Episode description:

Welcome to In Reality, the podcast about truth, disinformation, and the media. I’m your host, Eric Schurenberg, long-time journalist and media executive, now the founder of the Alliance for Trust in Media.

You cannot talk about the future of media, or the future of anything for that matter, without talking about AI. What generative AI does is pretty much exactly what journalism does: digest information, highlight what matters and render it to an audience in a fetching, attention grabbing way. For journalism, is that a good thing or a bad thing? Well, judging from history, it’s both or either: all depending on how it’s used. There’s already alarming evidence that Google’s pivot to AI summaries in its search, rather than links, is decimating referral traffic to newsrooms. It’s also undeniable that those same newsrooms are using AI to automate the grunt work of information gathering for journalism.

To explore both the peril and promise, I’m joined by two guests who approach this upheaval from opposite sides of the media-tech divide. Jessica Chan is head of publisher partnerships at Perplexity.ai, a rising player in AI-powered answer engines. Troy Thibodeaux leads AI strategy at the Associated Press, a legacy newsroom that’s long been a leader in automating journalism with integrity.

This conversation—recorded live at my University of Chicago class—dives into the licensing dilemmas facing publishers, the safeguards newsrooms are building around generative content, and the hopeful ways AI is being used to personalize, streamline, and even monetize trustworthy journalism.

 

Transcript

Speaker 3 (00:00.308)
Now let’s bring on Troy and Jessica who’ve been with us since the beginning of the class. We are very lucky to have them on board with us. They are two experts on artificial intelligence and the intersection with media. And they come at it from very different parts of the business, very different parts of this issue. Jessica Chan is coming from the technology side. She’s the head of publisher partnerships at perplexity.ai, one of the biggest of the AI companies, and a resume that found her following a path through Meta and LinkedIn. So she comes from the big tech side. Troy Thibodeau is the head of AI products and services at the Associated Press, and his career has been all about helping journalists manage new technologies as they came along and using the tools that technology presents. Now with AI, he’s probably got the tool of all tools. So Jess and Troy, welcome to the class.

Thank you, Eric. It’s very nice to be here and it’s nice to see you all.

Thank you, Jess and Troy, are you? Okay, good to see you. All right, so here you are coming at this confrontational issue that from two different sides of the question, you heard kind of the range of views about artificial intelligence in the first question that the students answered. Maybe you guys could help us frame.

Speaker 2 (01:19.31)
Yes, I’m here. Thanks for joining me.

Speaker 3 (01:46.444)
What we should be thinking about when we talk about the intersection of media and artificial intelligence. I have to say that, I mean, the range of metaphors that you could use is pretty wide. It’s either, as someone alluded to earlier, it’s just math, it’s just computers working together. Others use a metaphor of it’s kind of an eager intern who just wants to please, but doesn’t have any judgment.

And then another way to look at it is that this technology is growing so rapidly. We don’t know exactly gonna end up. So Jess, I invite you first kind of frame it for us. How should we think about artificial intelligence and media?

I think maybe what might be helpful here is to explain a little background about perplexity, how we work and how we differ from other platforms. And then I can share a little bit from there what we’ve built in relation to news publishers and how we think about AI in the new space. This sounds good. So.

Take it away, Jess.

For those who don’t know, at the most basic level, Perplexity is an AI answer engine. So if you haven’t played with it, I encourage you to check it out. You know, when you look at the internet the last 20 years, someone types in a question, they get back 10 blue links. With Perplexity today, you ask a question, you get back an answer with citations and sources that are linked very visibly above the answer.

Speaker 1 (03:22.252)
Now, a big difference to note between Perplexity and the other AI platforms is really how we work. With Perplexity, what we are is we’re a combination of an LLM, so a large language model and a traditional search engine. When you think about wanting to answer someone’s questions, there’s really two things that need to hold true, and that’s accuracy and real-time information that’s useful in answering people’s questions.

With a lot of the other platforms, what they do is they train on lots of data and then they use that model as a store of knowledge to answer people’s questions. And as people have experienced it, perhaps you’ve experienced it yourself, it can make stuff up. And at some point that knowledge cuts off because it’s only as recent as the data that it’s been trained on. With perplexity, we’ve really taken this different approach where we use the model for synthesizing and summarizing and then we take the open internet.

And that’s where that traditional search engine part comes in for perplexity. That comes in for the facts and figures that’s useful in answering people’s questions, which is why it’s more accurate and more real time. And there’s not that knowledge cutoff. We actually don’t build our own foundation models. And unlike the other platforms that rely on a single proprietary model, partner, we instead partner with multiple language models simultaneously, including

GBT-4, CLAW, DeepSeqs R1, and then we select the most optimal model for different query types. But for us, Perplexity was fundamentally designed as an AI-powered answer engine rather than a conversational chatbot. I think that architecture really prioritizes information accuracy and kind of retrieval over creative or open-ended dialogue, making it really well-suited for fact-finding missions, which is why I think you know, when I have conversations with publisher partners and I have conversations with journalists, they tend to find perplexity to be a lot more advantageous or useful in, you know, in helping them do their work more efficiently. It’s not about replacing them or, you know, replacing their work. It’s really about how do we make things more efficient in the newsroom?

Speaker 3 (05:38.926)
I’m going to ask Troy to take a swipe at that question and we’ll get to some of those things that you’ve alluded to in a few minutes.

Yeah, so one of the things I noticed in the in the discussion at the beginning, you know, I like the way that you frame the question. Is it it going to be? I can remember the exact terms used, but one of the I’ve seen often it paired as is it peril or promise, right? Or is it peril and promise? And I think what we’re looking at now is that it’s both right from the news industry. Certainly there’s a risk involved and we there’s a risk of loss of trust. There’s a risk. You know, a lot of things you all mentioned about some of the problems with disinformation and misinformation certainly are there.

The business model, I think there’s a challenge there as well for a lot of publishers. So in my job, I’m trying to maximize the promise and minimize the peril. And one of the things that Jess said I think was really important is that these are large, and we talk about AI, we’re talking about kind of an umbrella term for lots of different technologies. But what has everyone excited right now are these large language models. And one of the things I like to remind folks about is that these are language models, not knowledge models. And so with Jess is talking about where people rely on the foundation model and get the answer from its training data, that’s a lot riskier than something that’s going to use the language model to generate an answer based on a subset that you trust. So here is where the answer should come from. Now, language model, you can phrase it.

There are still risks. They still do hallucinate, right? We know that they still can misconstrue something within that constrained environment, but it really minimizes the risks. So that’s a lot of things that we think about. At AP, trust is everything for us. Accuracy is everything for us. Our tolerance for inaccuracy is zero. People get fired if they’re inaccurate too many times in a year, right? So we have a really low tolerance threshold. Someone in a different field, like marketing, for example, they might have a little more tolerance for this. There are other places where close is good enough.

Speaker 2 (07:40.916)
And with journalism, that’s not the case. So we have a policy right now that anything that’s produced by generative AI, not some of the other technologies we’ve used, but anything produced by generative AI has to go through an editor or reporter, be vetted and make sure that it’s correct before it goes to our members and customers. So we are both a B2B. We sell our content to other news organizations and we do have a smaller B2C. We have apnews.com. So before it goes up on our website, where it goes to our members and customers, a human being is looking at it. And that’s where we are right now. That’s the level of trust we have for the technology.

Okay, great. All right, so human in the loop for trust. We have talked in this class a lot about the business models that underlie media and the effect they have on the health of the information environment. So I’d like to start there with the business model challenges. And I guess the thing that comes to most people’s minds when they think about this right now is the friction between publishers and… and AI models. That’s what gets all the news. There’s plenty of cooperation, which we’ll talk about too, but the thing that’s in the news is the lawsuit that the New York Times has leveled against open AI and perplexity has its own issues with other publishers. The Times’ argument and arguments that other publishers like Forbes have made is that the

The AI model creators to create the training data that they need to use to train their LLMs are basically taking the hard work of journalists, scraping it off the internet with no permission, no compensation, and then using it to train these models that are based on their market caps worth billions. And then that violates copyright and fair use.

Speaker 3 (09:46.926)
But copyright and fair use are really ideas that are created for a different era. And the application to this particular technology is hard to see. Jess, I would give you the opportunity to respond to that claim by publishers like the New York Times that the use of AI to consolidate information that they have.

generated from their reporters and so forth is a copyright violation. Do you want to take a swing at that?

So, look, I can’t speak to specific, you know, I can’t speak to the lawsuits, you know, given the ongoing legal proceedings. But what I can say is that with perplexity aligned to what Troy actually just said about AT about AP is that we really focus on trust, both in technology and in the way that information is shared and credited.

With Perplexity, our mission is to make information more accessible and useful by synthesizing knowledge from a wide range of sources. And we always strive to provide clear attribution and links so that users can verify the information that they receive. I think Perplexity’s approach really emphasizes transparency and users obviously are encouraged to check original sources, which supports both user trust and publisher recognition. I think, you know, we believe

that synthesizing information in new ways is transformative and can coexist with original journalism. And we aim to help users understand complex topics by drawing on multiple perspectives. Trust is built when users know where information comes from and can verify it themselves. Perplexity’s model is designed to really foster that trust by providing those up-to-date answers with source links rather than hiding or obscuring where information would…

Speaker 1 (11:43.446)
was originated. think this transparency is a cornerstone for building a responsible AI ecosystem. I think where it all shakes out, no one really knows, but I believe that we’ll continue to see new forms of collaboration between AI companies and publishers.

Thank you. Thank you. That is, yeah, it is important to emphasize that point that perplexity, unlike the other large language models that you think of immediately shares the links. So it tells you where the information comes from. And you sort of are left to guess and trust in the designers of the model with say, chat GPT. Troy, AP was one of the first publishers to strike a deal with large language models and basically gets a large

Fell off.

Anyway, that you get a licensing fee, AP does, from OpenAI. I guess that agreement might be expiring soon. I think it was only two years. Maybe it’s already been renewed. But you chose to go that way rather than suing and standing up for copyright. You know the company’s philosophy about why it was done that way?

and whether this is either a stock gap until a more permanent arrangement is settled on, or is this kind of the way the relationship will go in a business sense?

Yeah, I I will say upfront, I don’t know what the future will hold with these relationships, right? But I was part of the conversations about the deal we did with OpenAI, which was, you with a press release about it. And honestly, for the AP, it was clearer than it might be for some other publishers because at our core, we are a B2B, right? So we have a member model and we also have customers who are not members and we license our content. It’s something we’ve done for decades, right? So…

For us, this was a licensing deal. It was a source of revenue. So we were able to, and there are two different kinds of deals here. One is for the training data. So we can say, here is a piece of our archive. You want reliable, clean data that’s non-toxic. We’ve got some really good clean data that’s been fact-checked and is high value. So we will license that to you as we would to any other, to a publisher, and you will train your model on it you’ll have better results.

Speaker 2 (14:27.362)
The other is a display use case, is something we’ve talked with publishers about as well, which is, you know, now you have the right to publish our content, attribute it to us in your answers. And so that’s another kind of way that we can partner with technology companies. And for us, think that was kind of a natural thing to do. You know, we’re used to selling our content, we’re used to licensing our content. For other publishers who don’t do this routinely, maybe that wasn’t as obvious a path for them.

But I think for us, it has been a benefit. We’ve been able to generate revenue from this. I don’t know in terms of the larger news ecosystem how this plays out. I don’t know how the court cases are going to play out. So there’s a lot of things I just don’t know. But I know for us right now, it felt like the right thing to do. And it really has helped us kind of keep things, keep the ship afloat and keep moving forward. It’s because we’re always looking for new forms of revenue.

Okay. One more question about the business models. And I take your point that AP’s business model is different from someone who, some publication that speaks directly to consumers for the largest part of income and reason for being. Jess talked about blue links showing up in the traditional search engine. So the way that worked was you put in a query to

Google and you got back a bunch of blue links, you could click and go to the newsroom that created that story that may or may not directly answer your question. But what it meant for the news publisher was that it gets an eyeball, which it can monetize. If the answers are now all embedded in AI models, that traffic is pretty much dried up.

And I had a particularly kind of fraught or memorable, I guess I should say conversation with Nick Thompson, the CEO of the Atlantic, who also struck a deal with OpenAI, a licensing deal not unlike AP’s. His reasoning was different. He said, we’ve seen our traffic dry up as much as 80 % of our referral traffic disappear because of

Speaker 3 (16:50.872)
Google’s transition to an answer model rather than a link through model. And that the technology companies have now offered us this pot of cash and I could not turn it down under the circumstances. I just wonder if you guys have some insight on how things will evolve if those blue links no longer translate to traffic for the… for the media companies.

Should I take that? Yeah, I would say when I first saw some of the generative search coming out, it scared me, quite frankly. And I thought this is going to be devastating. I expected people started talking about a phenomenon they called Google Zero, When the referral traffic from Google, which is the dominant source of traffic for most news publishers today is from search, Google being other search engines, but primarily Google.

Troy, go ahead.

Speaker 2 (17:52.938)
If that were to drop off to zero, it would have a devastating effect. This was the fear. And I think that’s legitimate fear. However, the research I’ve seen recently, there was a recent study, I think it was from Northwestern University that said it’s not happening yet. That what they’re seeing across publishers is not a huge drop off in traffic yet. The anecdote you mentioned is what I was expecting. It’s not happening yet. That doesn’t mean it won’t happen, but it isn’t happening yet. And so I think there are a couple of things.

Going on here, one is there are questions and there are questions, right? And what are you asking? If you’re asking what was the score of last night’s game, well, there have been search engines that would tell you that where you didn’t have to click through. That’s been in existence for a very long time, right? So that kind of traffic to those kinds of simple answers, I don’t think people need to go to new sites anymore. They’re not going to new sites.

There are deeper questions where you really want a more discursive answer, where you really want to dig in a bit more, where you do want to click through, you do want to read more. And I think what we have to ask ourselves as publishers is what are we offering that goes beyond what you can get from a simple answer to a question? And primarily, what kind of relationship are we developing with our audience so that they come to us, so that they’re not stumbling upon us from social or search?

But really, how do we build that direct relationship and that direct traffic? I think that’s really where we need to be focused going forward.

Okay, Jess, I see you’re nodding. Do you have a thought to add to that?

Speaker 1 (19:27.02)
Yeah, so I think Troy raises some really interesting points. From my discussion with publishers, it’s hit or miss. I’ve had a number of discussions with publishers who say that their traffic to Google has declined significantly and some where it hasn’t been impacted so much. But I think for us at Perplexity, a big focus is really how do we work with publishers, whether or not your traffic might be declining?

You know, I kind of, I alluded to this earlier, I would say, you I think you’ll see new business models emerging. For us, we launched our publisher program, I think as this kind of new model or attempt at this new model or first evolution of that, where for us, what we’re doing with publishers is we are sharing in revenue with participating outlets when their content contributes to an AI generated answer. And we’ve monetized on that quarry.

You know, we’ve had pretty major publishers sign on as part of this program already. And so, you know, it’s for publishers, I think it’s not just about gaining new revenue streams, but it’s also how like the way that I look at it is how can we support them beyond revenue in some of their other initiatives? So for example, as part of our publisher program, we also give publishers access to our API, which allows them to create their own version of

know, perplexity or essentially enhance their sites with AI and we power that infrastructure for them. And we support them from a, you know, a technological perspective. We also hear from publishers that they are thinking about, you know, or they’re not even thinking of, they’re already leveraging AI in their newsrooms. And so how can we support those efforts as well? And so we also make that part of our publisher program. That in my opinion is just kind of the first iteration of our.

our program, my remit in coming in is really to talk to our publishers, figure out what kind of things work for them, what kind of things make sense, and to continue to evolve and build programs and partnerships, product partnerships that will help them hit those true Norths. And so, for example, if it is things like driving subscriptions, for example, how is it that we can think about evolving our program and our products to support things like that?

Speaker 3 (21:47.79)
Okay, great, great. Well, that’s a great segue to the next kind of phase of the conversation that I hope we would get to, which is about the positive uses of AI in newsrooms. We heard at GPT just tell us that there were all sorts of efficiencies to be had from AI in the newsroom. Troy, that’s really kind of your specialty right now. So tell us about some of the things that AP, which by the way is a real leader in this field.

about some of the things that get you excited about the way AI can be put to use to enhance journalism.

Sure, and I noticed that Chachi B.T. mentioned a few of the things that we’ve done, so that was nice to hear. So we started where I think a lot of publishers did. We were a little bit early on this, but we started with news production.

So what are the repetitive tasks, the kind of boring tasks or the things that are just time consuming that we can take out of our journalists way? So we did translations and transcriptions and we generated headline ideas and we did summarization, both bullet summaries and kind of a broadcast summary for our broadcast customers. All of these things were, I think of them as all versioning tasks. I think the LLMs are really good at versioning. Here’s one version of a thing, make it Spanish.

Here’s one version of thing, make it shorter, make it bullet points. It does a pretty good job with that, as you saw with your voice assistant. It did a pretty good job of summarizing, you know? It took us a lot of work to get it to our standards. We couldn’t just use the off-the-shelf model. We really had done a lot work on the prompt. We did what’s called fine tuning, where we kind of used some really specific examples and got a better version of the model that worked for us. So it took a lot of work and effort, but we were able to get all of these tools spun up. And then we were thinking about really

Speaker 2 (23:35.63)
what’s the next wave of things that we should be doing with this? And this is I’m pretty excited because we think about kind of the value, the news value chain, right? So it goes from news generation, from news gathering to news production, to kind of presentation and distribution. So we started with that middle chunk and now we’re kind of expanding, looking at, we’re doing a lot of things with news gathering. So for example, when the JFK papers were published, we were able to throw that to a model, get it to do some summarization for us, pull out key keywords.

We also had a model look at what had been unredacted. So we had the old version of these things. We were able to see what was different, right? And we built a nice little tool that let journalists look at that, reporters look at that and say, someone’s social security number’s in here, or phone number, whatever it was, right? So that was a cool tool. We didn’t get any huge revelations from it, but now it’s a toolkit that we can use for future data sets like that. And then on the other side, we’re looking at personalization.

We did our first personalized newsletter. Now this was simply, it was about personalizing content for the user, kind of the things that they’re interested in that they haven’t read yet. And that the things that not only just kind of the topics, but the depth of topic that they tend to study. And so really looking at how do we get a newsletter that’s really hitting the things that you typically want to learn about, but haven’t seen yet, or things that may be adjacent to it in interesting ways. This is tied to some of the older technology.

Kind of recommend your engine, but we built on that and now we can think about using that versioning capability How do we you know, what what shape of story do you really like most? How do you like consume your news? And they could be it could differ for different topics So we’re interested in that personalization not only for newsletters But also for our website as well and one of the ways that we can use Vetted content, right? So this is the kind of the struggle that we have we want to make sure that everything is vetted

by an AP editor. Nothing is going, is being published without that set of eyes. At the same time, we want to create personalized versions of stories or videos. And so trying to work between that and thinking about what are the ways we can break our content down to smaller, interesting modules, modular chunks, and then recombine them in ways that best suit that user’s needs, right? That best, that way that they want to consume the news. I think that’s super exciting. News gathering and news presentation, I think.

Speaker 2 (25:57.614)
We’re gonna see a lot more going on there and we’ll continue. We’ll add new languages to the translation. We’ll do different kinds of summarization. So the news production will continue, but those two areas are gonna be really exciting.

Okay, great. All right, thanks. So personalization and summarization and versioning all things that the models do very well. Jess, from what you’ve seen with your partners, anything to add to what AP is up to?

Yeah, think, I mean, I’ve just been hearing the most creative use cases. Of course, I think besides from the internal use cases of time saving and on research and being able to process data efficiently, or some of the examples Troy actually just outlined in which, and how AP is using it, I’ve had a number of publishers begin launching external use cases, really helping their websites become more sticky, improve user experience, and in some cases really,

power products to increase revenue. So, I’ll call it a couple of examples that are ones that I love to share, but we make our API available to our publisher partners. And then for them, again, they get very creative with how they’re integrating it. So, LA Times is a really good example. They actually integrated our API and what they do is they actually provide a perplexity powered insight or perspective, if you will.

alongside their editorial articles. And this allows readers to see balanced perspectives of a news story. Mostly I their end goal there was to really, you know, I talked to the owner of the LA Times and he was a really big proponent of just not allowing people to stay in an echo chamber of like content. And so that was really their goal around building this use case with AI.

Speaker 1 (27:47.86)
Another really fascinating one is that one of our publishers actually built a tool to make it accessible for paying subscribers of theirs to generate sponsored content. And so, you know, one of the challenges that they found that they were facing was, you know, they were working with, you know, small businesses and these small businesses wanted to generate sponsored content and then in turn do a one-click sponsorship of this content.

it was taking a lot of time, a lot of efforts and resources. And so they built our API into a tool that allowed them to create this offering. And then it’s essentially this paid offering. And so that publisher is actually seeing a significant increase in ARR through the offering of this tool. And so those are probably.

Just let me just jump in for people who are not familiar with the term sponsored content is what you might refer to as some advertorials. So it’s basically information provided by a company that wants to appear as an expert on that or wants to suddenly promote its products.

Exactly. Yes. Thank you for specifying that. But yes, so this publisher generated this tool and they recently shared just the increase in ARR was significant. It was in the millions. Just from them launching this tool, within, I think it was within maybe six months, I’m sorry, six weeks that they saw like a pretty significant increase in ARR.

ARR being annually recurring. Correct. And the holy grail for any media business, the stay in business. right, you. Those are great examples. Thank you. And creative uses. Let’s just talk about the guardrails around AI. Troy, you’ve mentioned there’s always got to be a human in the loop. It’s got to whatever output comes out of these models has to be from vetted information.

Speaker 3 (29:51.854)
You know, we have seen that it is quite possible to misuse generative AI and it doesn’t take a lot of imagination. You saw some of it with Claire’s piece just even recently. There was an AI generated image of Trump dressed up in papal clothes, which offended some people. And there was famously an AI generated

video that was broadcast by a hack to Ukrainians of Zelensky saying, down your arms, we surrender. So that can happen. Apart from the good intentions and the integrity of newsrooms, what are the guardrails that can prevent really damaging misinformation from being spread by AI? What do you think should happen? Is it regulation? Is it something else?

think it’s going to be difficult, right? Because it’s going to be an arms race. There are tools that can try to identify, you know, generated visuals. But they’re always going to lag a bit behind the bad actors. How do you get ahead of that? How do you? And so I think it’s going to be extremely difficult. a lot and there are also some things I think, interestingly, there are some efforts to

tag things at the source. So you’re able to say going from the camera, for example, and then with a chain of maybe this is actually a good use of blockchain, right? We’re still looking for one, but so that you can say kind of keep the provenance of that image from camera to consumer and be able to tell all the way along the way any edits that have been made. So there’s some efforts kind of at both ends, right? Detecting things that are fake or verifying things that are true from the beginning.

But it’s going to be difficult because the technology is changing so fast and because it’s, even though maybe we aren’t seeing some of the giant leaps we saw last year, we’re still seeing pretty significant improvements. How do we keep, excuse me, how do we keep up with that? And so I think that is going to be a challenge. And so I think it’s gonna come down to the consumer, know, that was the Claire’s advice, right? Was to make sure before you share something that you know for sure where it’s coming from. And it’s an old fashioned way of thinking.

Speaker 2 (32:14.572)
But, know, how do we do in journalism? Who’s my source? How do they know that? Why do I trust them to give me this information? And I think a lot of education is going to have to go into this. And I’m working on my kids are 11 right now and they’re learning some of this stuff in school and they’re seeing deep fakes. That’s where it’s going to have to start to have consumers with a critical eye. And I think that’s one of the failures we’ve seen around social media is that critical intelligence has sort of been put on the shelf for the ease of use of social media.

That is a good point. I mean, you could argue that putting critical thinking on the shelf is just human nature, that we are cognitively lazy as a species for a whole number of reasons that we’ve talked about in other classes. Jess, I wonder where you come down on this. It’s one…

One other conclusion you could draw from the case that Troy just made is that the responsibility lies on the consumer to recognize fake news, but also to be able to identify trustworthy sources. And that may be in the long run, rather than being a cudgel to beat down standards-based journalism, this will be the thing that leaps it up because you just have to be able to trust the brand.

Yep. I mean, I think my position, my view on this is like regulation is good, but it’s actually not a solution. know, Troy alluded to this earlier as well. I think the speed of technological advancement cannot compare to the speed of law, meaning regulations will never be able to keep up with a bad actor armed with AI, armed by AI. And so, you you, talked about, you know, I think

We’ve kind of talked about the liars dividend and I think that’s always been a problem. And as media gets more fragmented, I think it becomes even more of a problem. think my position is that, technology or not my, but I guess my position at Perplexity is really that technology is the only thing that will solve this problem. I think regulation can’t keep up. Regulation is a technology for trust, meaning.

Speaker 1 (34:29.844)
regulation functions as a mechanism, much like technology designed to create, to kind of create and maintain and reinforce trust amongst its participants. And so what perplexity is building is that technology for trust. You I think we are the people that will help solve this. You know, I’ll be honest, we haven’t solved it yet, but we know knowing that technology is the only thing that can solve it in this age of AI and that we’re focused on it.

I think that’s really where, you what my position is on this.

Okay, technology as an antidote, good technology as the antidote to bad technology. Let’s let that thought simmer for a moment and let me turn the mic, such as it is, over to the class and invite members of the class to ask questions of Jess and Troy.

---

Created & produced by: Podcast Partners / Published: Jul 11 2025

---

Share this episode: