers alone?
Jim O’Shaughnessy
Pistachios.
Brian Chau
Yeah, yeah. If you ask me, like, okay, for Lent, you have to give up everything that you use data centers for, or you have to give up almonds. What would be the harder choice? And it would obviously not be almonds.
And it was this incredibly effective campaign of not even laundered, like, misconstructions, necessarily. You could call them misconstructions, you can call them misreadings or misrepresentations of denominators, but really what they were was, there was this one author who basically made it the fuck up and was promoted by the New York Times, was promoted by all these journalistic institutions who clearly had a story they wanted to tell and was grasping at straws in order to tell it.
And I think it’s in theory possible that they could do the same thing using the tools that we have. But in some sense that would be a restraining force, not an amplifying force. And here’s what I mean. We end up making literally hundreds of falsifiable claims when we set out. And actually not even falsifiable claims at the level of, here is the actual amount of water a data center uses, but falsifiable claims in that we’re saying, here is this financial record that is proving what we said, and here’s the link to that record. And really, if we were just full of slop, if we were just making shit up, then you could click on those links and you could verify that and you could very easily dismiss us. You could very easily generate this kind of reputation obliterating point, which is like, oh, you claim to have found this invoice, but actually this is just completely fake.
And that is actually something that has happened in some cases when it comes to either defendants or lawyers, defense attorneys, or attorneys in general, making ChatGPT hallucinated citations, or other news stories. Or famously, in the Nathan Cofnas case, the person who suspended Nathan Cofnas had this university speech which made up quotes by people like Albert Einstein. And apparently this was because that person used, this is Petra De Sutter, used ChatGPT to write the speech, which itself is a little bit suspicious, I think, because if you’re doing that with modern ChatGPT, it will not fabricate quotes so brazenly in that way. I almost think it’s some kind of deflection.
But it’s this information ecosystem where it is the “I made it the fuck up” information ecosystem, and there are all sorts of historical institutional legitimacies that are basically tied to the practice of making things up at the moment. And also, to be fair, a lot of new institutions or just influencers or whatever who are tied to making things up. And I think that when you get to the systematic production effort of actually trying to verify and source and evaluate large scale records, that imposes a discipline on you that doesn’t make it impossible, but that just by having this quality filter makes it less likely that’s the kind of story that you’re doing.
Jim O’Shaughnessy
You make a very logical argument. But the fact is we live in a world where a lot of people are very busy, they have their priors, and we all, myself included...
Brian Chau
Yeah, I love my priors, I’m a huge fan.
Jim O’Shaughnessy
We all love our priors, and that turns us into confirmation bias machines where we literally are blinded to information that conflicts with our priors. So let’s just say that’s the average voter, right, if we keep this in the political realm. So an unkind term would be low information voters, right? And nevertheless you could make, I think, I’m asking you really, I think you could cherry pick real facts, not “I made it the fuck up”, but real facts that are verifiable in, let’s say, one of these many government data sets or private data sets or whatever, and you could turn that into an incredibly persuasive story.
And look, I am 100% on board with AI. I think it is potentially the greatest tool that has certainly been invented in my lifetime. I’ve been waiting for it since I was in my 20s. I keep journals, handwritten journals, and I found one from when I was 22, and I didn’t call it AI, obviously, but this supercomputer, et cetera. But every technology is dual use in my opinion. And so I worry about the cognitive load that is coming for your average person. And I don’t know that we are adequately prepared for it, because as you said, the idea that logically you can go back and tie it into all of the footnoted, et cetera. By the way, I’m amused by most conspiracy theorists, because the real conspiracy is there really is no conspiracy except in filing.
Brian Chau
Yeah, it’s a conspiracy of incompetence. My least favorite phrase in the entire English language is “do not attribute to malice what can be attributed to incompetence.” Because if you’ve ever spent a moment in the real world, you would know that the most malicious people are also extremely incompetent. And very often incompetent people are also malicious. It’s not mutually exclusive. And yeah, I think you’re totally right. You’re totally right on this. You’re totally right on the human nature points.
Jim O’Shaughnessy
And so I think that there’s going to be a lot of potholes, there’s going to be a lot of things that were, whoops, we didn’t mean to do that. But rather than subscribe to the idea, no, shut it all down, make it illegal, et cetera, it’s like the example I always give is fire. Boy, fire was incredibly useful technology for we humans. I’ve read reports that when we started cooking our food, it’s what developed the prefrontal executive function of the prefrontal cortex. But fire is also really fucking dangerous. And so rather than try to ban fire, what we got was fire departments, fire alarms, fire extinguishers, fire exits, et cetera.
And I think that we’re going to have to think out really thoughtfully. We’re going to have to do this here as well, because I could definitely, I don’t know whether you saw the report that said that, I haven’t read the full report yet, so let me add that footnote, but that they were finding in multiple studies that large language models are actually more persuasive for a majority of people than human beings are. You do not have to be a very insightful person to think, oh, what would happen if I put that in the hands of my greatest enemy, right?
Brian Chau
Oh, but you should do a double click on those reports. I’ve read very many of those papers. It’s an area that’s fascinating to me, practically useful to me as well, as you can imagine. But do you know how the LLMs tend to be better persuaders than humans?
Jim O’Shaughnessy
No.
Brian Chau
You know, what are the causal mechanisms that they found? No, it is by being patient, not being condescending, ironically empathizing with the other person’s views. It’s basically by the same epistemic technologies that we’ve known since the time of the Greeks make for good persuasive rhetoric and good persuasive conversation.
Jim O’Shaughnessy
Yeah.
Brian Chau
And a lot of people just don’t want to put in the effort to do that. And that might be the frontier at the end of the day, is that the LLMs, no matter what your problems with them, are infinitely more patient, are by design infinitely more patient than humans are. And that’s going to come with, as you say, its own set of consequences, both positive and negative.
Jim O’Shaughnessy
And do you have any ideas about how that unfolds? And how...
Brian Chau
Oh, yeah. I’ve been thinking about this for so long.
Jim O’Shaughnessy
Enlighten me. Enlighten me.
Brian Chau
Yeah, I remember there were these two dueling camps back when I did AI policy in DC. I was one of the few people who were doing AI policy in favor of having more AI. This is a funny anecdote as well, but when we were deciding on the name, the org ended up being called Alliance for the Future. But when we were deciding on the name, we decided not to put AI in the name, because every single organization that had AI in the name was anti AI. This is an amazing thing. But that really gave me an insight into how policymakers were thinking about this.
And there tended to be two camps. There tended to be one camp which was something like, oh, AI is not a big deal. Maybe it’ll be good for GDP on net, but it will be fundamentally uninteresting, and therefore we shouldn’t regulate it. And that ended up being most of my allies in that situation. And then there was another camp that was something like, AI is going to fundamentally change the world, and therefore we must ban it.
And I don’t know if it’s because I’ve always been a pathologically optimistic person, but my way of thinking about it, that really solidified after leaving DC and building more with AI for myself, was something like, actually, the apocalypse was in the past. This is like reinventing postmillennialism, right? But actually that there has been this massive wave of antisocial and destructive behaviors, and actually most of that is going to be revealed.
There was this concept that was floating around for a long time of something like an encryption jubilee, a time in which we finally crack RSA, maybe that’s with quantum computers, maybe that’s somewhat with the assistance of AI, and we leak everyone’s encrypted messages, and then everyone can be okay with tolerating each other’s strange porn tastes and Google search history and whatever, right? And I think that there is something analogous with people’s either laziness or biased forms of argumentation, or even just, like we were talking about, the lack of effort that many political commentators have in really trying to persuade people.
And when the LLMs become widely diffuse, I think you’re going to see a wave of things that could be, that are on their surface incredibly appealing and invigorating, in a sense that just gets demolished on the thinnest pushback. A lot of those arguments are just going to vanish. Or are going to become these anti-markers of intelligence or of status. They’re going to be much better socially penalized, because any single person can go, you know, “@grok, is this true?” And there’s a certain class of arguments that will just completely be evaporated by Grok or by ChatGPT or Claude or any of these models. Probably even by these 7B models, these tiny models from Google or whoever, like Gemma, where even not even just existing technology, but technology that is well below the frontier of what we have now, can just demolish these arguments. And I think that the biggest societal change that we will undergo is unveiling all of this essentially clown behavior that has gone on for the past really centuries.
Jim O’Shaughnessy
Yeah. And I agree that I joke with people that we’ve always been living in The Truman Show, just now we are having the tools that say, oh, wait a minute, this all seems a lot more managed than we would have thought. And I think part of it came from this anomalous period, and I stress anomalous period, after World War II, through the ending of, I don’t know when the FCC changed the Fairness Doctrine, but we’re literally in America...
Brian Chau
It was under Reagan, right? It was...
Jim O’Shaughnessy
I think so.
Brian Chau
I think ‘82.
Jim O’Shaughnessy
Yeah, the 1980s. But I definitely think that...
Brian Chau
‘87.
Jim O’Shaughnessy
Thank you.
Brian Chau
It was second, second Reagan term.
Jim O’Shaughnessy
One of, I guess I would call it a bug, other people might call it a feature, of human OS is the desire for certainty and a hatred of uncertainty or chaos or what have you. And so we had this period between the end of the war and 1987 when in America there were three major networks that were the source of the news for the vast majority of Americans. There were two national newspapers, the New York Times and the Wall Street Journal. And that was it. And you think, like shooting fish in a barrel, if you want something buried and you have access to those three networks and/or those two newspapers, you can manage society in a much more seemingly orderly way. Now, that started to break down right when television went and started filming the shit we were doing in Vietnam. And that’s another example. And I’m wondering your take on this. Is what you’re doing in any way kind of analogous to what happened when we introduced reporters who, by the way, for much of Vietnam were not censored. They were not, they were just riding along, and the American public for the first time saw what was actually happening. And they’re like, holy shit, we don’t want any part of this. Could you see something similar happening with things like Effort News?
Brian Chau
Yeah, I think it’s totally revolutionary. And it was actually this unfulfilled promise of the Nixon era. The idea was “sunlight is the best disinfectant.” We’d finally have transparency and that would bring a whole new wave of accountability to governments. And funny enough, we got the transparency, at least much more than we had before. And that led to almost nothing. It led to the Bill Clinton sex scandal. And there was some degree that was useful in some degree, but it led to no real systematic change. And some people would argue, I think correctly, that there’s been since then a systematic decline in accountability in all sorts of government areas.
And I think in part that is because the infrastructure to take advantage of that transparency never really materialized. You had all of this dependency on sources reporting. And once again, this isn’t to fully condemn sources reporting, but especially if you’re censoring your own stories as an institution in order to get better sourcing, that itself would be something that is a lot more worthy of condemnation, where the transparency was there. But at that point the news organizations, the centralized news organizations, I think actually by that time both TV and print, were so attached to getting stories through source reporting that they’re no longer able to effectively comment and really effectively counterbalance the measures that were taken in place to control information. And that only became stronger in the COVID and the post social media platform era as well.
And I think that is changing for a number of reasons. Number one is just underlying technology, right? It’s a lot easier to make these queries. It’s a lot easier to just process a sheer amount of documents, many of whom will be hits, right? Or sorry, many of whom will be misses. There will be a much greater number of misses per hit, right? Where for example, if we’re just looking through every grant in the HHS database, a lot of those grants would just be fully legal, and not just fully legal, but fully non controversial stuff that we, at least on balance, I don’t know if it’s fully good as a policy goal, but on balance, do not see as contradicting the purposes of HHS.
And that cost benefit calculus has totally changed, where if it’s something like, oh, maybe we’ll get one scoop, but we’ll really have to burn through thousands of hours and hundreds of thousands of documents of things that we’ll completely not get anywhere with. Before it would just have been, you know, are you kidding me? Let’s move on to the next area. But now that can actually be investigated and actually investigated fairly diligently. So I think the overall trend, like I said, my underlying theme to all of this is that yeah, there are going to be problems, there are going to be misuses of the technology, but the overall trend is towards something better than what we had before.
Jim O’Shaughnessy
Yeah. And that brings me to the economic question I had for you. So if in the past there was a big story that a news organization wanted to own, so to speak, like the Boston newspaper that basically broke all the stories about the child abuse rampant in the Catholic Church, and they made a movie about it, et cetera, that you can quantify. You could say, okay, we need to hire six, a dozen, however many journalists, it’s going to take them six months, a year, okay, we can tote up exactly what their salaries are, what their overtime is, what the cost of travel, et cetera. But you have costs too, right? In other words, you have costs of inference, engineers, a variety of things. How cost advantaged are you over the model I’ve just described?
Brian Chau
The stories that we’ve broken so far, whether it’s the Preferred Communities grant, whether it’s the degrowth donors, all of them are stories that would have just been straightforwardly infeasible under the previous model, which would have, if you were to put a dollar number on it, and it would probably not even be feasible to just recruit that many journalists. It would be this kind of market that eats the entire old market. It would have been like upwards of tens of millions of dollars at least, if you could even physically assemble, like find that many journalists who had those competencies in those areas, to pay. So it is a total sea change in terms of what is capable and what is cost.
Speaking of cost, though, this is actually more of my traditional educational background. I had a degree in pure math and started in machine learning at a really young age, basically when I was still in high school. Anyways, I had been a machine learning engineer before, and this is mostly what we optimize on, because a lot of the harnesses make extremely inefficient queries to databases. It’s shocking, because at the same time I can look at some of the cyber security applications, I can tell that there are a lot of things that the LLMs are a lot better than me on. But in terms of making API requests, not all of them, but many of them, or making efficient database queries, they just will put out tons of fucking slop queries and will consume way more tokens, will just take longer, will have higher latency on these tool calls. And there is just a lot that you could optimize for in many dimensions.
That’s part of what’s so exciting about this. And that’s been super exciting for me. It kind of points to that environment that a lot of people talk about, this is actually before my generation, but of the early Internet, where it seemed like everything was so unoptimized. You could just sit down and think about it and you could optimize these things. You as basically anyone who was reasonably smart, who had a computer, was reasonably familiar with these things. And that’s just this massively inspiring thing. And that’s something I feel like we’re picking up a ton of momentum on every single day on our team.
Jim O’Shaughnessy
So what about the idea that journalism has been misclassified for a century? And by that I mean, I think of H.L. Mencken, right? He was the king of the muckrakers, and he was always great with his quotes and everything else. But I, so at least I, maybe others don’t, but I sort of associate journalists with writers, with the actual act of writing or producing, if it’s TV or radio. But maybe they were actually search and verification engines, and using writers was just a hell of a lot cheaper because we didn’t have the technology. Your model changes the economics of finding needles in haystacks, right?
Brian Chau
Yes, that was one thing that we thought of putting on the website. We find needles in haystacks.
Jim O’Shaughnessy
But my question and pushback on that is, if I was your opposition, right, if I was trying to get up to something nefarious, why wouldn’t I, if I have my own tools that are at least as good as, if not better than yours, why wouldn’t I just make a fuck ton more hay and make it much more difficult to find the true needle, so to speak?
Brian Chau
Oh, it’s already happening. Yeah, I think that’s totally fair, and that’s totally happening. And actually that’s been happening for 30 years. That is the intelligence agency, Defense Department, that is the documented strategy of just producing so many documents, of maliciously complying with FOIA requests. That’s totally happened, and I think there will be another escalation of this. Imagine you make a FOIA request to the CIA and they just give you petabytes of AI slop. That could totally happen, and I think that probably will happen. So I don’t really have a solution to that necessarily. To some extent we will have to AI harder, and it is this Red Queen’s game where you’re staying in the same place, but you’re running faster and faster.
I think in some sense we can outcompete the CIA, at least in this narrow area, or we can outcompete all of the people who would otherwise be trying to obfuscate information. And in part that’s just the pace of technological development. Of course, I think a lot about incentivizing our organization and building up our capacities in that way. In part that’s innovator’s dilemma. There are all these tech theories and why the incumbents will always win, or why the startups will always beat the incumbents. It’s always treated as this kind of predetermined natural course of history thing. I think we just have to win. I think we just have to fight as hard as we can. And I think you’re absolutely right that it’s not just going to be us, it’s going to be people on both sides. It’s going to be people both maliciously complying, maliciously producing information, subverting transparency laws, or at least kind of maliciously complying with transparency laws. That’s totally going to be happening. And at the same time we’ll just have to race on faster and faster.
Jim O’Shaughnessy
Yeah, but again I come back to good old human OS, and I don’t know anyone who reads the terms and conditions before clicking okay. And I agree with you, by the way, that the patient large language models will read all of them and they will point out all of the ones that go against you. But I’m worried about the people, you know, they’re busy with other things in their life, right? They’ve got a family, they’ve got a job, they’ve got hobbies, et cetera. And meanwhile I kind of see this massive battle between two philosophical points of view, if we keep the extremes, right, and the weapons are getting a lot more precise.
And I just wonder, and it’s like, again, the Pandora’s box is opened, and so trying to say no, we want to rethink, that’s not going to happen. But I do wonder if we should spend a little bit more time on cog sec, cognitive security, right? Like maybe freely disseminated. If we could figure out a way to really be able to judge deepfakes, for example, wouldn’t it be cool if you just freely distribute it to everyone, something on their phone that was like, the minute a video showed up, it just went “deepfake” across it. At least that would condition them to start thinking. Oh, you’re absolutely right about history. People have been manipulated by religions, by political ideologies, by a variety of incentives for all of human history. That’s nothing new. But now we’re doing it at weapons-grade and it scales. And so I definitely think some form of cognitive security for everyone might be a good idea.
Brian Chau
Yeah, you’re definitely the best kind of critic, Jim, because I think I really share the same fundamental view of human nature. I have a very Hobbesian view of human nature. Hobbes is still my favorite philosopher to this day.
Jim O’Shaughnessy
Brutish and short, right?
Brian Chau
Yeah, and so much more than that. Both the dual powers of the sovereign, the origins of human reason, goes so far. But I think you’re right in that I think that you want to draw a distinction between the scale of deployment, where you’re absolutely right, it’s bigger than ever before, and the equilibrium harms. I think something that every major tech policy fight has sort of gotten wrong in the past few decades has been this conflation of the possible harms and then the equilibrium, where people will say this, for example, with cyber, that’s something that’s more played out. People will say, oh, there are so many potential hacks. And then you ask the straightforward next question, which is, well, how many of those potential hacks will be prevented by AI? And not only that, how many of the potential hacks from past generations, or just possible with previous levels of software or previous levels of human hackers, will be prevented by AI? And then that equilibrium, that’s not zero, right? That’s not zero harm, but it’s very different. And man, this could really be a long answer. I will try to keep it from being like a 30 minute Hobbes monologue.
Jim O’Shaughnessy
Yeah, yeah, no, obviously, let’s try to make it a little more compressed. But I think this is important. I think that people who are in favor of AI and all of the tools and the powers that they unlock need to examine and acknowledge the other side of the argument. Don’t get me wrong, I’m maybe one of the most pro AI guys out there in terms of the potential for what it can do. We just made a movie about how the Vesuvius, the volcano that destroyed Pompeii...
Brian Chau
Is this the scrolls?
Jim O’Shaughnessy
Yeah, yeah. And love the scrolls. We made a movie about it because to me it’s the coolest thing in the world. It required a particle accelerator and AI to be able to translate these scrolls, but now we’re able to transcribe ancient Greek and Roman scrolls that if you touched them in the past, they just disintegrated. And I’m giddy about the use cases for AI in terms of expanding our knowledge, expanding innovation, expanding discovery, et cetera.
But I definitely think that you need to be, I call myself a pragmatic optimist. And by that I mean I think you need to consider the unintended consequences and try to come up with good solutions for them. Because if you don’t, the people who are on the other side of the argument are going to make hay with those, and they’re going to say, yeah, but, right, like. And most of the objections are going to be way deep in the tail and very unlikely to happen. But if we who are in favor of this new expansive rocket ship for the mind that AI is. And also I hate the artificial intelligence. Brian Roemmele calls it intelligence amplification, which I much prefer, because really that’s, in my opinion, one of the things it’s doing and/or uncovering, right, like you’re doing. But I do think we need to think about these other use cases and have good ideas, because it’s going to happen.
Brian Chau
Oh, absolutely. It’s totally going to happen. And I think one of the most important questions for any philosopher to ask is to step back and ask, is this a matter that will be decided by philosophy, or is this a matter that will be decided by action? And to a large extent, I think that the part that has been decided by philosophy we’ve already talked about, it’s kind of the nature of the story and that kind of epistemic field of what comes. And the part that will be decided by action is essentially, what are the names in the media ecosystem people will trust, what are the solid ground, right? And that can never be fully reliable. You shouldn’t put your absolute faith in any kind of news reporting or any kind of human institution in general.
But you have to have something, and I think about a lot of that on a daily basis as well, where Effort is not merely trying to produce a good technological platform, right? You’re not merely trying to say, like, oh, here is this technology, everyone will have this technology, voila. I’m really concerned, and I think this is very different than a past generation of more optimistic, more maybe Rousseauian founders, even. I don’t know if they’re explicitly Rousseauian, but implicitly Rousseauian. But I’m a lot more concerned about institution building, that this is the specific group of people who will be employees and founders and contributors to Effort. This is a specific kind of culture that I want to calibrate on epistemic openness and of hypothesis testing and of the scientific method and just a very high standard for everything.
Not publishing slop. That’s the most important thing. We do not publish slop, and we do not publish slop in all sorts of ways. We do not publish slop in that we do not publish AI writing, at least at this point, writing still, at least in our experience, not great. We rigorously, we human write all of the articles and thoroughly check for any improvements we can make to the phrasing, any improvements we could make to the clarity. We also, of course, don’t publish slop in terms of records. We verify all of the records, both with a lot of AI tools and by hand. We don’t publish slop in the sense that we don’t bury stories, but we want to make sure we take our stories to completion. We don’t want to just be wildly flinging accusations. We want to get to the point where we have a clear border of, this is what we found, this is what we didn’t find, here is how we can prove what we found, here is the steps that led to us not finding these other things.
And I think that at the end of the day, part of that will be decided by technology. Of course, we would not be able to do most of what we’re doing now without it. But there’s also a large part that will inevitably come down to a form of human judgment. And that is something, in my opinion, that you have to be proactive about. You can’t just completely leave the human judgment to the wind. That you have to be proactive about as we’re building out Effort. And to me in the long term there is no perfectly philosophically consistent defense against all bad actors. That’s just not feasible. Maybe there’s a solution to it. Maybe GPT 7 will come up with one. But it is just beyond me and it’s beyond every philosopher in history. But the practical solution to that looks more like the institution building that I think a lot of people in the previous generation have neglected.
Jim O’Shaughnessy
Yeah, and that’s an interesting point, because it also leads to objectivity. I think that’s a term that we like to use a lot and misuse mostly, because I think that your organization, for example, right, like I would love to see that you don’t get captured because you’re a subscription model, right? There’s always audience capture. And are you dedicated to exposing whatever kind of corruption and/or story that got buried regardless of which political side it helps?
Brian Chau
Yeah, for sure. And I think that at the very beginning you’re going to see actually a different kind of selection bias, which is that we’re reporting on essentially stories that have been dropped. Because for example, the refugee resettlement story, that’s something that could have been reported on for literally five years. So the fact that we are able to report it is a direct consequence of it not having been reported. And honestly, it’s funny because this is going to be a justification for stories that are perceived as more right coded. But honestly it’s because a lot of the right wing media ecosystem are just a bunch of muppets who do not seriously complete their investigations and are comfortable with just publishing slop, publishing accusations, not actually digging into the documents.
And I mean, digging into the documents may not be completely easy, especially without AI. But it is just this completely different equilibrium where, you know, there are obviously biases with the New York Times, but they do actually do investigative reporting at least, and they do actually break stories, and they break stories in a certain direction, and that’s going to leave actually a gap that is not going to be equal. So I think that in the short to medium term you should expect that to be unequal via negativa.
Let me give a more concrete example related to that, though. We’re going to be breaking a story, I don’t know exactly when this will come out, but we will be breaking a story within a few days, reporting on something that the Trump administration has been trying to keep secret, which is the whole Freedom Fuel network. Do you know about this? You know about Freedom Fuel?
Jim O’Shaughnessy
No, I don’t.
Brian Chau
It’s this fascinating mystery. I think it’s been this fascination of journalism world because it’s this particularly enticing thing. Essentially the Trump administration has been promoting this series of gas stations in New Jersey and Pennsylvania that have been selling gas almost certainly at a loss. They originally sold at $3.47, right? This is because Trump is the 47th president. And they’ve in general been somewhere like 40 to 50% below the going gas rates in those areas, almost certainly losing money compared to selling that gas even on the open market. So a classic case of fascinating mystery.
Jim O’Shaughnessy
Yeah, and economists would call that dumping to gain market.
Brian Chau
Yeah, yeah. And so there’s this fascinating mystery of where’s the gas coming from. And all the king’s horses and all the king’s men have tried to figure out where the gas is coming from and they have not figured it out. And we have not, I think we have meaningfully moved forward the Freedom Fuel case, I’ll put it that way. Meaning we’ve not fully found evidence of government subsidy, but we have found fairly conclusive evidence of self-dealing.
Jim O’Shaughnessy
Interesting. Of the stories that you’ve dropped so far, which one got the most reaction, both positive or negative? And then I do want you to, because when I was getting ready to chat with you, I read your story about how to hijack the new X, or as us old timers call it, Twitter, algorithm.
Brian Chau
But wait, am I an old timer?
Jim O’Shaughnessy
Maybe in spirit, maybe learning a lot today. But which stories are the ones that really the lightning hit the rod?
Brian Chau
Yeah, it was definitely the refugee resettlement story, because this one was just something where the financial record was so damning, and also where the players involved were these already very big media entities. US Conference of Catholic Bishops, Church World Service, HIAS, Episcopal Migration Ministries. For a long time they’ve been trotted out. They are the most public of public figures or public entities. They’ve been trotted out to be very critical of any kind of migration restrictions.
And for literally, for upwards of five years in terms of the grants that I reported on, but upwards of 10 years at least for the total body of grants, it had just gone completely unnoticed that they were being majority, that the vast majority of their revenue, 81.8% in the case of USCCB, 89% in the case of Church World Service, was being funded by these resettlement grants. In other words, they were explicitly benefiting from the policies they were advocating for.
It would be like taking a dairy farmer and having them be the object, treating them as the objective moral arbiter of whether the government should subsidize milk. It’s like, there are these well known names who go on TV every time there’s a milk policy debate. And they’re treated as the representatives of Christian Americans worldwide, or Christian Americans across the country. And they come out and they say, “we morally condemn in the strongest terms the President’s opposition to subsidizing milk.” And then it turns out five years later that all this time they’re selling milk. It’s like, what? This is a totally crazy story. And it also goes to show, despite all of the media attention, despite all of the endless commentary on these organizations, for a long time no one went and actually dug into the financials. And now we did, we finally done it.
And yeah, the reaction to that was overwhelming. Interestingly, there’s a certain type of story, there’s a certain type of, I mean, I don’t want to be mean to any individual specifically, but as an institution, there is a certain type of New York Post story that very effectively draws kind of counter criticism, that breaks a real thing, but also taints it in a way and fills it with other sort of inaccuracies or bias, or basically gives this massive attack surface to critics of that story. And of course the most famous one of these was the Hunter Biden laptop, which at its core was a truthful story, it was a real finding, but was distorted in all of these ways, which eventually became used in order to weaponize social media against it.
Anyways, we got very little of that with the refugee resettlement story. I think actually for that reason. And there’s actually a perverse incentive there, I’m going to do my best to avoid it. I think that we shouldn’t be lowering the quality of our story intentionally in order to get more counter reaction from the other side. But I think there’s a certain type of story that gets a ton of negative counter reaction because it, I don’t know if it intentionally introduces flaws, but because of the flaws that are there. And I think that this was sort of the opposite kind of story, where the financials were so clearly laid out, the conflict of interest was so undeniably established, that the other side didn’t even want to touch the story.
Jim O’Shaughnessy
Interesting. And we’re talking a lot about incentives, right? And another point of view that is kind of consistent with my quip about we’re all living in The Truman Show and always have been, is like, what percentage of, in quotes, news is just propaganda, and it’s literally the various sides vying for power trying to win the news cycle? Like that’s another one, winning the news cycle. That’s as old as we have the media, right?
And I just wonder, one of the things that I do worry about is all of this is so new. And I’m not just talking about AI, I’m talking about, Bill Bryson has a great book called America One Summer, in which he’s talking about the first time a single human spoke to more than a thousand humans, right, was the guy on the radio announcing Lindbergh’s return from flying across the Atlantic. And so telegraphs, like, it’s in the scheme of human history, all of these new technologies are brand spanking new. And now we’ve got, we’ve gone from spears to thermonuclear multiple warheads really quickly. Is there going to be some form of way that humans just...
Brian Chau
Yes, I think it’s all resolving now. I think it’s all not resolving to a final conclusion, but a lot of that built up technical debt is now being reevaluated, and that’s the conflict we see. That’s what people are angry on social media 24/7. And not all of that is directed in a productive way. A lot of that is directed towards slop. Those criticisms are totally fair, but largely what I see is a reevaluation of those fundamentals. As Nietzsche called it, “a reevaluation of all values.”
And I think where that starts is something I call the crisis of nominalism. There’s this big phrase that goes around. You’ve caught me at a great time, because I was just on a four-hour hike and I was thinking about this and I have all these thoughts fresh in my mind. But there’s what’s called a crisis of liberalism, which I think is really mislabeled. And the theory behind calling it a crisis of liberalism is like, oh, there’s a ton of premises, but the story is like, oh, we’ve lived in what we now call liberalism for roughly 250 years, basically since the American founding. And that is now coming to various flaws, whether that’s fertility, whether that’s social media, whether that’s foreign influence, whether that’s migration. And what people will say is that liberalism is collapsing under its own values.
And I think a point by point deconstruction of that claim, almost every premise is false. Where we are not living under, we are no longer living under liberalism. It’s not a continuation of older values. Liberalism is not causing the issues that we see. And actually it is not even a crisis. It is a reevaluation of a slave decline.
And I can get into each of those points specifically, but let me try to give a core, something that’s like a useful takeaway from this. Is that nominalism, what it actually is, is people taking a name for granted, right? This is science. It was published in a title, in a journal called Science, therefore it must be science, right? And not actually going looking underneath. Is this the scientific method? Are these neutrally tested hypotheses? Were these in a controlled environment? What was the statistical probabilities, that this was actually a relevant finding, and so on and so forth. That at the very surface level, you kind of look at the cover of the box and you never actually look inside the box.
That problem has persisted. It has persisted, certainly scientifically, it’s persisted in news media, in both the older institutions and the kind of influencer slop type media that’s come up to rival it. Both of those domains are domains where you kind of just get the slop and you never look inside the box.
And this is really the most disastrous version in spending. And you get this in a lot of government spending. I’m very sympathetic with the DOGE people, where they had a complete communication failure and they just got so internally overwhelmed that they ended up being driven crazy and not really being productive in their political actions. Where if you are someone who’s had experience putting together financials for a startup, I’ve done this both in my own company and with previous companies that I’ve worked for, I’ve done this personally, and you are used to that level of accounting, and you know that if you do not present this quality of information to auditors, you will go to jail. And you see the standard that is set in Treasury or in these other government departments, then you will have a meltdown. It is just a completely different standard. It’s a total double standard.
And they were expecting to go to the media or to go to Congress with this, and for them to react with, oh my goodness, this is a total, this is like a five-alarm fire. This is something that, if you were done at a private company, you would go to jail for. This is some crazy stuff. And they were completely not prepared for the reaction that they got, which is, this is how things have always been. And both of these sides are kind of acting rationally in their own narrow context. It is simultaneously true that if you were doing this in a private company or public company, if you’re doing this in any kind of normal corporation, you would go to jail for this. That’s both simultaneously true. And the kind of protests of the senators of, this is how it’s always been, this is maybe not good, but there’s not much we can do about it, this individually, I think that’s also true. So it was this huge kind of Shakespearean tragedy that played out.
But that’s a good illustration of what I mean by the crisis of nominalism, that we have these names, right? We have these names of accountability or auditing or financial record keeping. And under the surface they’ve completely diverged. Under the surface you actually look inside the box and it’s like, what is even going on? It is this chthonic mass. And a lot of that is going to be reevaluated. A lot of that is going to be either split into different names, or there is going to be this type of philosophical colonialism where people who have been more competent over the past 10 or 20 or 30 years will go in and have to clean up these institutions. Part of that is also this new trend of software roll-ups, which could be a completely different rabbit hole. But there’s a lot that’s going to shift, and that’s going to hinge both economically, politically and philosophically on this crisis of nominalism, of the big question of what happens when you finally look inside the box.
Jim O’Shaughnessy
Well, you make, now we’re down to linguistics in my opinion. So for example, I think it was Kierkegaard said, “when you label me, you negate me.” And you were using words like accountability, et cetera. Of course, good old George Orwell understood that a long time ago when he talked about Newspeak and, you know, “war is peace, freedom is slavery,” et cetera. And then, of course, the classic Bill Clinton under questioning during the Monica...
Brian Chau
“What do you mean is?”
Jim O’Shaughnessy
Yeah, it depends on what you mean by the definition of is, is right? But yourself, you’ve said slop I don’t know how many times. You see, we’ve always had slop. We, except it was always just human slop, right? Of the millions...
Brian Chau
When I say slop, I am inclusive. I am not discriminating based on species.
Jim O’Shaughnessy
I understand, but I want to say we get captured. One of the dangers of labels, in my opinion, is it puts thinking aside, right? You label something slop, it goes into that bucket. We call it slop. You no longer really try to disambiguate, is this really slop or is this real? I mean, would Joyce’s early works get put in a slop bucket? Probably. You know, Finnegans Wake and Ulysses. And it was because he was so new, right? His style of writing was so new, people literally didn’t know how to read it, right? And now we know how to read it now, because we evolved into understanding that kind of postmodern way that Joyce approached his novels.
But I just think that I worry too, because labels, the power of labels. You yourself have just made an excellent case, right? Accountability, transparency. When I hear transparency and accountability, I think, okay, wow, it’s going to be accountable, this is going to be transparent. And so the power of language in and of itself, the semantic meaning, the semiotic meaning, is something that we’re going to also have to kind of dig into.
One of the ideas I had would be, how cool would it be to have an AI companion that was, I’m calling it the Reader, right? And when you’re reading a news article, it just points out, like, factually incorrect, appeal to emotion. I don’t know if you’ve read Influence by Cialdini. We talked about it a little earlier. And I joked one of the reasons why the large language models were pretty good at it was because they’ve been trained on the entire corpus of influence. And Cialdini is kind of the king with his book Influence.
But I definitely think if you had a Reader there that was saying, like, who benefits from this? I used to say to younger people, and they’re asking me, like, how do you navigate reading the news, this pre AI, I was just like, ask yourself the question, who benefits if I believe this? What are the objectives of this particular writer? Are they for a particular outcome? Are they just informative? What are they? But if you actually had a reading companion right next to you, or a watching companion, right, I mean, if we had this, I’d put it on this podcast and it would probably contradict me a lot. Well, Jim is doing an appeal to emotion. Jim is doing an appeal to authority, all of those things. But I definitely think that organizations like yours, that’s why I was so excited to talk to you. You’re the beginning stages of this unfolding.
Brian Chau
Yeah, I was just, so when you talked about Joyce and his early works, I immediately flashed back to this experience I had two months ago. So I’m going to read something and tell me what you hear.
Jim O’Shaughnessy
Okay?
Brian Chau
“The old professor, em dash, Calguès by name, em dash, aimed his glass at one of the ships still lit by the sun, then patiently focused the lens until the image was as sharp as he could make it. Like a scientist over his microscope peering in to find his culture swarming with the microbes he knew all the time must be there. The ship was a steamer, a good 60 years old. Her five stacks straight up like pipes showed how old she was. Four of them were lopped off at different levels by time, by rust, by lack of care, by chance, em dash, in short, by gradual decay. She had run around just off the, she had run aground just off the beach and lay there, comma, listing at 10 degrees like all the ships in the phantom fleet. There wasn’t a light to be seen on her once it was dark, not even a glimmer. Everything must have gone dead, em dash, boilers, generators, everything, all at once, em dash, as she ran to meet her self-imposed disaster.”
This is from the second page, I believe, of Camp of the Saints. And I was going to read this, I was going to reread this, and I just couldn’t do. Was too AI coded for me. It just broke my brain. I know it’s obviously not AI, but it’s just like totally broken, broke my brain.
Jim O’Shaughnessy
It’s so funny, serendipitous. I was just rereading, or trying to reread that book. For people who are not familiar with In the Camp of the Saints, it was a book written, I think, in 1970s, right?
Brian Chau
Yeah, a long time ago.
Jim O’Shaughnessy
And it envisions the very migration problems that we are having now, done differently, if my memory serves. I read it back then and was horrified. And I tried to, I picked it up again and, like you, I’m like, wow. But I had a different reaction than you. My reaction was, why do large language models? And if you look at my Twitter account, you’ll see that I have used em dashes copiously my entire, I’ve written four nonfiction books, obviously all before AI. You’re going to find a lot of em dashes in them because I love them. And so on Twitter I say, you can come and pull my em dash out of my cold dead hands.
But isn’t it fascinating, right? It’s like, the same with the, it’s not X, it’s Y. Go back and read mystery novels and things. The reason they use that technique and the reason they use em dashes is because of the training literature they were trained on. We use em dashes all the time previously. And you have the Orient Express. Like it’s all, he wasn’t this, he was this. You know, when she’s describing the villain. And so it’s another part of that axiomatic reaction, right?
Brian Chau
I think I want to add a little nuance to that, because I think that these are actually different cases. I think for em dashes, you’re mostly right on. That is just part of the training data. For the, it’s not this, it’s that, that’s more downstream, I think, of a particular technique called, oh my goodness, just struggling to pronounce this correctly, called contrastive learning. There we go. Called contrastive learning, where the idea was that the models would have a better directional update if they were given both positive and negative examples of something. And this was a specific technique published in papers by OpenAI, by Google, all of the biggest names, that specifically use this format, and was actually effective, at least in the short term, for post training their models. So I think that those two cases actually have slightly different origins. But I think the underlying idea, if you, you shouldn’t necessarily dismiss these literary patterns, that they’re legitimate literary patterns. I think the underlying idea is right.
Jim O’Shaughnessy
Yeah. I think if you are an Agatha Christie fan, that is who I was referring to when I talked about Murder on the Orient Express. She does that. It’s not this, it’s this, all throughout her books, right? And because she wants to surprise the reader, right? You thought that he was a good guy. No, he’s actually the villain. You thought he was steadfast. You thought he was noble. No, he was the villain.
And so I just think that there’s a lot of ignorance about the idea that the models do what they do ultimately, right? Agrippa’s trilemma. You familiar with Agrippa’s trilemma?
Brian Chau
No, I’m not.
Jim O’Shaughnessy
Well, that would take us forever. But the short version of, he destroys logical systems because he ultimately, and Lewis Carroll, a mathematician and logician, actually wrote a book called, not a book, he wrote a piece called What the Tortoise Said to Achilles, and it made the case even better than Agrippa’s trilemma. But basically it’s like every time you have a logical rule, you gotta say, well, why do we have that rule? And then you keep going down, down, down. And you finally get to that base rule where you said earlier in our chat, I just made it the fuck up. But in Carroll’s What the Tortoise Said to Achilles, is ultimately everything is a human choice or a decision, at least now, right? This may change, right, going forward. You mentioned ChatGPT 7 or 10 or 12 or whatever. But for now, and historically, Agrippa’s trilemma is the problem of the infinite regress of. There’s a, let’s, it would take us an entire different podcast to talk about Agrippa’s trilemma, but I think you would definitely be interested in it.
Brian Chau
Yeah, I’m reading it now. Yeah, it is this, it reminds me a ton of formalized set theory. This would be okay. Like it would not be entirely surprising to me, but it would be like a life changing moment, I think, if the LLMs discovered a reduction in the set theory axioms. That would be a moment that I’m here for. It would be like my personal 9/11 is when AI discovers a way to reduce the set theory axioms.
Jim O’Shaughnessy
Yeah, well, I mean, again, because I am very optimistic about the, even knowing that we’re going to have a lot of bad with the good, I definitely think that the power and discovery that AI allows us to use as tools. That’s the other thing that drives me a little crazy, right? Like, I was looking up the thing on our system which we’ve trained on a bunch of different things. And you had said like 80 to 100% of news is sourced, going to the newspaper. Which, well, it gave me the whole history and it was like, yeah, media scholar Jim Macnamara reviewed 150 to 200 studies and he found that up to 80% of media content is sourced from or significantly influenced by PR, with estimates of 50 to 75% common. But then it keeps going. It goes back to Herbert Gans’s classic 1979 content analysis, and it gives me all of the notes that I can actually go and go to the actual book or study and find. And I definitely think that for a certain type of person, the inveterate rabbit hole diver, that is really cool.
My question is, is it cool for non nerds? Like, I’m totally willing to call myself a nerd, right? And I love rabbit holes and I love doing all that. And I would guess that you would be coded nerd.
Brian Chau
Yeah, sure, yeah.
Jim O’Shaughnessy
But I just wonder how much of a difference it’s going to make to the non nerds. In other words, I think of the flat Earthers. If you could do a news story based on all of the data and convince the flat Earthers, right, that would be of interest to me. There, again, we come back to kind of the scientific method, right? If it’s not falsifiable, it’s a religion, in my opinion, right? It’s a belief system.
And what I’ve seen, historically at least, and I don’t mean to pick just on the flat Earthers, there are a lot of cult like beliefs that are patently untrue, but when confronted with data, and I made my career in asset management on doing algorithmic investing using data, so I’m a big believer. But one of the things you see is, here is Mr. Flat Earther, the million reports showing you why you’re wrong. And what happens is, in many cases they double down on the belief and they’re like, yeah, no, that, even though you can show every footnote, every study, everything, now that’s all wrong. Aggressive, I don’t know, I don’t want to say stupidity, but aggressive disbelief. Do you think you ever change those minds?
Brian Chau
I think on operational grounds, on anything where it’s like, oh, you actually have something at stake, you can change those minds. I think when it’s this kind of like, when it’s this Truman show thing and it’s pretty clear that, you know, if you think the earth is flat and the entire system of physics that’s been developed around a circular earth is wrong, like, do you avoid flights? Like a lot of the time the answer is no. And the ones who the answer is yes, actually I would probably respect more, because of course that entire system is based off of an assumption of a flat Earth. Not an unfounded assumption, of course, but that all comes along with the system in play. And so you have this really fake virtualized game of debating about the flat Earth.
And I think in that environment you’re never really going to get anywhere with most of the time. And maybe the AIs will have enough patience to do that and eventually win someone over. But on all practical grounds, that is actually not a particularly effective use of our time, right? Like the flat Earther is not actually particularly harmful to anyone, and in a lot of cases can go about their own life without that cognitive dissonance, right? They’ll go on the flights, they’ll talk to their round Earth colleagues, they’ll in a lot of cases actually work a stable job and so on. And that’s basically fine.
And actually I think there will need to be a slow political reorienting around kind of what Leo Strauss calls the cave below Plato’s cave, right? That we have all this scientism. We’ve now gone to the extent of applying the scientific method to all sorts of ideas and concepts and philosophies and all kinds of new moral rights that are not the ground of the scientific process, are not falsifiable, are not things that you can decide via microscope, and are slowly being unrolled and seen as such.
And I think that political process involves actually, it’s interesting, this is kind of a, I think it’s a tension, but not necessarily a contradiction. We can have the philosophical maturity to realize that there are forces acting on the world that act in opposite directions at once, or that the different forces will act in different directions even if they’re acting at the same time. But I think there’s both a tendency towards developing greater standards for truth, as well as actually an admitting that there are these issues that have been treated as truth or as treated as scientific that are really not scientific at all and are rather more moral or theological issues.
And I think, to add one more thing on this, I think that there’s a big unraveling that goes back to Mill, who I think was actually a traitor to classical liberalism by redefining the goal of classical liberalism around consensus. This was the classical Mill’s trident, right, which was an argument for free speech that I think ultimately undermines free speech when it comes to the modern day. And the idea was, okay, if you think that what someone is saying is wrong, then they should be allowed to say it, because then the free speech, the debate will prove them wrong. You’ll have good information to beat the bad information. And if you think it’s correct, then obviously they should be allowed to say it, because they should be allowed to tell the truth. And then if you think that it’s somewhere in between, or if it’s uncertain, then it’s either case one or case two. So you should still let them have free speech.
And to be clear, I’m for free speech. I’m not anti free speech. But what this did was it redefined the old classical liberal rights established by Locke, and I think established before we even had the term classical liberalism by Hobbes, which was much more around survival and around peace. That the idea was that you would have classical liberalism, you would have freedom of speech, freedom of conscience, the acknowledgement of reason, and you would have that under the grounds of avoiding war. That there were these intractable beliefs that people would have, particularly around religion, that if not dealt with in a mutually tolerant way, would inevitably lead to war. And this was not a hypothetical, of course. This was the Hundred Years War. There’s many religious wars before that. And the idea was that you would have these natural rights that were enshrined in a mutual understanding, that if we did not have these rights, it would be a violent free for all.
And that slowly got displaced with this idea of consensus, that if we have these rights, then not only is it possible, but it is inevitable that we will reach a total consensus. And that broke down in all sorts of ways. Most notably that people realized that wasn’t what was happening. That as social media gave more people not just the right, but the potential to speak, that there was less consensus. And, you know, as we talked about at the beginning, some of those people were just making shit up. But it actually undermined this new theory of classical liberalism, which I think was not a theory of classical liberalism at all. It turned into a theory that was deeply antithetical to classical liberalism.
Jim O’Shaughnessy
You know, Brian, we could make this how long? I’m not a listener usually, but how long does, like... Chris Williamson. Well, I do listen to Chris.
Brian Chau
This could be the eight-hour Lex podcast.
Jim O’Shaughnessy
Yeah, this could be the eight hour. But I am getting my hook for my producer. So what I will do is, we’ve left so much left to be discussed. We’ll arrange for you to come back on. But for now, first off, tell everyone where they can find Effort News and how they can subscribe.
Brian Chau
Yep, effort.news is where you find us. effort.news, slash subscribe. And to find some of the news stories that I’ll publish on socials, you can just find brianchau57 on all of the socials, like X, Instagram, so on. That’s B, R, I, A, N, C, H, A, U, 57.
Jim O’Shaughnessy
Perfect. And next time around we’ll have to talk about all these other fascinating things, because I think we are in a very tumultuous and yet fascinating and innovative time. And I think the more we talk about these things and discuss them, the better people can understand them and reframe them.
You know, we didn’t even get into the difference between the real scientific method and scientism, but I guess the shorthand there could be, if anyone tells you the science is settled, they’re an idiot, because science is never settled, right? It just keeps like, yeah, this works, this works. Oh right. Newtonian physics worked and worked until we, the quantum guys came along and said, actually. So it’s an ongoing process, right? And it should be falsifiable, it should lead to better explanations, it should lead to more explanatory power. But it’s a process, right? It’s not anything that is settled, right? I often think of the true scientific method as being like pure punk rock energy. It’s like, no, I’m not going to take your word for it. No, I’m going to actually...
Brian Chau
Exactly, exactly.
Jim O’Shaughnessy
If you’ve seen or listened to the show, Brian, you know that we do have a final question for you. And that is, we’re going to make you the emperor of the world. You can’t kill anyone, you can’t put anyone in a re-education camp. But what you can do is, we’re going to hand you a magic microphone and you can say two things into it that are going to incept the entire population of the world. Whenever their next morning is, they’re going to wake up and they’re going to say, you know what, I just had two of the greatest ideas, and unlike all the other times, I’m actually going to act on these two ideas. What are you going to incept in the world’s population?
Brian Chau
First has to be the crisis of nominalism. I think that’s right. You have to ask the question of what’s in the box. And not everyone will come to good answers. Some people will ask that question and they’ll answer it by making stuff up. But I think even asking that question, a lot of things cannot survive contact with that question. And almost all of them that cannot survive contact with that question of, is this nominalism, should not survive contact with that question.
Second thing is, I think, maybe this was primed by listening to your podcast with Jonathan Bi, but I think that’s like a question for everyone, and then a question for people who are already doing things about it or who have set out to do things about it, is more like the question we started on. What is the domain of philosophy versus what is more the domain of action? Where there are definitely problems that can be solved by philosophizing around them. I love those problems. Those problems are great in theory. But there are also tons of problems where you just need to go out and do the thing. And I would have them very clearly distinguish, what is the thing, which of these problems are problems of action versus problems of philosophy.
Jim O’Shaughnessy
Both good ones. On the nominalism, it’s funny, my son and I were just having a long conversation about, like, what is wrong with people. And one of his observations was, they have sufficiently made it so that they don’t have to come into contact with reality. In other words, they have sufficiently made it so that the box is there, as you put it, and they’re not opening it. They’re not going to look in that box. They’re just quite happy with assuming what’s in the box and saying, yeah, what’s in the box is true. And that is not going to get you very far. You’re going to have a. Yeah, it’s...
Brian Chau
A very Lovecraftian theory of knowledge, right?
Jim O’Shaughnessy
Yeah. And you’re going to have a look. George Box said “all models are wrong. Some are useful.” If you don’t upgrade yourself to a more useful model, you’re going to suffer, right? If you, as my son puts it, refuse to engage with and contact reality, guess what? Things aren’t going to go your way. And I just put up on Twitter the other day the Jed McKenna quote. If you’re having a little tea party between Lord Lion and Lady Gazelle and somebody comes along who doesn’t buy your fantasy narrative, it isn’t that they are mean, it’s just that your fantasy narrative is a bit fragile, and contact with reality destroys that fragility, I think.
Brian Chau
Oh, absolutely.
Jim O’Shaughnessy
Brian, this has been a ton of fun. I wish you the greatest of success with Effort News, and we will reschedule you to continue the conversation.
Brian Chau
Yeah, this was fantastic. You’re really an amazing interviewer, Jim. And see you next time.
Jim O’Shaughnessy
See you next time. Brian, thanks for coming on.
