The Epstein Files, AI journalism and the future of truth - The Sunday Story

Summary

---

The Story – The Epstein Files, AI Journalism and the Future of Truth

Podcast: The Story
Host: The Times
Episode: The Epstein Files, AI journalism and the future of truth - The Sunday Story
Date: March 29, 2026
Summary By: The Times

---

Episode Overview

This episode investigates the emergence of The Epstein Files, the world’s first AI-generated investigative podcast covering the massive, recently released trove of Jeffrey Epstein documents. Hosts Manveen Rana and Luke Jones explore how artificial intelligence is used to process and present overwhelming data volumes that would take journalists years to fully unpack. The episode also interrogates the implications, limitations, and pitfalls of AI-native journalism—not just for reporting, but for the future of media accuracy, truth, and ethics.

---

Key Discussion Points & Insights

1. The Challenge: Unpacking the Epstein Files Data Dump

• Context:

  • The US Department of Justice released approximately 3.5 million documents related to Jeffrey Epstein, ranging from emails and flight logs to court transcripts.
  • Processing this quantity would take journalists several years (01:27).

• The Opportunity:

  • Adam Levy saw a unique moment: “It was a challenge to take all that data, contextualize it, and build a storyline—a true-crime investigative approach that documents the facts, the stories, the people, and the themes into a comprehensive series.” (05:32)

---

2. Creating the AI-Native Podcast

• Origins & Motivation—Adam Levy:

  • Not a journalist by trade, Levy describes himself as a builder of audiences and content products (04:39).
  • He created a personal app, Distil, to turn various data sources into audio digests—leading to the concept for the Epstein podcast (05:32).

• AI Workflow & Safeguards:

  • The AI ingests diverse file formats, synthesizes information, and constructs story episodes driven by central investigative questions.
  • “It's like bowling with guardrails. I need to add guardrails so that the quality of the content comes out to be the utmost to its potential… Only the highest quality episodes are getting pushed out.” (09:54)

• Content Curation:

  • Rather than random summaries, episodes are built around questions Levy is curious about (e.g., Who was Epstein? What was his operation?)—solving them through “interviewing the files themselves.” (09:02)
  • An internal filtering system ensures only solid episodes get published (11:26).

---

3. The Podcast’s Launch and Reception

• Slow Start, Viral Growth:

  • “The reaction was nobody cared… then three days in, wow, people are actually listening. Seven days, we hit 100,000 downloads.” (11:40)
  • Within 30 days: 2 million downloads; over 100 episodes published (13:51, 13:58).

• Transparency about AI:

  • The show is clearly labeled as AI-generated on all platforms, though some listeners still miss the distinction (13:19, 12:29).
  • Memorable quote: “Maybe the production is that good that some people can't tell until they hear the nuances.” (13:35)

---

4. A Critic’s Perspective: James Ball on AI Podcasts

• Assessment of the Show:

  • Tech journalist James Ball notes, “It was the best AI podcast I’ve heard—far better than the joke ones that had come before—but still sounds a bit off, like US breakfast radio on 1.5x speed.” (16:05, 17:16)
  • The format leverages its vast source material for focus, but lacks human touch and depth of explanation.
  • “Most people, if they listened to even 30 seconds, would probably clock that something’s off… that it isn’t actually people.” (17:16)

• True Crime Genre Issues:

  • Ball observes the podcast taps into the pop-culture, “soapy” side of the Epstein story, sometimes at odds with journalistic ethics and victim sensitivity (20:07).

    “There’s a very real story of human suffering … but there’s a much more, quote, unquote, fun story… and people do treat [the latter] as entertainment, and I think this taps into that.” (20:07)

---

5. The Journalist’s Dilemma: Ethics & Legal Risks

• Vetting and Legal Peril:

  • Ball expresses alarm:

    “No journalist with even basic training would go near publishing this in the way it’s been done. One sentence being wrong can cost hundreds of thousands… There are lots of unverified, unvetted allegations in those documents.” (21:29)

• AI’s Limitations:

  • Ball cautions: “AI does make mistakes… they will often just misattribute a quote… It just takes one episode to say the wrong thing and your life could be ruined.” (23:04)

• Guardrails and Fact-Checking:

  • Levy reiterates he built safeguards and disclosures, programmed “the truth machine” to optimize for accuracy and nonpartisan tone (27:15, 30:37).
  • Admits: “You can’t make it non-biased 100%... but if you program the guardrails correctly, you could get a quality output from the black box.” (30:37)

---

6. Implications for Journalism and the Future of Truth

• Human Element vs. Automation:

  • Manveen raises concerns that human nuance is lost:

    “On the human understanding—war, mass trafficking of female victims—these are very human stories. Are you missing something by having podcasts entirely generated by AI?” (34:57)

  • Levy maintains optimism that software can, over time, capture the human element (35:33).

• Impact on Media Ecosystem:

  • Levy contends: “Most newsrooms already use AI tools in workflows. Maybe the end product is a human, but I’m very doubtful they’re not using AI for production.” (34:18)

• Host’s Closing Reflection:

  • Manveen voices a broader worry:

    “If everybody becomes so reliant on AI to synthesize [information] … you almost forget how you do the basic skill of finding stuff out on your own. In the end, it's kind of the truth that disappears. There will be a version of the truth, but there won't be the same search for it.” (36:45)

---

Notable Quotes & Moments (with Timestamps)

• Adam Levy on motivation and experimentation:

  "This was just another idea in the shed… then three days in, wow, people are actually listening" (11:40)

• James Ball on AI’s current limits:

  "Most people, if they listened to even 30 seconds, would probably clock that something's off." (17:16)

• Ethical warning from James Ball:

  "One sentence being wrong can cost you hundreds of thousands, if not millions." (21:29)

• Manveen Rana’s fundamental question:

  "Is it possible for AI not to have biases? ... we can almost identify what their biases are ... it still has a way of presenting information which... represents a bias." (30:14)

• Adam Levy on truth and optimism:

  “I’m optimistic… while there are bad actors, I'd like to think that some want to see this technology win in the right manner to benefit society.” (35:33)

• Manveen’s final caveat:

  "In the end, it's kind of the truth that disappears… There will be a version of the truth, but there won't be the same search for it." (36:45)

---

Key Segments & Timestamps

• [01:27] — Introduction: The data dump and the challenge it presents
• [04:22] — Adam Levy: Background & creation motivation
• [07:54] — Technical explanation: AI processing, guardrails, and content curation
• [11:40] — Launch impact: Podcast statistics, viral growth
• [14:29] — James Ball: Tech journalist reviews the podcast and its culture place
• [21:29] — Ethical and legal dangers: Journalism vs. AI automation
• [27:15] — AI fact-checking, bias, and truth mechanisms
• [34:57] — Media landscape risks, journalism's future, the AI/human divide
• [36:45] — Host’s closing reflection on the fundamental search for truth

---

Tone and Language

The episode blends journalistic skepticism, clear-eyed curiosity, and philosophical debate. Adam Levy brings a product-building, experimental tech optimism. James Ball is skeptical, informed by newsroom rigor and legal caution. Manveen Rana offers balanced questioning and thoughtful reflection on the risks and rewards.

---

Conclusion

The Epstein Files podcast embodies a crossroads: AI’s promise in rapid data processing and content generation is undeniable—demonstrated by millions of downloads and a product no newsroom could match for speed. Yet, this episode highlights the profound risks: legal pitfalls, loss of human nuance, and the peril that “the search for truth” itself could be subordinated to whatever the algorithm presents. Both excitement and apprehension characterize the rapidly approaching AI future in journalism.

---

Transcript

[Cachava Advertiser]

Craving the coffee flavor you love but without the caffeine, Cachava's got you covered with their newest coffee flavor. This all in one Nutrition Shake delivers bold, authentic flavor crafted from premium decaffeinated Brazilian beans with 25 grams of protein, 6 grams of fiber, greens and so much more. Treat yourself to the flavor and nutrition your body craves. Go to cachava.com and use code smoothie new customers get 15% off their first order. That's K A C-H-A-V-A.com code smoothie lets be completely honest.

[Strawberry Me Career Advertiser]

Are you happy with your job? The fact is, a huge number of people can't say yes to that. Too many of us are stuck in a job we've outgrown or one we never really wanted in the first place. But we stick it out and we give reasons. Like what if the next move is worse? And I've put years into this place and maybe the most common one. Isn't everyone miserable at work? But there's a difference between reasons for staying and excuses for not leaving. It's time to get unstuck. It's time for Strawberry Me. They match you with a certified career coach who helps you get from where you are to where you want to be, either at your existing job or by helping you find a new one. Your coach helps clarify your goals, creates a plan and keeps you accountable along the way. Go to Strawberry Me Career and get 50% off your first coaching session. That's Strawberry Me slash career

[Manveen Rana]

from the Times and the Sunday Times. This is the story on Sunday. I'm Manveen Rana. Thousands of documents from two criminal investigations into the late sex offender Jeffrey Epstein have been released in the US which offers a glimpse into the sheer number of his wealthy and powerful associates. The Epstein Files have unleashed one of the biggest and most complex investigations of modern times. They dropped a colossal 3.5 million documents ranging from emails, flight logs, witness testimony and court transcripts. It's predicted to take several years of full time uninterrupted reading for journalists to go through them all, and even longer to conduct thorough investigations and fact checking to work out what they really mean. It's a frustratingly long and intensive process, but in that problem, one man saw an opportunity. Here at the Story, we like to keep an eye on the podcast charts and we found one new entry particularly intriguing. A new series called the Epstein Files hit the top of the Apple charts. And while at first it might look like any other podcast when we click play, it was definitely different.

[James Ball]

You are listening to the Epstein Files, the world's first AI native investigation into the case that traditional journalism simply could not handle.

[Manveen Rana]

A podcast with no reporters, no producers, in fact, no humans attached at all. Using AI to bring you the very latest in a time frame that most journalists could only dream of. But does it work? How does it sound? And what does AI led journalism mean for the future of investigations and the search for truth? We managed to track down its creator in America for a chat. The story today, AI journalism and the future of truth.

[Adam Levy]

My name is Adam Levy. I'm the creator of the Epstein Files podcast as well as, more recently, War Desk. And I love the idea of treating media as building product.

[Manveen Rana]

Tell us how you get there. What's your background? Are you. Are you a media person? Are you a product person?

[Adam Levy]

Yeah. So my whole career has been building owned audiences. I grew my newsletter to over 100,000 subscribers. I built a community on chain of over 100,000 collectors and sold nearly half a million pieces of content. And I've learned what it takes to build an audience that comes back every day, not just for a moment. And when I was building the Epstein Files, I kind of tapped into my background of building data products, creating content, and melding the two together to create a source of truth in documenting everything that the dog released around Jeffrey Epstein.

[Manveen Rana]

So, Adam, you weren't a journalist. What was it that drew you to this? I mean, what made you think. I have an idea.

[Adam Levy]

You know, during the holidays, December of 2025, I built this personal app for myself called Distil. And the premise of Distil was taking a variety of sources, extracting the meat and potato of those sources, and distilling them into an audio form that I could then learn and listen from. I learned best through audio. So I built me this app to do that accordingly across any topic. So when the Epstein File started trending, I thought it would be cool and honestly, quite interesting and a challenge to figure out a way to take all that data, contextualize it, and build a storyline through it. A true crime like, investigative approach to documenting and picking out the facts, the stories, the people, the themes into this comprehensive series that I call the Epstein Files.

[Manveen Rana]

So when the Epstein Files are released, you sort of. You think this is the perfect place to experiment with that format?

[Adam Levy]

Yeah, I thought it was the right medium. I thought, you know, going back to my background of building product and data products in particular, I thought there was a unique opportunity to get the millions of data points that were, you know, kind of abruptly dumped on society from the doj, and take the responsibility to create something more digestible, because there's no way that any human would be able to go through those files, find meaning through those files, context to those files, to understand exactly what happened around this character. I felt like it was an interesting opportunity. Not from a place of, oh, I think the public will enjoy this. More from a place of, I'm interested in this, but let me build this for myself and then I'll just publish it and I'll see what happens.

[Manveen Rana]

You know, having been on the newsroom side of the story, you know, I can assure you there was such a huge dump of data and documents, no one newsroom could go through it and get through it in, you know, a year. It would take much longer. So here you have an AI solution that will go through it in minutes effectively. How does that turn into a series of podcasts?

[Adam Levy]

It's a great question. You know, once you contextualize the data and you. You bring the data to a point where it's formatted to be digested in a unified manner. So what does that mean? When the DOJ released all these files, there's videos, images, screenshots, PDFs, you have an entire database of just his emails and communication from different, know, people in the timeline. You kind of ingest all that information, and through that you work with the AI to figure out what's the interesting angle to tie all this data together in a compelling story. AI is the perfect tool to synthesize that media or that information to create the media around it, to then tell the story around it.

[Manveen Rana]

But you sort of have to tell the AI how to order that information. So, you know, how do you go about it? What makes an episode? Is it one character and you follow all of their correspondence through the Epstein files? Or how did you decide to sort through all of that, to turn it into something digestible?

[Adam Levy]

You need to come from a place of curating the questions. Again, I always stem from what I'm interested in. So there were key questions that I wanted answered more around the fact of who was Jeffrey Epstein? What was the operation? What was the business model? What role did the island play? What role did Manhattan play? Is it Manhattan Apartment? I kind of took that approach to kind of interviewing the files themselves, all the data themselves, to get questions to the answers that I had and then turn those questions into this comprehensive content outline that would then be generated as a series.

[Manveen Rana]

I mean, just tell us a bit about that on the technical side. And this is certainly not. I'm certainly not A technical audience, but I imagine some of our listeners won't be either. So just explain in very layman's terms how you went about doing that.

[Adam Levy]

So there's an entire product that we built around this engine. And not to get too technical, but the goal of the engine is to take a bunch of inputs, emails, documents, files, construct a comprehensive outline as to what an episode could entail based off a question or an insight that I have. And then the output of that is the final form factor of what the audio would look and feel like. And then gets, you know, you gotta make sure that you're producing truthful content and the correct content and that you're not speculating or creating conspiracy around the information. And the AI is really good at doing that. I like to use this analogy of when we're building this product, I think about it as a bowling ball, for example. And when you go bowling, you have the, you know, the pros that you kind of throw the ball, you hit a strike, it goes across the grid, no problem. And then you have the other folks, like myself, who need the guardrails on the left and right side so that when you throw the ball, you know, it hits the left and right guardrails to eventually hit the middle pin. I need to add guardrails around a product so that the quality of the content comes out to be of the utmost to its potential and to not get too technical. That's kind of like the analogy that I like to use. There's safeguards in place, quality control, filters and metrics to identify what would then be considered a high quality episode versus a low quality episode, and to ensure that only the highest quality episodes are getting pushed out.

[Manveen Rana]

Right. So there is a filtering system there. Some of them don't make it. They don't get published.

[Adam Levy]

Correct.

[Manveen Rana]

Tell me about going live with this. What happened when you released the first few? What was the reaction like?

[Adam Levy]

The reaction was nobody cared.

[Manveen Rana]

Oh.

[Adam Levy]

And that was. That was totally okay because I didn't do it for that. I did it for the sheer fun of doing something that was creative, something that was challenging and something that kept me super curious. And I'm the kind of person, I spend so much time tinkering with my ideas. This was just another idea in the shed. And then three days are into it and I was like, oh, wow, this is actually getting listenership. Four days. Okay. Whoa. Like, this is actually. People are actually listening to this. People are commenting, leaving reviews on Spotify and Apple podcasts. And they're not just like bad reviews. They're like four star reviews, five star reviews. People really love this. Six days. In. Seven days in. In seven days, we hit 100,000 downloads.

[Manveen Rana]

Wow.

[Adam Levy]

Seven days, two weeks into it, it's

[Manveen Rana]

very clear that everybody listening, they know this is AI. They know it's AI they're listening to.

[Adam Levy]

Yeah, they do know. They do know. Some don't know. Although I try to be very clear about it. In every pre roll, I say that's an AI native investigation. In the description, I say that this is AI native in the channel, I say it's AI on the website, I say it's AI. So maybe the production is that good that some people can't tell until they maybe hear the nuances of what an AI conversation could sound like.

[Strawberry Me Career Advertiser]

Welcome to the Epstein Files.

[Cachava Advertiser]

I'm joining you as the lead analyst on this project.

[Adam Levy]

Today is Monday, February 2, 2026, and if you've been following the news over the last month, you know why we're here. But we need to be very clear from the outset, this isn't a retelling of the stories you've already heard.

[Cachava Advertiser]

No. This is not a true crime podcast. It's not for entertainment.

[Adam Levy]

This is a forensic audit.

[Cachava Advertiser]

Precisely. And the reason we're doing this now

[Manveen Rana]

is because of a piece of legislation,

[Cachava Advertiser]

the Epstein Files Transparency act, or eft,

[Adam Levy]

to try not to trick anybody and to be forward that this is AI

[Manveen Rana]

generated and it hasn't stopped it taking off. You talked about it getting to 100,000 listeners. It's had a few more since then.

[Adam Levy]

Yeah, yeah. In about 30 days. We crossed 2 million downloads by then.

[Manveen Rana]

How many episodes had you put out?

[Adam Levy]

Well over 100.

[Manveen Rana]

Were you surprised by its success?

[Adam Levy]

I was so surprised. I was. I was in utter shock. I was not expecting this to happen at all.

[Manveen Rana]

An AI generated podcast made in a single weekend, topping the charts. So is this just a clever gimmick or is it the start of something that will change the media landscape?

[James Ball]

I'm James Ball. I'm a tech journalist and author, and I'm doing a PhD on AI.

[Manveen Rana]

Just the man. Are you also, by any chance, a fan of podcasts?

[James Ball]

I am an obsessive listener of podcasts. I dread to think quite how many I get through in a week.

[Manveen Rana]

Including ours, I hope.

[James Ball]

Oh, of course. First. First one I listen to.

[Manveen Rana]

When did you first come across the Epstein Files as a podcast? What did you make of the idea? You know, as both a podcast aficionado and somebody with an interest in AI,

[James Ball]

I think I first encountered the podcast just on my Social media feed. And I've sort of had a little running interest in AI generated podcasts because I made one of my own.

[Manveen Rana]

Ah.

[James Ball]

I saw a joke from a webcomic a couple of years ago where this Guy imagined a 247 continuously running AI podcast just talking about polling numbers.

[Manveen Rana]

I kind of somebody's idea of fun.

[James Ball]

Well, I thought, hey, the text's come on enough in the time between him writing this and now. And so I made it. And it was the single worst thing I've ever listened to. Cause it was just the. Every five minutes, it generated a new five minutes of audio of these two voices just endlessly discussing Donald Trump's approval ratings. And I have to stress, mine was a joke. It was like a proof of concept, and it was awful. And so I was really curious as to how good this one would be. And I did have to say it was a damn site better than mine was.

[Manveen Rana]

Tell me about it. You know, if you were reviewing it as a podcast, what was it like as a listen?

[James Ball]

So I think the thing with the Epstein podcast was it's quite a concentrated set of documents. It's quite a concentrated story. And so there was this ability for it to have a huge amount of source material. And so especially the early episodes, it actually seemed to have a bit of a focus. It could say, we're going to say, what do these say about Epstein as a person? What do these say about his business operations? What do they say about the trafficking operation? And so it actually was quite focused. It could talk to sources quite a lot. It was a bit weird to listen to. It sounded like it was default on 1.5 speed.

[Adam Levy]

And.

[James Ball]

And it wasn't very good at explaining things. It didn't tend to understand where you might want to jump back and explain something or give some background. But, you know, it sounds like two hosts talking. It sounds a bit anodyne. It sounds a bit like a US breakfast show.

[Manveen Rana]

Does it sound quite human? Does it sound like the sort of show you would hear if you tuned in anywhere else?

[James Ball]

I think most people, if they listened to even 30 seconds of it, would probably clock that something's off with it that it isn't. Actually, people, before we can even begin

[Adam Levy]

to talk about the network or the

[Cachava Advertiser]

crimes, we have to answer the one question that's been at the summer of this for 20 years.

[Strawberry Me Career Advertiser]

Who is Jeffrey Epstein?

[Cachava Advertiser]

Not the monster, not the caricature.

[Adam Levy]

No. We have to strip that away for a moment.

[Cachava Advertiser]

Yeah.

[Strawberry Me Career Advertiser]

We know the ending, we know the headlines.

[Adam Levy]

To understand how it all happened, we have to look at the man on paper, the man before the infamy.

[Cachava Advertiser]

And that's where it gets interesting, because when you pull the vital records from

[James Ball]

the FBI's conspiracy consolidated, there's lots of little convincing bits. There's pauses, there's hesitations. They'll make little asides about filing their own taxes or having to fill out a power of attorney form, which is a bit odd when you realise that it's a computer generating it all.

[Manveen Rana]

Is this the best version of an AI podcast you've heard so far?

[James Ball]

It is by far the best version of an AI podcast I've heard so far. And I think, you know, the. The creator deserves a bit of credit. He's managed to do things like a couple of extra passes on scripting that seem to make sure it sort of plays quite clever, that I think it ties very closely to the gimmick of it being about the Epstein files. Now, these are very important documents and a very important story, but they are also a pop culture phenomenon in a way. And they're also easy to treat as if they're quite self contained. And so the podcast makes a big deal about referencing primary documents. The hosts, quote, unquote, hosts always flag up that every document named is linked in. The show notes that you can go and check them out. There's too many documents to read. People have the sense that the documents are important. The idea that an AI could read them all and find the most important to you, I think that sounds right psychologically to us. You know, maybe our AI hosts can read more than a human set of hosts ever could. I don't think that's true. And I think the narrative descends and it's not actually a good podcast, but I think as a conceit, it's a lot better than other AI podcasts.

[Manveen Rana]

Yeah. The idea of using it as a tool to trawl through endless amounts of information, which newsrooms are struggling to do, you can see how that's a benefit, how that might be a good thing. One of the weirdnesses of this as a podcast, I mean, and we should say it has been unbelievably successful. Two million downloads. I mean, that's the most successful AI podcast, surely. It's more successful than most human podcasts. It is labeled as sort of entertainment and true crime, which for journalists, looking at this story, which is so much more about sort of grooming and victims and holding power to account, it feels slightly at odds. What did you make of that?

[James Ball]

I think this is probably more a problem of the True crime sort of genre, which there's no shortage of podcasts there than it maybe is the AI issue itself. I think the problem is there are two Epstein stories. There is a very real story of human suffering where hundreds of girls and women were dragged into a sex trafficking ring. But there's a much more, quote, unquote, fun story with Epstein around. Well, which intelligence agency was he working for? Was it all blackmail? You know, which other people I don't like were in there. And people do treat it as a soap or as entertainment, and I think this taps into that. But oddly, that might actually make it quite authentic. There are quite a lot of podcasts with pretty poor ethics out there, and so it doesn't necessarily, for me, stand out on that front.

[Manveen Rana]

Talking of ethics, James, you're a journalist. You know, if you were publishing something on the Epstein files, you have to be so careful. Legally, you have lawyers trawling over it because of all the things that we can definitely say and the things are still uncertain. You know, the files are presented to us as just a data dump. It's a really difficult, thorny subject. How would you feel as a journalist about publishing an AI version of those files? I mean, legally, is that a risk?

[James Ball]

I don't think anyone with even basic journalistic training would go anywhere near publishing this podcast in the way that it has been published. It absolutely terrifies me as someone who has worked as an editor of an investigative newsroom. You know, I've published investigations on Epstein and Andrew and they go through an awful lot of lawyers, because one sentence being wrong can cost you hundreds of thousands, if not millions. We know that there are all sorts of unverified and unvetted allegations in the Epstein documents. There are sort of reports from FBI tip lines that are just recorded, unvetted. And sometimes there are pranksters in there. Sometimes it's people who are sort of battling mental illness who hallucinate things. There are just also unverified allegations that can't be proven. And that's sort of fine for the government to publish with its legal protections. If you repeat it, it doesn't matter that a public document said it. If you spread that allegation and a rich person takes objection, you can be sued for it. If you do things that lead to victims being identified in some jurisdictions, like the uk, that can actually be a criminal offence, that can actually get you on the Sexual Offenders Register. And even if you think you might have some legal defenses on it, even if you think it might be accurate, defending these can cost a huge amount of money which you don't necessarily get back.

[Manveen Rana]

Yeah.

[James Ball]

And AI does make mistakes. You know, I am not a sort of AI doomer. I like these tools. I think they're interesting. But one thing I have found is they will often just misattribute a quote in a document, and that's when they're reading a hundred pages, not several million pages. And it just takes one episode to say the wrong thing about the wrong rich person and you could find your life absolutely ruined. And yeah, I think you have to really not know very much about the law at all to sort of jump in and assume it will be okay.

[Manveen Rana]

Coming up, the creation of this AI podcast has raised some serious questions. Who takes responsibility if the AI gets it wrong? And how do we know what's real and what isn't? We'll have more with Adam Levy in just a moment.

[Adam Levy]

Foreign.

[James Ball]

AI is transforming customer service. It's real and it works. And with fin, we've built the number one AI agent for customer service. We're seeing lots of cases where it's solving up to 90% of real queries for real businesses. This includes the real world, complex stuff like issuing a refund or canceling an order. And we also see it when FIN goes up against competitors. It's top of all the performance benchmarks, top of the G2 leaderboard. And if you're not happy, we'll refund you up to a million dollars, which I think says it all. Check it out for yourself at fin.AI

[Cachava Advertiser]

craving the coffee flavor you love, but without the caffeine. Caciava's got you covered with their newest coffee flavor. This all in one nutrition shake delivers bold, authentic flavor. Crafted from premium decaffeinated Brazilian beans with 25 grams of protein, 6 grams of fiber, greens, and so much more. Treat yourself to the flavor and nutrition your body craves. Go to kachava.com and use code smoothie. New customers get 15% off their first order. That's K A C-H-A-V-A.com code smoothie.

[Strawberry Me Career Advertiser]

Let's be completely honest. Are you happy with your job? The fact is, a huge number of people can't say yes to that. Too many of us are stuck in a job we've outgrown or one we never really wanted in the first place. But we stick it out and we give reasons. Like, what if the next move is worse? And I've put years into this place, and maybe the most common one. Isn't everyone miserable at work? But there's a difference between reasons for staying and excuses for not leaving, it's time to get unstuck. It's time for Strawberry Me. They match you with a certified career coach who helps you get from where you are to where you want to be, either at your existing job or by helping you find a new one. Your coach helps clarify your goals, creates a plan and and keeps you accountable along the way. Go to Strawberry Me Career and get 50% off your first coaching session. That's Strawberry Me Career.

[Manveen Rana]

Adam, clearly you'd hit on an incredibly good and incredibly popular idea. While newsrooms were struggling to make sense of the Epstein files, you had AI that was able to sort through the information and put out very polished AI podcasts about what it found along the way. As a journalist doing it on the other side, we had to be quite slow with, with our coverage because it's such a legally fraught subject and also sort of such a sensitive subject. You know, we know that some of the earlier data drops, they were releasing information about some of the victims that shouldn't have been there. So there was a lot of very careful sorting through the files to make sure we weren't publishing anything, you know, that would damage one of the victims, but also that we weren't going to get sued because there was obviously not all of the information there was good. How did you approach that problem?

[Adam Levy]

Part of these LLMs, they are a black box to an extent, so there's only so much that is in your control, but you could build guardrails to the best of your ability to ensure that they do follow a particular format and structure. So when it comes to the Epstein files, there's a lot of fact checking that you need to do and that you can also program and build into the product itself to ensure that it filters information, doesn't speculate on information and documents, what's shown in emails, in the videos, in the images, in the court releases. You know, that's fair game. That's all fair game.

[Manveen Rana]

I mean, we found, you know, there's such complexity around this particular subject. There were some things that you could report as fact and others that even if they appeared in the files, even if they were in an email, you couldn't because, you know, we know some of the emails, for example, were drafts that Jeffrey Epstein had written to himself, or because you didn't have the context around them to know whether they referred to the thing you thought they referred to. You know, there was so much complexity. How does AI deal with that?

[Adam Levy]

I think the thing that you're pointing out, if I understand correctly, is it's a biased opinion around what one believes is true or not. And I really tried to build this series around just seeing what the data shows and talking about the data at hand and not speculating on the data, ensuring that it's sticking to what was said in the emails, conversations that happened between people, and just bringing light as well to who Jeffrey Epstein was.

[Manveen Rana]

And is it able to sift between, you know, because it's about accuracy, but it's also the fact that not all of those documents are created equal, if you know what I mean. You know, some of those will be documents between people, some will be notes to themselves. You know that there are many documents that we haven't been allowed to report legally as being accurate, even though they're part of the Epstein files because of the quality of the information. Were you worried about the legal risk? I mean, could lawyers still come for you?

[Adam Levy]

I don't think anybody's ever fully safe. I think the approach that I took in the guardrails that I put into place and the disclosures I included to document the story and providing additional sources so that I'm not that the AI is not rather building conspiracies for what it's worth. I mean, I can go into what ifs all day, you know, what if a meteor hits? You know, like, there's only so much that's beyond my control in terms of who comes after me and what happens next.

[Manveen Rana]

Is it possible for AI not to have biases because of the models that we've seen so far? You know, we can almost identify what their biases are. Whether that's GROK or, you know, Gemini, we kind of know that, you know, it still has a way of presenting information to you which in some ways represents a bias.

[Adam Levy]

Yeah, look, you can't make it non bias 100%. I've actually learned that most of these LLMs, they're more left leaning, and GROK actually tends to be the one that's more in the middle more often than not. And it's super important that you have what we call as part of the engine, the truth machine. Trying to get to the source of truth and doing it with a degree of confidence that you are presenting information with the utmost amount of truth. And if it doesn't pass a certain mathematical score that it is as truthful as it can be, and that the tonation and the word choices don't lean too much to the left or too much to the right and lean more nonpartisan, you know, then it's good to pass. I think AIs could actually do a better job at doing that than humans, because AIs are really good and could also be really bad. I've learned at following rules, but if you know how to program the guardrails correctly, you could get a quality output from the black box

[Manveen Rana]

on that. There was a review on Apple of, of the Epstein files and, and I have to say, as a fellow podcaster, never read the bad reviews, but this one raises sort of an interesting philosophical question. It just, it says it's fine to use AI to trawl the data, but obviously we still want humans to be in charge of the investigation, nuance and telling the story. You know, Is that a worry for you? Is that part of the, the picture that you're missing?

[Adam Levy]

No, because I think humans are. Have a big part in curating this, and I think the point that they're referencing is more so the human voices. That tends to be the underlying factor. People want to see humans speaking in humans conversing because there's a plethora of other reviews of people really enjoying this and liking the approach of it being quick, straight to the point, no spin. I, I think there's validation in what this person is saying, and I think that's just the, the headwinds that new, new technologies face. Frankly, it's just part of the, it's part of the growing pains.

[Manveen Rana]

Adam, would you see yourself as a journalist? What do you make of journalism right now?

[Adam Levy]

Not by trade. I'm not a journalist by trade. I'm a content creator. And I have a set of values that I run my life with. I like, to be honest, I like the facts. I don't like bs. I like when things are said for what they are. And I try to create content and build product around some of those values. So I can't speak to journalism as a whole because that's not my trade. I just know what I would like to get from the media and information that I consume. And if nobody else is going to do it, I will.

[Manveen Rana]

If this becomes the way people synthesize their information, you know, if they're listening to your podcasts on the war or on the Epstein files and not listening to journalistic ones instead of that sort of steals from, I suppose, the business model of journalism a bit. Are you afraid that actually you could end up damaging some of the sources that your AI relies on?

[Adam Levy]

I mean, that statement goes with assuming that today newsrooms don't touch AI at all in their production process. So, like, they're using Google Search, not using Gemini or ChatGPT or Claude to bring a product to life. Maybe the end product involves a human reading and conversing and, you know, doing their, their segments, whatever it may be. But I'm very doubtful that whether it be the Times, New York Times, you know, the New York Post, or whatever other media outlet that has authority with the mainstream public is not already using AI in their day to day workflow. So frankly, what's the difference?

[Manveen Rana]

I suppose the difference would be that while you can use AI on certain parts of ordering information, it's still having a human over it to make sense of what they're seeing, which may be very good on what it can see, but on the human understanding, you know, war, mass trafficking of female victims, these are very human stories. Are you missing something? By having podcasts that are entirely generated by AI, where that entire process, a very human story, is told entirely by something that isn't human.

[Adam Levy]

I like to think that you could build software around that and capture the human elements of doing so. Yeah, I remain optimistic that you can program those guardrails in the future. Look, I'm not here to completely be dismissive of what risks this could, you know, what its limitations are. You know, I've listed some of them on this as well. But we're only as good as the tools that we have at our disposal. So there's also a responsibility on these LLMs and the companies at large like OpenAI Anthropic to produce accurate models, models that are reliably producing information that we can then rely on, you know, with confidence. And I'll remain in the camp that I believe I'm, I'm, I'm optimistic, you know, I think that's going to pan out. And I think while there are bad actors in the world, I'd like to think that there's some good ones as well that want to see this technology win in the right manner to benefit society. In the context of media, you know, speaking in the context of media, How do you feel about this, this and veen?

[Manveen Rana]

You know, I obviously see the potential, but I also see so many potential pitfalls. And my fear is that the potential will seem so appealing that media companies, everybody will jump on it very quickly to the point where you almost forget how you do the basic skill of finding stuff out on your own. I worry about the future of truth in that world. I think we become so reliant on AI to synthesize it for us and not necessarily to check because you can't at every possible turn. And in the end it's kind of the truth that disappears. It's, you know, maybe if everybody believes something, it's fine for it to be the truth. I don't know. I personally don't. You know, I personally, as a journalist, you're slightly obsessed with getting accuracy and fact and trying to find out the truth behind something. And I sort of suspect that's probably going to be the first casualty. There will be a version of the truth, but there won't be the same search for it.

[Adam Levy]

I think it's a fair, fair, fair opinion. Yeah, I think that's a very fair, very fair reflection for sure. And I think about that as well.

[Manveen Rana]

That was Adam Levy, creator of the Epstein Files podcast cast. And before that, you heard from James Ball, a technology journalist. The producer today was Robert Wallace, the sound designer was Dave Creasy, and the executive producer was Kate Ford. Thanks for listening. We'll be back tomorrow.

[Cachava Advertiser]

Craving the coffee flavor you love, but without the caffeine? Cachava's got you covered with their newest coffee flavor. This all in one nutrition shake delivers bold, authentic flavor crafted from premium decaffeinated Brazilian beans with 25 grams of protein, 6 grams of fiber, greens and so much more. Treat yourself to the flavor and nutrition your body craves. Go to kachava.com and use code smoothie. New customers get 15% off their first order. That's K A C-H-A-V-A.com code smoothie before

[Adam Levy]

we had ATT business wireless coverage, our delivery GPS wasn't the most reliable. Once our driver had to do a

[Strawberry Me Career Advertiser]

14 point turn to get back on route. A 14 point turn.

[Adam Levy]

An influencer even livestreamed the whole whole thing. Not good for business. Now with AT&T business Wireless routes are updating on the fly and deliveries are on time. And the influencer did get us 53 new followers though.

[Strawberry Me Career Advertiser]

AT&T business Wireless connecting changes everything.