Latest / Razib Khan's Unsupervised Learning / Brian Chau: noticing with technology + investigative journalism
Transcript
- 0:00SPEAKER_03: This podcast is brought to you by the Albany Public Library Main Branch and
- 0:03SPEAKER_03: the generosity of listeners like you.
- 0:09SPEAKER_03: What is a podcast?
- 0:16SPEAKER_03: God, Daddy, these people talk as much as you do.
- 0:26SPEAKER_03: Razib Khan's Unsupervised Learning.
- 0:29SPEAKER_01: Hello, everybody.
- 0:30SPEAKER_01: It's Razib Khan with the Unsupervised Learning podcast.
- 0:33SPEAKER_01: This week,
- 0:34SPEAKER_01: I am recording a podcast with Brian Chow,
- 0:37SPEAKER_01: a returning guest, entrepreneur, technologist, cultural critic in various ways.
- 0:44SPEAKER_01: And he has a new project related to the intersection of technology and media,
- 0:48SPEAKER_01: using some of the technological tools that we have today to try
- 0:52SPEAKER_01: to break problems Basically do investigative journalism,
- 0:55SPEAKER_01: but do it in a quantitative and rigorous way, I think it's pretty interesting.
- 1:00SPEAKER_01: I actually have wondered who was going to do this,
- 1:04SPEAKER_01: because what you're doing, Brian, it's interesting.
- 1:09SPEAKER_01: It's novel, but it also seemed like it was going to be inevitable.
- 1:11SPEAKER_01: And so I was wondering, and I guess that was you who was going to do it.
- 1:13SPEAKER_01: So talk to me about your new project.
- 1:15SPEAKER_01: Actually,
- 1:16SPEAKER_01: you can talk about your background,
- 1:17SPEAKER_01: whatever that matters to people,
- 1:18SPEAKER_01: but then talk to you about your new project, because that's really why you're here.
- 1:21SPEAKER_02: Yeah, I mean,
- 1:22SPEAKER_02: most people,
- 1:23SPEAKER_02: One of the things that they look at and stand out is I did IOI Gold in 2017,
- 1:30SPEAKER_02: this big programming contest.
- 1:32SPEAKER_02: I started Alliance for the Future,
- 1:34SPEAKER_02: which was this NGO in DC that was actually advocating for pro-AI policies,
- 1:40SPEAKER_02: not one of the anti-AI ones.
- 1:46SPEAKER_02: And I just left my old software position at network school.
- 1:54SPEAKER_02: Yeah, the project is a lot more interesting.
- 1:56SPEAKER_02: So the project,
- 1:58SPEAKER_02: I saw a lot of people trying to do investigative journalism by having AI automate
- 2:03SPEAKER_02: the writing.
- 2:05SPEAKER_02: And it's shocking to me that AI writing still sucks.
- 2:08SPEAKER_02: The best people in the world have been working on it.
- 2:10SPEAKER_02: They've cracked the hard math problems.
- 2:12SPEAKER_02: They've cracked coding.
- 2:14SPEAKER_02: But the writing is still terrible.
- 2:17SPEAKER_02: But instead of doing that, we use AI to query these massive troves of documents.
- 2:22SPEAKER_02: So in some cases,
- 2:23SPEAKER_02: literally millions of documents,
- 2:25SPEAKER_02: financial records, legal records, filings, stuff like that.
- 2:29SPEAKER_02: And we use that to break stories.
- 2:31SPEAKER_02: And we have broken stories, and there'll be a lot more on the way.
- 2:35SPEAKER_01: Yeah.
- 2:36SPEAKER_01: So talk about the...
- 2:38SPEAKER_01: I mean,
- 2:38SPEAKER_01: I think the high level people understand like how this is going to work in terms of,
- 2:41SPEAKER_01: okay, so,
- 2:43SPEAKER_01: you know,
- 2:43SPEAKER_01: technology,
- 2:44SPEAKER_01: you know, software, you know, algorithms, programs, whatever you want to call it.
- 2:48SPEAKER_01: They're good at munging through big piles of data.
- 2:52SPEAKER_01: Talk about the logistics of,
- 2:54SPEAKER_01: What you had to do,
- 2:56SPEAKER_01: because I'm assuming someone can't just go onto research grade chat GPT
- 3:02SPEAKER_01: and just do everything you did, do this, do some command prompt or whatever.
- 3:07SPEAKER_01: What are the nuts and bolts here?
- 3:10SPEAKER_02: Yeah, the main thing that we found is that LLMs are very inefficient.
- 3:15SPEAKER_02: When they're making queries to government databases, things like that,
- 3:21SPEAKER_02: They'll just run out of usage incredibly quickly.
- 3:25SPEAKER_02: And maybe if you're willing to pay tens of thousands of dollars,
- 3:30SPEAKER_02: and that's probably feasible, actually.
- 3:32SPEAKER_02: But a lot of what we ended up doing was optimizations around creating offline copies
- 3:37SPEAKER_02: of the database,
- 3:38SPEAKER_02: streamlining those queries, optimizing those tool calls in some other ways.
- 3:42SPEAKER_02: A lot of that is the infra stuff.
- 3:45SPEAKER_00: And then there's also the sort of judgment component where it does help a lot
- 3:51SPEAKER_00: to have an idea of where to point the LLMs.
- 3:55SPEAKER_02: For example,
- 3:55SPEAKER_02: for the ORPC story,
- 3:58SPEAKER_02: the Office of Refugee Resettlement Preferred Communities grant,
- 4:01SPEAKER_02: this was like the widest story, you know, had thousands of likes on Twitter.
- 4:07SPEAKER_02: This one, we kind of got it, it kind of came about by creating the HHS grant table.
- 4:13SPEAKER_02: So something like that,
- 4:14SPEAKER_02: where we have a defined data source and we kind of know at least
- 4:19SPEAKER_02: where we're looking, if not what we're looking for.
- 4:21SPEAKER_02: This one, we didn't know what we were looking for.
- 4:22SPEAKER_02: It was literally just,
- 4:23SPEAKER_02: you know,
- 4:24SPEAKER_02: identify grants that spiked during this period,
- 4:28SPEAKER_02: during the most recent six years under the HHS grant table.
- 4:33SPEAKER_02: When we have a unified data source, that becomes much easier.
- 4:36SPEAKER_02: So it's both the human judgment around how to use it and some of
- 4:39SPEAKER_02: the more technical optimizations around how the LLMs are fetching data.
- 4:45SPEAKER_01: Okay.
- 4:45SPEAKER_01: And so, actually, I didn't do the research on your particular project right now.
- 4:52SPEAKER_01: Is it, are you trying to...
- 4:54SPEAKER_00: It's actually, yeah,
- 4:56SPEAKER_00: it's really interesting that,
- 4:57SPEAKER_00: like, A lot of people have thought that this era was coming
- 5:03SPEAKER_02: for a long time.
- 5:05SPEAKER_02: You talked about it as inevitable,
- 5:07SPEAKER_02: but people have said that for a while,
- 5:09SPEAKER_02: where even under Nixon, there was this whole era of sunlight is the best.
- 5:15SPEAKER_02: It's time for transparency and so on.
- 5:17SPEAKER_02: And people thought, oh,
- 5:18SPEAKER_02: if we make these laws mandating that governments will have
- 5:20SPEAKER_02: to make certain disclosures,
- 5:22SPEAKER_02: then that'll usher in a whole wave of truth and reconciliation.
- 5:26SPEAKER_02: People will now know the truth.
- 5:28SPEAKER_02: People will now know what's going on with their government.
- 5:31SPEAKER_02: And that never really materialized.
- 5:32SPEAKER_02: People thought it was going to materialize.
- 5:34SPEAKER_02: People also thought it was going to materialize in the 2010s.
- 5:37SPEAKER_02: I'm sure there's been more historical cycles where that's happened.
- 5:41SPEAKER_02: But it didn't really materialize until now.
- 5:45SPEAKER_02: And I think that's the thing that we're fixing, is that we now have the raw data sources.
- 5:50SPEAKER_02: We've had the raw data sources for a long time,
- 5:53SPEAKER_02: but for the first time,
- 5:54SPEAKER_02: it's becoming feasible to actually do this kind of investigation.
- 5:58SPEAKER_01: Okay, so I guess one thing I was going to ask is, you know, is this going to be a startup?
- 6:04SPEAKER_01: Is this for profit?
- 6:05SPEAKER_01: Like, do you have a moat?
- 6:07SPEAKER_02: Yeah, it's a startup.
- 6:08SPEAKER_02: It's for profit.
- 6:09SPEAKER_02: We are already making money.
- 6:11SPEAKER_02: The monetization is super fascinating around this.
- 6:14SPEAKER_02: I think that we're kind of in an era.
- 6:17SPEAKER_02: It sort of reminds me of like PayPal or where like people don't really know how
- 6:22SPEAKER_02: to monetize this stuff.
- 6:24SPEAKER_02: So right now we have subscribers, but there's a lot more we could do.
- 6:27SPEAKER_02: One interesting thing was Semi-Analysis recent raise.
- 6:31SPEAKER_02: They raised at, I think, a $200 million valuation.
- 6:37SPEAKER_01: Really quick, tell them what Semi-Analysis is.
- 6:40SPEAKER_01: We know what it is, but not everyone.
- 6:41SPEAKER_02: It's a trade publication for the semiconductor industry.
- 6:45SPEAKER_02: And a big way they make money is by doing these business intelligence reports
- 6:50SPEAKER_02: for specific companies.
- 6:51SPEAKER_02: And you can see where I'm going with this.
- 6:54SPEAKER_02: With the technology that's available to us,
- 6:56SPEAKER_02: it's actually not too different to query private sources of data as opposed
- 7:00SPEAKER_02: to public sources of data.
- 7:01SPEAKER_02: There's nothing in our pipeline that says
- 7:05SPEAKER_00: we can only Fetch from public sources.
- 7:09SPEAKER_00: So I think that as we establish,
- 7:13SPEAKER_00: and I think that this also matters a lot to our positioning,
- 7:16SPEAKER_02: as we establish ourselves as the highest grade standard of published information,
- 7:22SPEAKER_02: And I've thought a lot about that as regards to sourcing,
- 7:26SPEAKER_02: as it comes to the aesthetics of the site,
- 7:29SPEAKER_02: as it comes to what stories we'll cover and what sources we'll use.
- 7:33SPEAKER_02: As we establish ourselves as the highest,
- 7:37SPEAKER_02: essentially like highest quality,
- 7:38SPEAKER_02: highest refinement of information,
- 7:41SPEAKER_02: that becomes something you want for your own organization.
- 7:44SPEAKER_02: And so that becomes a path towards revenue that I think is really viable.
- 7:49SPEAKER_02: And there's a lot of other adjacent techniques around that.
- 7:54SPEAKER_02: So one other thing I was thinking about is like a newswire.
- 7:57SPEAKER_02: This is something that would just be like super awesome from like
- 8:00SPEAKER_02: a visual standpoint where we could literally publish a story and we can see
- 8:03SPEAKER_02: the chart of like, oh, here's the tick that went up after we
- 8:06SPEAKER_00: published the story.
- 8:07SPEAKER_00: This is kind of like what Bloomberg already does.
- 8:10SPEAKER_00: And something that could be even more interesting is that
- 8:15SPEAKER_02: many startups
- 8:15SPEAKER_02: from like Series A or Series B onwards spend a lot of time doing audit preparation,
- 8:19SPEAKER_02: right,
- 8:20SPEAKER_02: or diligence preparation,
- 8:21SPEAKER_02: essentially gathering all the internal financials,
- 8:24SPEAKER_02: all the data that otherwise might have been messy from multiple data sources
- 8:27SPEAKER_02: and kind of joining them in order to create baseline financials that they present
- 8:30SPEAKER_02: to investors or,
- 8:31SPEAKER_02: you know,
- 8:32SPEAKER_02: in the case of public companies,
- 8:33SPEAKER_02: they have to
- 8:34SPEAKER_00: file and,
- 8:34SPEAKER_00: you know, this is actually under, you know, under the threat of law.
- 8:43SPEAKER_00: I've been in these experiences before,
- 8:44SPEAKER_00: and I would estimate that it costs millions,
- 8:48SPEAKER_00: if not tens of millions of dollars, even for a medium sized startup.
- 8:52SPEAKER_00: And for a public company, it could easily cost up to billions of dollars.
- 8:57SPEAKER_00: So I think the market for this kind of technology,
- 9:01SPEAKER_00: it's Really unclear at the moment what the best way to monetize it is.
- 9:05SPEAKER_02: I think at the end stage, we'll have a kind of,
- 9:09SPEAKER_02: you know,
- 9:10SPEAKER_02: we'll kind of have a throw everything at the wall strategy in terms of monetization.
- 9:15SPEAKER_02: But there's a ton there.
- 9:16SPEAKER_02: And yeah, the whole thing is for profit.
- 9:18SPEAKER_02: Right now, we don't have any investors, but I would expect that to change pretty soon.
- 9:22SPEAKER_01: So you're bootstrapping it.
- 9:24SPEAKER_00: Yeah, we're bootstrapping it.
- 9:25SPEAKER_00: The burn is really low, especially with...
- 9:28SPEAKER_00: It's interesting because I've kind of drifted around.
- 9:31SPEAKER_00: I've not had a lot of jobs where I
- 9:33SPEAKER_02: can combine the media stuff with technical
- 9:39SPEAKER_02: or just kind of like programming background.
- 9:42SPEAKER_02: But it's interesting.
- 9:43SPEAKER_02: I'm able to kind of switch between the technical optimization and kind
- 9:47SPEAKER_02: of harness building And the kind of
- 9:51SPEAKER_00: more human judgment side of these things
- 9:54SPEAKER_00: in this company really often, which is something that's super fun.
- 9:58SPEAKER_01: So let's talk about what are the biggest things you have discovered right now
- 10:06SPEAKER_01: in terms of what have you uncovered just in the short,
- 10:09SPEAKER_01: you've been around for what, like a week?
- 10:10SPEAKER_02: Yeah, so we've been out for a week.
- 10:13SPEAKER_02: The launch was last Monday.
- 10:16SPEAKER_02: And the company has been around for around a month.
- 10:20SPEAKER_02: The biggest story, I actually discovered this story while I was doing just random experiments.
- 10:26SPEAKER_02: I was actually building, to speak to the audit point, I was building the audit software.
- 10:30SPEAKER_01: Real quick, I should interrupt.
- 10:31SPEAKER_01: Probably we should just say it's effort.news, right?
- 10:35SPEAKER_01: Effort.news.
- 10:35SPEAKER_02: Yeah, effort.news is where you find all the stories.
- 10:38SPEAKER_02: Effort.news slash PC is the biggest one.
- 10:41SPEAKER_02: These asylum grants,
- 10:43SPEAKER_02: these migration and refugee services grants that have gone to some of
- 10:47SPEAKER_02: the biggest critics of the Trump policy.
- 10:50SPEAKER_02: I don't know if you followed this story,
- 10:51SPEAKER_02: but every time there's a Trump migration policy story,
- 10:55SPEAKER_02: they will bring out these religious NGOs that claim
- 10:59SPEAKER_02: to represent Christian Americans.
- 11:01SPEAKER_02: And as we found,
- 11:02SPEAKER_02: 81.8% of their revenue in one case, USCCB, comes from these migration contractors.
- 11:08SPEAKER_02: It's like, you know...
- 11:09SPEAKER_02: You're bringing out the bakery to comment on how,
- 11:11SPEAKER_02: you know,
- 11:12SPEAKER_02: every American should buy bread,
- 11:14SPEAKER_02: how it's a federal right to subsidize bread,
- 11:17SPEAKER_02: how it's a moral good to, you know, bake bread.
- 11:21SPEAKER_02: That's what is happening with bringing out these organizations that are,
- 11:24SPEAKER_02: you know,
- 11:26SPEAKER_02: The vast majority of their funding comes from migration and refugee services grants,
- 11:29SPEAKER_02: yet they're presented by the media as like neutral arbiters or
- 11:33SPEAKER_02: as like somehow morally representative of Christians in America.
- 11:37SPEAKER_02: It's this kind of crazy conflict of interest that was completely undisclosed for decades.
- 11:44SPEAKER_00: And we actually found this story.
- 11:48SPEAKER_02: This was the story that I found by querying the HHS grant table.
- 11:51SPEAKER_02: It was actually like a lot fairer than how a lot of investigative journalism comes
- 11:55SPEAKER_02: about because a lot of journalism,
- 11:56SPEAKER_02: you know,
- 11:57SPEAKER_02: even going back to Nixon,
- 11:58SPEAKER_02: it's like a source comes to you with a story that they want you to publish.
- 12:01SPEAKER_02: And sometimes that's newsworthy.
- 12:02SPEAKER_02: I'm not saying it's all it's never right to publish in that case.
- 12:06SPEAKER_02: But that's following someone's agenda.
- 12:08SPEAKER_02: In this case,
- 12:09SPEAKER_02: we literally queried,
- 12:10SPEAKER_02: you know, find newsworthy findings in the HHS grant table from 2020 to 2026.
- 12:18SPEAKER_02: Look in particular for massive spikes in grant funding.
- 12:22SPEAKER_02: And this was a program that went from around 90 million,
- 12:26SPEAKER_02: not a lot,
- 12:26SPEAKER_02: maybe a lot to,
- 12:27SPEAKER_02: you know,
- 12:27SPEAKER_02: you and I, but nothing compared to the 1.2 billion that it rose to under Biden.
- 12:33SPEAKER_02: And so it was this like massive...
- 12:36SPEAKER_01: That's a 10x.
- 12:37SPEAKER_01: Wait, that was a 10x?
- 12:37SPEAKER_02: Yeah, it's more than 10x.
- 12:39SPEAKER_02: It's more than 10x.
- 12:40SPEAKER_02: 90 to 1.2 billion.
- 12:42SPEAKER_02: 90 million to 1.2 billion.
- 12:44SPEAKER_02: And we found this by just querying the HHS grant table for anomalies.
- 12:49SPEAKER_01: So how many tables... How big is the data that you're querying?
- 12:53SPEAKER_02: Yeah, I think it's some number of hundreds of thousands of rows.
- 12:58SPEAKER_01: Okay.
- 12:58SPEAKER_01: And so, I mean,
- 12:59SPEAKER_01: you know,
- 12:59SPEAKER_01: the old school days back before even computers,
- 13:02SPEAKER_01: like there were investigative journalists,
- 13:03SPEAKER_01: but obviously these sorts of things were not possible before you had automation
- 13:08SPEAKER_01: and computation to actually like look through the patterns, right?
- 13:12SPEAKER_00: Oh, absolutely.
- 13:13SPEAKER_00: When we launched, a lot of people compared us to ProPublica.
- 13:16SPEAKER_02: And that's a super interesting comparison because ProPublica has teams of 10, 20 journalists.
- 13:26SPEAKER_01: Tell the listener what ProPublica is real quick.
- 13:28SPEAKER_02: Yeah, they do a lot of tax stuff.
- 13:31SPEAKER_02: They investigate essentially like tax documents and filings, right?
- 13:35SPEAKER_02: They're this kind of like somewhat left of center NGO.
- 13:41SPEAKER_02: It's a nonprofit.
- 13:42SPEAKER_02: And they do a lot of investigations into like,
- 13:45SPEAKER_02: for example, like what the Koch funding network is donating to, right?
- 13:50SPEAKER_02: And they've done this for a lot of like NGO networks.
- 13:54SPEAKER_02: And they basically go and they read all of the tax filings from all of these organizations.
- 14:00SPEAKER_02: And then they kind of manually input those numbers into some kind of spreadsheet
- 14:06SPEAKER_02: or some kind of graphic.
- 14:08SPEAKER_02: And, you know, it's this very labor intensive work.
- 14:11SPEAKER_02: And they will have to have,
- 14:12SPEAKER_02: you know,
- 14:12SPEAKER_02: large teams of lawyers and journalists and researchers do this for many years
- 14:18SPEAKER_02: on end.
- 14:19SPEAKER_02: Their investigations don't come out in a week.
- 14:22SPEAKER_02: They don't come out in even three weeks.
- 14:24SPEAKER_02: They come out, it takes a matter of years.
- 14:26SPEAKER_02: So I think the scope and the quantity of what's possible is totally different now.
- 14:32SPEAKER_01: And so, I mean,
- 14:34SPEAKER_01: in terms of totally different,
- 14:35SPEAKER_01: like you say,
- 14:35SPEAKER_01: like 10x,
- 14:36SPEAKER_01: 100x, like basically is your rate limiter at this point just the resources you put?
- 14:41SPEAKER_01: Because that's what it sounded like.
- 14:43SPEAKER_01: In terms of like token,
- 14:44SPEAKER_01: I mean, whatever, like, you know, it could be tokens, it could be GPUs, whatever.
- 14:46SPEAKER_01: But I mean, are you limited basically by your spend, by your infrastructure at this point?
- 14:53SPEAKER_02: I think we're limited in a few ways.
- 14:54SPEAKER_02: We're not limited by Infra at all at the moment.
- 14:58SPEAKER_02: We're limited in part by writing, actually.
- 15:02SPEAKER_02: The AI writing is pretty bad.
- 15:05SPEAKER_02: And we're onboarding more writers as we speak onto the team I'm hiring.
- 15:10SPEAKER_02: If you are someone who's interested in joining,
- 15:13SPEAKER_02: jobs at effort.news is where to send us an email.
- 15:16SPEAKER_00: But we're onboarding more hiring and we're onboarding more...
- 15:22SPEAKER_02: We're trying to figure out more ways to expand the search and identify what's newsworthy.
- 15:28SPEAKER_02: I think that the LOMs are pretty good at identifying clear, quantitative changes.
- 15:34SPEAKER_02: Things like the ORPC program,
- 15:37SPEAKER_02: which went from $90 million to $1.2 billion,
- 15:41SPEAKER_02: that's objectively newsworthy, although it evaded
- 15:44SPEAKER_00: coverage for five years.
- 15:48SPEAKER_00: What's more, I think like either getting those tips, right?
- 15:54SPEAKER_00: Getting those areas where it's like, oh, this is something to investigate.
- 15:56SPEAKER_00: This is something that's very important.
- 15:58SPEAKER_00: This is something that will actually see a lot of attention, right?
- 16:00SPEAKER_00: That people will want to hear about.
- 16:03SPEAKER_00: That's one key skill, maybe the most important skill.
- 16:06SPEAKER_00: And another part of that skill is communicating it to LLM.
- 16:10SPEAKER_00: So figuring out,
- 16:12SPEAKER_00: you know,
- 16:12SPEAKER_00: at a harness level,
- 16:14SPEAKER_00: how to expand
- 16:14SPEAKER_02: the kind of qualitative search that we're
- 16:16SPEAKER_00: capable of doing.
- 16:18SPEAKER_02: And I think we're improving at that every day,
- 16:22SPEAKER_02: but there's still a lot more to get better on that.
- 16:26SPEAKER_01: Yeah, real quick, explain what a harness is again.
- 16:28SPEAKER_01: This is a term that's coming up a lot in the AI space,
- 16:32SPEAKER_01: but a lot of the normal people, normies don't know this.
- 16:33SPEAKER_01: You talk about what a harness is.
- 16:35SPEAKER_00: Yeah, for sure.
- 16:38SPEAKER_00: I really appreciate these questions.
- 16:41SPEAKER_00: So a harness is essentially a system that organizes LLM output, right?
- 16:49SPEAKER_02: So take, you know, Claude Code or Codex or OpenClaw.
- 16:54SPEAKER_02: You guys might have heard of those.
- 16:55SPEAKER_02: It's like...
- 16:58SPEAKER_02: Very often, although not always,
- 17:00SPEAKER_02: it's a system that runs in your terminal, right,
- 17:02SPEAKER_02: that uses some kind of file system,
- 17:06SPEAKER_02: and it can take things from the LLMs and,
- 17:10SPEAKER_02: you know,
- 17:10SPEAKER_02: write them to a file,
- 17:11SPEAKER_02: or it can,
- 17:12SPEAKER_02: you know,
- 17:12SPEAKER_02: read from your files and send that output to LLMs,
- 17:16SPEAKER_02: or it can put together a lot of other tools like, you know, using a Google search.
- 17:20SPEAKER_02: Or, you know, running a Python script, running a JavaScript script.
- 17:24SPEAKER_02: It's essentially a way to kind of glue together multiple pieces
- 17:27SPEAKER_02: of traditional software in a way that the AI models can use.
- 17:32SPEAKER_02: And they've been something that's been responsible for creating a massive amount
- 17:35SPEAKER_02: of revenue for OpenAI, for Anthropic, and for a lot of other companies.
- 17:41SPEAKER_01: Yeah, it's interesting.
- 17:43SPEAKER_01: You know,
- 17:43SPEAKER_01: some people thought that,
- 17:46SPEAKER_01: you know,
- 17:47SPEAKER_01: when we had LLMs,
- 17:49SPEAKER_01: basically, you know, you do command prompts and, you know, let's just go.
- 17:54SPEAKER_01: But it turns out that there's a lot of layers of analysis,
- 17:59SPEAKER_01: a lot of layers of,
- 18:01SPEAKER_01: you know,
- 18:02SPEAKER_01: Technological work between the LLM and the output that's human-digestible
- 18:07SPEAKER_01: or human-com principle at this point.
- 18:10SPEAKER_01: Let's talk about what's on deck, what are you thinking about?
- 18:16SPEAKER_02: We are working on a big story right now that's going to be 10 times larger than
- 18:23SPEAKER_02: the $1.2 billion refugee grant program.
- 18:27SPEAKER_00: And that's going to be a $20 billion,
- 18:32SPEAKER_00: essentially like system, an entire ecosystem funded by Medicaid.
- 18:38SPEAKER_02: To essentially do wealth transfers to reduce racial disparity.
- 18:44SPEAKER_02: So it's this entire program using Medicaid,
- 18:48SPEAKER_00: this $21 billion program that is essentially Essentially,
- 18:53SPEAKER_00: transferring money or incentivizing the
- 18:56SPEAKER_02: transfer of money from essentially,
- 18:59SPEAKER_02: you know, whites and Asians towards blacks and Hispanics.
- 19:02SPEAKER_02: That is practically the effect of this outcome and we're graphing all of that as well.
- 19:07SPEAKER_02: And this is something that we'll be breaking, I think, by the time this episode is out.
- 19:13SPEAKER_01: Okay.
- 19:14SPEAKER_01: And so, I mean,
- 19:15SPEAKER_01: in terms of the growth prospects for this sort of technology,
- 19:21SPEAKER_01: do you think,
- 19:22SPEAKER_01: like, obviously, you're out there, you're doing it, you're bootstrapping.
- 19:26SPEAKER_01: Do you think you're going to have imitators?
- 19:28SPEAKER_01: Like, how are you feeling in terms of just, like, the business landscape?
- 19:30SPEAKER_01: Like, you know,
- 19:32SPEAKER_01: it's going to take a while for people to catch up in terms of all
- 19:34SPEAKER_01: the processes you have in place.
- 19:36SPEAKER_00: Yeah.
- 19:37SPEAKER_00: I remember there's one pro gamer who's like,
- 19:40SPEAKER_00: you know,
- 19:41SPEAKER_02: I can talk about this technology because by the time they catch up
- 19:44SPEAKER_02: with this technology, I'll have tons of other new stuff.
- 19:47SPEAKER_02: And I feel the same way.
- 19:49SPEAKER_00: Our team, it's a small team, but we are just discovering new things every day.
- 19:53SPEAKER_02: It's like so fun.
- 19:56SPEAKER_02: It's just like a fascinating experience.
- 19:58SPEAKER_02: And yeah, I think it's one of those things where it'll be easier to snowball ahead of time.
- 20:03SPEAKER_02: And the incumbents are just...
- 20:04SPEAKER_02: They just don't care.
- 20:05SPEAKER_02: A lot of them hate AI.
- 20:06SPEAKER_02: It's like this shocking thing.
- 20:08SPEAKER_02: It's this crown in the gutter that's been lying around for so long.
- 20:10SPEAKER_02: Just been like, no one's been doing this.
- 20:13SPEAKER_02: It's been feasible probably for upwards of at least two years at this point.
- 20:21SPEAKER_02: And it hasn't quite come together for anyone else.
- 20:23SPEAKER_00: And I think that we can just take the entire cake.
- 20:26SPEAKER_00: I think it'll be massive for us.
- 20:30SPEAKER_01: Okay.
- 20:30SPEAKER_01: So, I mean, we're talking like a zero to one scenario.
- 20:32SPEAKER_01: You're pretty optimistic about it.
- 20:33SPEAKER_01: Yeah.
- 20:35SPEAKER_00: It's like just as an experience, it's amazing.
- 20:38SPEAKER_00: And maybe that's clouding my judgment.
- 20:40SPEAKER_00: That's like totally fair.
- 20:42SPEAKER_02: But it is a different experience than like any other startup that I've worked on.
- 20:48SPEAKER_01: Mm-hmm.
- 20:49SPEAKER_01: Well, I mean, so, okay, that sounds really good.
- 20:52SPEAKER_01: You know,
- 20:53SPEAKER_01: one thing that,
- 20:54SPEAKER_01: you know,
- 20:54SPEAKER_01: it is a for-profit,
- 20:55SPEAKER_01: it is a business, and businesses do great things, obviously, I'm not.
- 20:58SPEAKER_01: But what is, how is this going to transform the public sector?
- 21:03SPEAKER_01: How is it going to transform the public in terms of what they know in terms
- 21:05SPEAKER_01: of investigative journalism?
- 21:06SPEAKER_01: Like, what are you already seeing?
- 21:10SPEAKER_00: So, what we're seeing, there's already been a few changes.
- 21:15SPEAKER_00: The biggest change is that
- 21:19SPEAKER_00: They announced an update to this after,
- 21:21SPEAKER_00: I don't know how much of this was already planned,
- 21:23SPEAKER_00: but after our request for comment,
- 21:26SPEAKER_00: they announced this update to this refugee grant program that essentially excluded
- 21:32SPEAKER_00: the previous round of recipients from further participation in the program.
- 21:38SPEAKER_00: So these religious NGOs like Church World Service,
- 21:41SPEAKER_00: like USCCB,
- 21:43SPEAKER_00: like Episcopal Migration Ministries,
- 21:46SPEAKER_00: they're no longer able to receive the ORRPC grants.
- 21:51SPEAKER_00: And, you know, maybe that's something that they already had in motion.
- 21:56SPEAKER_00: I don't know for sure.
- 21:58SPEAKER_00: But that was something that really kind of
- 22:00SPEAKER_02: like made my day.
- 22:02SPEAKER_02: Yeah.
- 22:04SPEAKER_00: I think that at a broader level,
- 22:06SPEAKER_00: what's going to happen is that the fruits of all
- 22:09SPEAKER_00: of these transparency laws will actually come to pass.
- 22:12SPEAKER_00: And I think that you may begin to see more and more people turn against the concept
- 22:18SPEAKER_00: of transparency.
- 22:19SPEAKER_00: And that kind of sounds like a crazy thing to say,
- 22:25SPEAKER_00: kind of sounds crazy,
- 22:26SPEAKER_00: you know,
- 22:26SPEAKER_00: it's sort of this platitude at the moment,
- 22:29SPEAKER_00: but I think you'll see very concrete ways,
- 22:32SPEAKER_00: first in which,
- 22:33SPEAKER_00: you Politicians will go about trying to repeal transparency laws
- 22:38SPEAKER_00: without hoping anyone will notice,
- 22:41SPEAKER_00: and maybe using covert language to tie that up,
- 22:45SPEAKER_00: and then later on more explicitly come out against transparency as a concept.
- 22:51SPEAKER_00: And I think that's because...
- 22:54SPEAKER_00: There was this quote from Colbert a while ago.
- 23:01SPEAKER_02: His claim was that reality has a left-wing bias.
- 23:04SPEAKER_02: And I think the accurate version of that is that sources reporting has a left-wing bias.
- 23:09SPEAKER_02: The New York Times actually just does a better job than the New
- 23:12SPEAKER_00: York Post or
- 23:13SPEAKER_00: the Wall Street Journal or whoever has sources reporting.
- 23:16SPEAKER_00: And there's all sorts of systematic reasons for that.
- 23:19SPEAKER_00: But the equivalent to that is that data reporting,
- 23:26SPEAKER_00: transparency reporting has a natural right-wing or libertarian bias.
- 23:32SPEAKER_00: The more you look at the government, the more the normal person thinks, hold on.
- 23:39SPEAKER_00: It's doing what now?
- 23:42SPEAKER_00: What is going on here?
- 23:43SPEAKER_00: It's just this crazy thing that people don't know about.
- 23:47SPEAKER_00: And the more they know about it,
- 23:48SPEAKER_00: the more it becomes a problem for those governments and those programs specifically.
- 23:56SPEAKER_01: Yeah, you know,
- 23:57SPEAKER_01: in terms of transparency,
- 23:58SPEAKER_01: I was thinking,
- 23:59SPEAKER_01: actually,
- 24:00SPEAKER_01: you know,
- 24:01SPEAKER_01: Cambridge,
- 24:01SPEAKER_01: the Education Department took down its listings of the faculty pages
- 24:07SPEAKER_01: because they don't want people investigating.
- 24:09SPEAKER_01: Obviously, there's the Wayback Machine.
- 24:11SPEAKER_01: Anyone with trivial technological know-how could get around this,
- 24:15SPEAKER_01: but I guess they didn't think about that, which makes them look like fools.
- 24:17SPEAKER_01: But, you know,
- 24:18SPEAKER_01: there are a lot of institutions that have a lot of things that they want to hide
- 24:21SPEAKER_01: that they don't want the public to know.
- 24:23SPEAKER_01: And so this will probably trigger some sort of Oh, absolutely.
- 24:27SPEAKER_01: And we can already see that.
- 24:28SPEAKER_01: And there's this like Pareto frontier where there are ways
- 24:31SPEAKER_01: where it's technically legal to hide
- 24:32SPEAKER_00: it, but there are other ways
- 24:41SPEAKER_02: Kind of like countermeasures, like other disclosures that allow us to break through.
- 24:46SPEAKER_02: And here's a very good example.
- 24:47SPEAKER_02: We did the story on the 19th century family fortunes that are funding degrowth in Europe.
- 24:54SPEAKER_02: So there are all of these eras,
- 24:55SPEAKER_02: like literal, like,
- 24:57SPEAKER_02: Great-grandchildren or great-great-grandchildren of people
- 24:59SPEAKER_02: who invented like pharma fortunes in the 1800s who are now funding degrowth.
- 25:06SPEAKER_02: And they actually hid their money in these Swiss bank accounts.
- 25:10SPEAKER_02: This organization called Swiss Philanthropy Foundation, which is essentially a money mixer.
- 25:15SPEAKER_02: They host 106 different nonprofits.
- 25:18SPEAKER_02: You donate to any of the nonprofits,
- 25:19SPEAKER_02: it just registers on the
- 25:20SPEAKER_00: disclosures as a donation to Swiss Philanthropy Fund.
- 25:24SPEAKER_00: But we were able to use those records to trace back all of the donors,
- 25:29SPEAKER_00: and the donors have the donor-level disclosures.
- 25:32SPEAKER_02: So this degrowth organization, P4NE, they didn't disclose the donor amounts at all.
- 25:39SPEAKER_02: So they didn't disclose how much they got from each donor,
- 25:43SPEAKER_02: and they kind of hid it through the Swiss Philanthropy Foundation.
- 25:46SPEAKER_02: But in a lot of cases, the donors themselves had to disclose that information.
- 25:51SPEAKER_02: So we're able to cross-reference that and retrieve a lot of the transactions.
- 25:55SPEAKER_02: Actually,
- 25:55SPEAKER_02: not entirely all of them,
- 25:57SPEAKER_02: interestingly,
- 25:57SPEAKER_02: because some of the donors had different tax structures and things like that,
- 26:01SPEAKER_02: and they didn't have to do the same filings.
- 26:03SPEAKER_00: But we're able to retrieve the donor end filings and use those to prove out all of
- 26:08SPEAKER_00: the transactions to this degrowth NGO.
- 26:12SPEAKER_01: All right.
- 26:12SPEAKER_01: I mean, so, yeah, I mean, so obviously there's a lot of things that you could do.
- 26:19SPEAKER_01: I'm kind of curious, like,
- 26:21SPEAKER_01: how much are you going to focus on America and how much are you going to focus
- 26:23SPEAKER_01: on other countries?
- 26:24SPEAKER_00: So obviously I'm American.
- 26:26SPEAKER_00: I, you know, care a lot about America.
- 26:30SPEAKER_00: We've had some stories about other countries so far, other countries in the West.
- 26:38SPEAKER_00: That's a good question.
- 26:38SPEAKER_00: I mean, I don't have, I obviously have a bias.
- 26:43SPEAKER_02: I mean, like, I'm American.
- 26:45SPEAKER_02: I follow American news.
- 26:47SPEAKER_02: Most of my friends are American.
- 26:49SPEAKER_02: Like, I live here, so it personally affects me.
- 26:52SPEAKER_02: But really, it is whatever news is fit to print.
- 26:55SPEAKER_02: If we see a big story like we saw with,
- 26:57SPEAKER_02: you know,
- 26:58SPEAKER_02: the European Family Fortune story
- 27:00SPEAKER_00: or with the UK censorship story,
- 27:03SPEAKER_00: we'll absolutely cover that.
- 27:04SPEAKER_00: So it is just depends on, you know, what we find and whether that's newsworthy.
- 27:12SPEAKER_01: Yeah, okay.
- 27:14SPEAKER_01: So what else would you like the public?
- 27:17SPEAKER_01: I mean, this is brand new.
- 27:18SPEAKER_01: It's been out for a week.
- 27:21SPEAKER_01: Obviously, you're making the rounds and putting yourself out there.
- 27:24SPEAKER_01: I mean, what do you want people to know about what you're trying to do here?
- 27:28SPEAKER_01: Obviously, let's just be clear again.
- 27:30SPEAKER_01: You are looking to raise.
- 27:32SPEAKER_01: What do you want the consumers out there?
- 27:34SPEAKER_01: What do you want people out there to know about what your company wants to do?
- 27:52SPEAKER_00: Yeah, I think that there's a lot of slop out there.
- 27:57SPEAKER_00: And we are trying to do the opposite of slop.
- 27:59SPEAKER_00: And we are trying to do the opposite of slop in a lot of ways.
- 28:02SPEAKER_00: Number one is that all of the
- 28:03SPEAKER_02: writing is human.
- 28:04SPEAKER_02: So, you know, we do not use AI.
- 28:08SPEAKER_02: We purely handwrite our articles and send them off in post notes.
- 28:16SPEAKER_02: We have people writing all of the articles.
- 28:18SPEAKER_02: Number two is of course the sourcing.
- 28:20SPEAKER_02: We don't say,
- 28:21SPEAKER_02: experts say,
- 28:22SPEAKER_02: we don't make shit up, we don't go making random conspiracy theories, no.
- 28:29SPEAKER_00: Everything we say is proven by literally traceable financial documents that all
- 28:34SPEAKER_00: of you guys read.
- 28:36SPEAKER_00: Or all of you guys can read.
- 28:42SPEAKER_00: And another aspect of that is aesthetics.
- 28:44SPEAKER_00: And I think a lot of people in the new media space,
- 28:49SPEAKER_00: even when they are actually breaking stories,
- 28:51SPEAKER_00: they just don't really put any effort into the aesthetics or into the design.
- 28:57SPEAKER_00: It's not purely a vanity thing.
- 29:04SPEAKER_00: Having good visual design actually communicates information better to the audience.
- 29:09SPEAKER_00: And so you'll have things like the New York Post.
- 29:12SPEAKER_00: And I bring up the New York Post not to especially shit on the New York Post,
- 29:14SPEAKER_00: but because the New York Post actually breaks stories, right?
- 29:17SPEAKER_00: New York Post actually has good sources reporting in some cases,
- 29:21SPEAKER_00: but they also look like the fucking New York Post.
- 29:26SPEAKER_00: And I think a lot of people in this space don't understand that when you look like
- 29:31SPEAKER_00: the New York Post, it just looks ugly.
- 29:34SPEAKER_00: People will just not want to read this,
- 29:36SPEAKER_00: even if there is a genuinely important and newsworthy scoop.
- 29:41SPEAKER_01: Yeah.
- 29:42SPEAKER_00: So there's a lot of friends in the new media space that I think people have begun
- 29:48SPEAKER_00: to criticize, have begun to kind of grow bitter about.
- 29:53SPEAKER_00: They see, you know, some of the conspiracy stuff, some of the schizopository stuff.
- 29:57SPEAKER_00: I'm as guilty as anyone of complaining about it,
- 30:04SPEAKER_00: but now I am actually trying to fix these things.
- 30:07SPEAKER_00: I'm trying to not necessarily act with contempt.
- 30:10SPEAKER_00: I think there's been...
- 30:12SPEAKER_00: A tendency of people to divide themselves into two camps with regards to Joe Rogan
- 30:18SPEAKER_00: or podcasters in general,
- 30:20SPEAKER_00: which is either you hate the podcasters and you have utter dripping contempt for all
- 30:25SPEAKER_00: the podcasters or you believe everything that's said on Joe Rogan.
- 30:29SPEAKER_00: And I think that's not right.
- 30:31SPEAKER_00: I think that you...
- 30:33SPEAKER_00: There's a reason for the suspicion,
- 30:35SPEAKER_00: and as we are proving out in official government documents and sourcing,
- 30:41SPEAKER_00: there are genuine,
- 30:42SPEAKER_00: not exactly conspiracies,
- 30:44SPEAKER_00: but there are certainly misuses of power and conflicts of interest,
- 30:48SPEAKER_00: and in some cases, literal crimes committed by government actors.
- 30:55SPEAKER_00: And...
- 30:56SPEAKER_00: The solution to that is not just to shit on these people and not just to complain and gatekeep.
- 31:03SPEAKER_00: It's kind of like trying to centrally plan or trying to ban your way to innovation.
- 31:09SPEAKER_00: You actually have to do something about it.
- 31:11SPEAKER_00: And we are doing something about it.
- 31:13SPEAKER_00: That's part of the mission.
- 31:15SPEAKER_00: I think part of the mission as well is actually to change what is being discussed.
- 31:22SPEAKER_00: Because a lot of these new media channels,
- 31:25SPEAKER_00: they're still at the end of the day commenting on all of the same stories
- 31:28SPEAKER_00: that are being broken by the New York Times or whatever.
- 31:30SPEAKER_00: They're not breaking news stories.
- 31:32SPEAKER_00: And I'm not saying that just to complain.
- 31:34SPEAKER_00: I'm saying that to actually pilot out the system of breaking news stories
- 31:39SPEAKER_00: and talking about those stories and having them go live with new media partners.
- 31:45SPEAKER_00: So if you yourself are a new media brand,
- 31:47SPEAKER_00: a new media company, and you want to break some stories together, let me know.
- 31:51SPEAKER_00: Support at effort.news.
- 31:52SPEAKER_00: And I think that there's this whole cascading wave
- 32:16SPEAKER_00: In part that we will ride,
- 32:17SPEAKER_00: but in part that will also push forward,
- 32:20SPEAKER_00: that's going to change the new media landscape a lot for the better.
- 32:24SPEAKER_01: All right, man.
- 32:25SPEAKER_01: It was great talking to you.
- 32:26SPEAKER_01: I'm excited.
- 32:26SPEAKER_01: This is early times, early times for your company, you know, early times for this field.
- 32:31SPEAKER_01: And so hopefully we'll hear a lot of great things about it.
- 32:34SPEAKER_01: And we'll hear a lot of great things, you know, from it.
- 32:36SPEAKER_01: I mean, great.
- 32:37SPEAKER_01: Depends on how you define great.
- 32:39SPEAKER_01: But, you know,
- 32:40SPEAKER_01: that we learn how the world really is and make the appropriate decisions based on
- 32:44SPEAKER_01: that, you know.
- 32:45SPEAKER_01: So this is like a classic Silicon Valley story, right, for a better world.
- 32:48SPEAKER_01: But for real this time, right?
- 32:49SPEAKER_01: Yeah, the unveiling is coming.
- 32:51SPEAKER_01: Yeah.
- 32:51SPEAKER_01: Awesome, man.
- 32:52SPEAKER_01: Okay, Brian.
- 32:53SPEAKER_01: So that's effort.news.
- 32:56SPEAKER_01: Check it out online, everybody.
- 32:58SPEAKER_01: And I think by the time this posts, there'll be more new things out there.
- 33:03SPEAKER_01: So it's an accelerating field.
- 33:04SPEAKER_01: And obviously, if you're a funder or someone else like that, reach out.
- 33:08SPEAKER_01: I'll put all the links to Brian's info in the show notes.
- 33:11SPEAKER_01: All right, guys.
- 33:12SPEAKER_01: Thank you.
- 33:13SPEAKER_00: Hell yeah.
- 33:20SPEAKER_03: This podcast for kids...
- 33:49SPEAKER_03: This is my favorite podcast.