WEBVTT

1
00:00:00.120 --> 00:00:12.060
[on-hold music] Welcome to The Rubuti Show. I am Brian Morrissey.

2
00:00:12.220 --> 00:00:25.600
Today, I'm speaking with Pete Pachal, a veteran news editor who runs the reliably excellent Media Copilot newsletter. Pete is deep in the weeds about the intersection of AI and news publishers.

3
00:00:25.740 --> 00:00:32.900
It's a topic that is very near and dear to my heart, and I always rely on, uh, Media Copilot to keep me up to date.

4
00:00:33.360 --> 00:00:46.700
Pete is an expert in this area, and we discuss the recent news of DeepSeek's apparent AI Sputnik moment and what it means downstream for publishers beyond the fact that everyone's four oh one K took a hit today.

5
00:00:46.780 --> 00:00:53.820
We also talk about actual use cases of AI improving the news product. Spoiler, there really aren't that many to date.

6
00:00:54.400 --> 00:01:07.040
And the implications of agentic AI, which is where basically AI agents go out and accomplish tasks, and whether that will make websites obsolete. It's a real possibility.

7
00:01:07.500 --> 00:01:28.680
And also the wisdom or folly of site-specific chatbots that we're starting to see from publishers and the underwhelming example of AI news aggregator Par-Particle, as well as why Pete is obsessed with Google's Notebook LM and whether we will see the contours of a grand deal to iron out the economics of news content if AI keeps gaining ground the way it has been.

8
00:01:29.220 --> 00:01:39.120
This episode is brought to you by Exco, the machine learning video platform trusted by leading media groups like Advanced Local, The Arena Group, Hearst Newspapers, Nasdaq, News Corp, and more.

9
00:01:39.500 --> 00:01:49.240
Exco recently announced the expansion of its award-winning ad server to upgrade programmatic auctions in CTV and digital out-of-home environments. I'm a big digital out-of-home fan, by the way.

10
00:01:49.660 --> 00:02:01.160
Their solution empowers media owners to drive higher revenue through smarter automated ad auctions using a machine learning-based yield engine that dynamically optimizes auctions in real time.

11
00:02:01.500 --> 00:02:13.740
The translation of this is that basically it is taking what is fairly standard in the world of digital advertising and bringing it to the vibrant areas of CTV and digital out of home.

12
00:02:14.240 --> 00:02:23.660
So outdated programmatic pipes, fragmented technologies, and missed revenue opportunities that have plagued CTV and digital out-of-home media owners for far too long will go away.

13
00:02:24.160 --> 00:02:36.780
Try Exco's expanded ad server today to generate more revenue than you ever had before. Get in touch with Exco's video experts by visiting Exco. That is E-X dot C-O. Thanks so much, Exco.

14
00:02:37.360 --> 00:02:48.400
Now on to my conversation with Pete. [on-hold music] Well, Pete, thank you so much for, for joining me today.

15
00:02:48.540 --> 00:02:58.220
It was a busy weekend with this DeepSeek announcement, but I wanna get into, I wanna get into that a little bit, but-- and the overall sort of AI arms race.

16
00:02:58.240 --> 00:03:09.390
But really, you know, the idea of this for me was to get a look ahead at how AI is gonna impact overall the media industry in the, in the year to come, and we'll get into news, but beyond that.

17
00:03:10.020 --> 00:03:24.660
But this weekend, AI never sleeps, I guess, and a new Chinese AI model came out called DeepSeek, and it does a lot of the things these other models do. I played around with it a little bit. I read about it more.

18
00:03:24.780 --> 00:03:33.080
And, you know, to me, the, the big difference is how, is how much it costs to, to put together.

19
00:03:33.380 --> 00:03:41.220
At least the journal's reporting it had five point six million dollars versus the, you know, billion-plus that US models are making.

20
00:03:41.260 --> 00:03:49.549
I-- Which, by the way, is a little ironic because the tech people keep claiming that they're gonna eradicate wasteful spending from the government, and I start to wonder [chuckles]

21
00:03:50.100 --> 00:03:59.240
if, if what they're doing is a giant waste of money, then maybe they're not the people to do that, but that's for a different podcast. What, what is the significance of DeepSeek?

22
00:03:59.280 --> 00:04:14.320
It's, it's being painted, you know, very hastily as something of a Sputnik moment, and, and by that it, it meant that, like, you know, when the, when the Soviets came out with Sputnik, it was like, "Oh, no, they're gonna be able to compete technologically."

23
00:04:14.540 --> 00:04:21.399
And, and the Soviets actually did do a great job. Soviet engineering was amazing, uh, for what it could do within the constraints.

24
00:04:21.899 --> 00:04:36.140
And we're-- it seems like something a little bit similar here because the US, in its alliance with our tech oligarchs, has tried to control the transfer of, of technology and GPUs to, to China.

25
00:04:36.300 --> 00:04:46.920
So g-g-give me your, your, the, the, the assessment of, of the significance of this. Yeah. Well, first, thanks for having me. It's super thrilling to be here. Always love talking to you.

26
00:04:47.280 --> 00:04:53.580
So yeah, the DeepSeek thing, like you, I think you nailed it out of the gate in that the big factor here is, is cost.

27
00:04:53.920 --> 00:05:03.780
And what, what I try to do is I sort of think about this less in the sort of macro level, but we can get into that a little bit, and more on the sort of like what is the trickle-down effect of something- Mm-hmm...

28
00:05:03.790 --> 00:05:14.140
that is inherently cheaper just to run. And, you know, essentially, like, I even small, you know, small-time people like me feel the cost of AI sometimes.

29
00:05:14.240 --> 00:05:19.040
I try to automate a bunch of different things like social and whatever, and I have certain bots I've built.

30
00:05:19.140 --> 00:05:27.600
And sometimes in some of these platforms that I use, like, I notice my costs going up because of the model I pick, whether it's a four...

31
00:05:28.200 --> 00:05:37.770
GPT-4o versus the 4o Mini, and, you know, it's, it's, for a, a solopreneur like me, it adds, adds up to hundreds of dollars, not, not the tens of thousands.

32
00:05:37.820 --> 00:05:59.260
But you can sort of scale that for sort of publishers and, and media folks who are trying to, you know, essentially automate chunks of their operations, and it's like, "Oh, okay, so I need to use this other model that doesn't give me quite as good results," which might actually sort of place it below a threshold of quality, and therefore the AI thing is not viable.

33
00:05:59.820 --> 00:06:11.520
Now this creates a whole level of viability for those scaled projects. Potentially, if, if theyEnd up using it. So, so, so that's the, the most sort of immediate sort of trickle-down effect I see.

34
00:06:11.580 --> 00:06:16.130
Now, the other thing about this is that it's out of China, right? Like, and I kinda feel like- Yeah...

35
00:06:16.140 --> 00:06:26.040
well, didn't we just have a big row over TikTok and influence and having that power or a certain amount of ins-infrastructure? Now, it's obviously super early days for DeepSeek.

36
00:06:26.120 --> 00:06:31.300
I mean, uh, you know, just came out and no one's really, you know, using it in a, in a big way.

37
00:06:31.380 --> 00:06:53.170
But I feel like if it does become this thing where it's essentially as good as the current models and, and sort of keeps pace with whatever is, is coming out in an, uh, sort of open source commercial way, you know, like it's going to create some of these questions and influence and, and, uh, you know, is there

38
00:06:54.120 --> 00:07:01.880
some, some strings behind the scene, whether that's founded or not, you know, knowing that it's, you know, kind of the same kind of open source deal as Llama. It's, it's, yeah- Yeah...

39
00:07:01.900 --> 00:07:09.920
maybe those, a lot of those concerns are- I mean, it's, it's key that it's, it's, it's not open AI, which is not open. It is open source. Right. And so that would seem to get around...

40
00:07:10.040 --> 00:07:19.280
I know the-- there's already been some early, you know, whenever anyone gets access to one of these, these chatbots, they try to get it to like say something bad or whatever.

41
00:07:19.300 --> 00:07:35.120
And so if it's from China, you go and you ask it about Tiananmen Square or Xi Jinping and, and, and you see that it's surprise, it's not gonna return some sort of, you know, answers about the massacre there or about Xi Jinping's record with the Uyghurs or whatnot.

42
00:07:35.680 --> 00:07:48.940
But I would guess, you know, because it's just a model and it's open source, that the idea is anyone can take that and, and build, you know, a, a model that doesn't have those kind of, quote-unquote, "controls" on it.

43
00:07:48.980 --> 00:07:50.820
Now, I, I, I'm a realist.

44
00:07:50.980 --> 00:08:05.940
I mean, I, I've, I've been in the midst of this PR campaign from Silicon Valley that, you know, they, they all of a sudden have clo- have, you know, wrapped themselves in the American flag after being these massive multinational corporations that use tax havens and all kinds of things.

45
00:08:05.970 --> 00:08:16.780
Mm-hmm. But okay, whatever [chuckles], it's a business. And it puts a lot of pressure on them because, you know, they have been firing-- telling us that we need to fire back up Three Mile Island.

46
00:08:17.400 --> 00:08:26.930
And, you know, Sam Altman at one point was talking about raising seven trillion dollars, which I don't know where he came up with that figure, in order to, to get more data centers up and running.

47
00:08:26.940 --> 00:08:35.860
And this is just, at least in the initial reports that I've read, is just far more efficient. So I think the, the biggest impact right now is like, you know, the stock market is...

48
00:08:36.120 --> 00:08:39.270
D-do not look, do not check your 401[k], everyone- [chuckles]... I guess.

49
00:08:39.710 --> 00:08:52.260
'Cause the stock market is dominated by these, these tech companies and, and it's, it's calling into question whether or not all of this spending is, is, is really necessary.

50
00:08:52.310 --> 00:09:04.920
That's, you know, a tech story and a finance story. It doesn't really impact me. You know, the idea that the costs will come down is just axiomatic to all of technology.

51
00:09:05.240 --> 00:09:14.340
You know, all of these products are gonna be, you know, very gi-jury-rigged, uh, in the beginning, and they're gonna be incredibly expensive, and they're gonna be clunky, and, and we all dealt with that.

52
00:09:14.350 --> 00:09:25.380
Anyone who used the, uh, Nokia N95 phone can attest to [chuckles] that these things do get better and they get cheaper. So it- Mm-hmm... it will be interesting to see how this...

53
00:09:25.450 --> 00:09:32.749
'Cause it would seem to speed up the deployment of this technology, like you were saying, into real use cases.

54
00:09:32.860 --> 00:09:52.280
Because I guess sometimes with this, I struggle a little bit with the AI story because it lands in it could, it could, it might, it should, you know, range when we start to talk about actual real-life use cases that are not incremental, you know?

55
00:09:52.740 --> 00:10:07.140
And so where, where are we with that, right? Mm. Because to me, a lot of the, the general coverage of this, it falls into either academic. I don't care what AGI. I don't care.

56
00:10:07.340 --> 00:10:19.450
Like, I don't know why I should care about some nebulous line where it has crossed and whether it's PhD level or whether it just has like a Master of Science or whatever.

57
00:10:19.860 --> 00:10:30.200
Uh, what-- The only reason any of this stuff matters is if it has real impact on how we live and how we work. Like, that's all that matters. Right. This other stuff is nonsense.

58
00:10:30.240 --> 00:10:43.320
And the, you know, pouring billions or trillions into data centers, great. I, I saw this with Broadcom and Global Crossing with laying broadband, and there was a ton of money lit on fire then.

59
00:10:43.340 --> 00:10:55.200
There's a lot of companies who went out of business. I assume it's gonna be the exact same, right, right now. But like, where are we with real-life adoption of these technologies and the impact? Because whenever I...

60
00:10:55.340 --> 00:11:04.360
Like, we have these dinners, and I ask people, "In the media industry, what are you... Give me like what you're using AI for." And like I just hear like, "ChatGPT." Yeah.

61
00:11:05.060 --> 00:11:14.650
Yeah, it's funny, like here, you know, when you're reading about AI, when you're kinda living it, like maybe, uh, like certainly I am and you as well, like you're kind of in this bubble- Yeah...

62
00:11:14.660 --> 00:11:19.840
and you think, "Oh my God, everyone knows everything. It's, it's going so fast, I can't keep up."

63
00:11:20.480 --> 00:11:36.420
And then, you know, I, I-- a big part of my business is teaching newsrooms, PR teams, creative teams how to use AI, and then I sit down with these groups and they're-- it's like, oh, everyone's still just kinda messing around with a few prompts here and there, right?

64
00:11:36.480 --> 00:11:47.870
So to your point of like when is this like real? When has it actually arrived? I think it's when that part of my job kinda goes away, right? When- Uh-oh... when it's so inherent and so embedded in everything.

65
00:11:47.880 --> 00:11:56.640
Let's stretch it out a little, a little bit longer then, Pete. Yeah. Well, I hope so. You know, like I, I definitely see demand, so I, I think it's a ways out.

66
00:11:56.700 --> 00:12:11.020
Like, you know, so everything we hear about, you know, DeepSeek and Stargate and everything else, yeah, it's, it's all exciting and, and a little scary, but when it-- you get down to the, the general level, like my wife doesn't even really use ChatGPT at all.

67
00:12:11.480 --> 00:12:28.728
You know, it's not really-Affecting the normies quite yet. I think we're going to get closer, so there's, there's a whole layer of ChatGPT use that I think layered up or leveled up when they introduced search.

68
00:12:29.488 --> 00:12:39.848
And, but, you know, it always could access the web in, in sort of oblique ways. But when search came out, it's like, oh, you know, this can actually potentially replace your, your browser experience in some ways.

69
00:12:39.868 --> 00:12:54.138
And cer- certainly, it's an early adopter thing right now to do that and sort of change your default, but I feel like that's, like, the first kind of step towards some future where you're just talking to the AI, and it's just bringing you things.

70
00:12:54.208 --> 00:13:00.788
This is sort of like the agentic- Yeah... um, dream. So l- let's, let's break down the a- agentic. So it's, you know, I think...

71
00:13:00.848 --> 00:13:07.977
Look, I, I stepped outside of my apartment the other day, and I, I saw, like, a Salesforce, like, Agentforce billboard, and I'm like, "Oh, no, it's on now." Mm.

72
00:13:07.988 --> 00:13:15.818
When [chuckles] when Salesforce is rolling out the giant, the, the out-of-home campaign. You know it- Well, have you seen the quad in the airports? Have you seen those? Yeah [chuckles] you know- The quads...

73
00:13:15.828 --> 00:13:20.567
when you, when you see airport out of home, you know they're gonna, like, jam this stuff down our throats, like, all year.

74
00:13:20.628 --> 00:13:31.488
So yeah, we're gonna hear a ton, a ton about agents and agentic AI, and that's basically, I mean, my sort of definition of it is they go out and they do things for you. It's, it's the...

75
00:13:31.508 --> 00:13:39.808
At least this is in theory, right? Like, the idea is you're gonna... It's not gonna, like, point you to a bunch of links or tell you how to do something. It's gonna go and do something.

76
00:13:39.888 --> 00:13:48.528
It will book the flight, et cetera. Now, when these [chuckles] things start, it's gonna be so much... This is my prediction. Maybe I'm too cynical on a Monday morning, Pete.

77
00:13:48.868 --> 00:13:51.848
[chuckles] It's gonna be so much easier to just do it yourself.

78
00:13:51.888 --> 00:13:59.838
'Cause when people talk about it being, like, infinite interns, I'm like, anyone who's had, like, interns knows that's, like, sometimes, like, it's more work [chuckles] than, than actually- Yeah...

79
00:13:59.848 --> 00:14:07.108
and I'm like, "Infinite? Are we sure we want that? Can we just [chuckles] start a little bit lower?" Can we get infinite managers to- You know, 'cause I, I, I...

80
00:14:07.248 --> 00:14:16.578
This is, like, a little bit of an aside, but I can remember I went to... I wonder, I got out of J School in 2000, and I just got a note about our 25th anniversary, and I joked that- Hmm...

81
00:14:16.628 --> 00:14:17.848
I joked- I just, I just had mine.

82
00:14:18.028 --> 00:14:28.408
[chuckles] I joked, I joked to the woman they should have, like, those of us still sort of in the profession, or I guess I still am, uh, like, you know, bring us out like D-Day veterans, like the last [laughing] D-Day veterans.

83
00:14:29.248 --> 00:14:34.568
But I don't even know if I would qualify. But the people that are, like, still work or produce papers, let's bring them out, give them a round of applause.

84
00:14:35.108 --> 00:14:44.358
But I can remember then we had a Super Bowl party at the, for the 2000 Super Bowl, and the, like, the guy who was the dean of, like, the new media lab or something, [chuckles]

85
00:14:45.048 --> 00:14:52.188
he, he insisted on ordering, like, the Domino's Pizza, like, through the internet because you could. Mm. And guess what, Pete?

86
00:14:52.788 --> 00:15:01.157
We had a keg party with no pizza 'cause that goddamn pizza did not show up, and something similar [laughs]- It was always... Well, did the, did the progress bar on the Domino's just stop at a certain point?

87
00:15:01.157 --> 00:15:10.888
This is before that. This is before Chris and Porter Bogusky got involved. This, this was- Okay... like, you know, very, very early, and it was like, everyone was like, "Can you just pick up the phone?"

88
00:15:10.918 --> 00:15:24.248
[laughing] So anyway, I think agents are gonna be something similar, but at the same time, it, you know, everything starts like this, and just like, you know, e-commerce took off, it didn't stay with the- Mm...

89
00:15:24.268 --> 00:15:33.228
just pick up the phone to order the pizza, goddammit, it's gonna get going. But what... Give, uh, give... Like, what does this mean for, for publishers in general?

90
00:15:33.328 --> 00:15:46.028
Because to me, if you play this out, like, let's just assume that, that what is claimed for, for this agents actually eventually comes to fruition in a real way, not the hacked together way- Mm-hmm...

91
00:15:46.217 --> 00:15:56.128
in, like, a couple years. Like, to me, it's like that's kind of the end of websites, right? Mm. Because, like, websites are just, like... This was said to, uh, me at a dinner the other night.

92
00:15:56.188 --> 00:16:08.488
Like, they're just some kind of UI front end to a database at the end of the day- Mm-hmm... and it's just like calling or retrieving data, and the website itself is just kind of like a clunky intermediary at some point.

93
00:16:08.588 --> 00:16:19.088
But, like, what do you think the big impact, uh, of this agentic vision of AI is for publishers? Yeah. Well, honestly, it ties back to a little bit what we were just talking about, 'cause, you know, there was a...

94
00:16:19.468 --> 00:16:27.828
We were talking about, like, the Stargate stuff and the big picture stuff and all this, you know, compute power that we're gonna get and, uh, the powers and infrastructure, all that stuff.

95
00:16:28.228 --> 00:16:33.668
What that's gonna empower is w- a lot better inference time compute, right?

96
00:16:33.748 --> 00:16:49.548
Which is to say, like, when you actually search for something online, some information, it's actually gonna be able to process so much more so quickly that right now, like, AI search as it exists is a little bit kind of like the baby version.

97
00:16:49.588 --> 00:16:58.908
Like, it seems like it's going out and, and reading all of these articles in real time and, and doing it, and it's not really. Like, there's a lot of schema and a lot of metadata that's, that's still involved in that.

98
00:16:59.508 --> 00:17:07.188
But once we get there and, and you, you... It can actually do this, that is 100% when, like, websites are kind of obsolete. You know?

99
00:17:07.248 --> 00:17:17.298
It's like, okay, now, like, as long as we work out the, the monetization [chuckles] and the, the, you know, the- It's a big asterisk... the licensing. That's... Yeah, it's a big asterisk. But if that...

100
00:17:17.688 --> 00:17:21.728
You know, the models are kind of emerging and, and you could see some- someone running with the ball with that.

101
00:17:22.268 --> 00:17:34.428
Then everything just kind of goes into this blob of information as soon as it's published, and then, you know, as soon as someone asks for something that that fits the pattern of, it's, it's gonna get summarized, and, you know, they can go and, and go deeper at some point.

102
00:17:34.488 --> 00:17:47.088
So that's a bit... Uh, obviously, it's a bit of a ways off, but I think that that is the, the train we're on. Uh- Do you see anything, like, that you would point people to that you can start to see this, like, happening?

103
00:17:47.118 --> 00:18:00.368
'Cause I think one of, one of the big questions is how much brand really matters. And the- Mm... you know, I think a lot of times, particularly in journalism, you know, trust, the trust flag comes out, right?

104
00:18:00.568 --> 00:18:12.468
And it's just used as kind of like, I don't know, gauze to some, some degree, I feel like, s- because it's never backed up with actual, you know, data or, or evidence. It...

105
00:18:12.508 --> 00:18:22.316
There's this just-Kind of like religious belief that, you know, our, our audience trusts us, they're loyal and everything. And, and I'm like, I don't know.

106
00:18:22.356 --> 00:18:29.296
I mean, the, the data I've seen is that people do not really, for the most part, have a primary news source.

107
00:18:29.855 --> 00:18:58.496
And the obvious-- to me, like the obvious sort of mainstream sort of news experience is that it's very personalized, it's very tuned to the, to the person's sort of interests, maybe to their proclivities and ideological biases or just beliefs, and that it is delivered in the format that the person, the user or the audience member, like consumer, I guess, prefers.

108
00:18:58.716 --> 00:19:08.876
The idea that it's gonna be packaged by a bunch of, you know, newsrooms seems, I don't know, I don't know if it's arrogant, it just seems like sort of outdated. Like- Right...

109
00:19:08.896 --> 00:19:20.696
okay, there's some people who want the, the packaged version, but there's some people who are gonna, you know, package it themselves, right? Like, I mean, that, that seems obvious or, or just like an obvious endpoint.

110
00:19:21.176 --> 00:19:26.846
Yeah, I think. And I don't know, what, what is... what is your sort of take on that? Is that sort of how you see things going?

111
00:19:26.896 --> 00:19:39.216
Because I, I just like wonder, uh, do you see things out there that point towards this, this agentic future for, for news? Yeah. So I'd point to two different trends here, which are kind of...

112
00:19:39.296 --> 00:19:48.056
one is they're almost like the opposite of each other, but they're actually, I think, different ingredients of, of what this agentic future might be.

113
00:19:48.176 --> 00:19:58.756
So one is just all the deals that we're seeing with all, you know, OpenAI and others, mostly OpenAI right now with publishers, and it's always kinda like the same group, but it is sort of happening, right?

114
00:19:58.856 --> 00:20:07.836
So it's like, you know, whether it's News Corp and everyone else and Reuters, et cetera, you know, we're starting to see that with Meta and Microsoft too now, they're, they're making these deals.

115
00:20:07.896 --> 00:20:26.896
So what this sort of, you know, portends to me is like, oh, there is a model emerging, but in the initial, in the initial sort of phase of it, there's gonna be a certain sameness to what you get that is news on a lot of these AI interfaces, and I think there's this counter trend that people don't want that, you know?

116
00:20:26.926 --> 00:20:44.856
And this is why, like Substack and, and individual creators, while, while that, while that sort of summary of sort of what's going on is helpful, it's like what I really, what I really want is the, the news and, and points of view from the people that, that I want, that, that, that I'm following, right?

117
00:20:45.316 --> 00:20:56.336
Yeah. So the other trend, this is really counterintuitive, but I would point to the popularity of NotebookLM. And what that is to me is- Do you wanna explain NotebookLM for the- Yeah.

118
00:20:56.356 --> 00:21:03.526
So, so Google NotebookLM, it's actually been out for a while, almost a year, I think, when Google first released it, and it's essentially like a folder.

119
00:21:03.556 --> 00:21:14.716
And the way I sort of talk about is like with ChatGPT, with, with your interaction with AI prioritizes the chat. I mean, you can attach files to it, your chat, that's fine.

120
00:21:14.936 --> 00:21:23.476
What NotebookLM prioritizes is the files, like the, the data you're giving it, and the chat and the AI interaction is sort of secondary.

121
00:21:23.506 --> 00:21:40.876
It sort of flips the script on the chatbot, which it, it was a super brilliant innovation because what that empowered and what really put it on the map is the idea you can take that data and re-version it into whatever you want, and their, their big product was podcasts.

122
00:21:40.886 --> 00:21:49.396
"Hey, let's have a couple- Yeah... of chatty people talk about the data we have here." And so- And they're very human sounding, by the way. I've talked about NotebookLM. They did a great job. They're... they, they...

123
00:21:49.456 --> 00:21:59.466
I mean, they [laughs] I, I've mentioned before, I have this like thing, I, I like keep like, like a journal, I guess, of like, of the stuff I'm thinking about with the business and beyond, but I do it like every day.

124
00:21:59.556 --> 00:22:08.916
I... and, and, you know, I'm a writer, so like I, I have a lot of data, like [laughs] over the last couple years. It's very strange to listen to a podcast that talk, talking about you. [laughs] Yeah.

125
00:22:08.936 --> 00:22:16.186
[laughs] I couldn't do it. I haven't done that yet. I've thought about it. I think I, I inevitably will. I've got a bunch of notebooks though. They're very nice. Uh, yeah.

126
00:22:16.216 --> 00:22:22.456
I mean, AI, I will say this, AI engine, they have tuned them to be suck-ups. They're all suck-ups. They're never gonna tell you- Mm. They're...

127
00:22:22.466 --> 00:22:32.695
So I think that's the problem with the, the sort of truth thing and trust thing because, like they're too, they're too positive. Like when I- Yeah... ask like, you know, "Oh, can I make, uh, this dish with this, this?"

128
00:22:32.796 --> 00:22:39.396
They're never like, "No, that's terrible. Like, that's a terrible idea." [laughs] There's actually like a few actual podcasts you can get in the...

129
00:22:39.616 --> 00:22:50.666
probably more than a few, in like the Apple Podcasts app that are just NotebookLM feeds, and if you listen to a bunch of these together, like, like a few, you, you start- Yeah... to really feel the weakness of this.

130
00:22:50.716 --> 00:22:56.446
It's like, oh, it's really cool that- Yeah... they're conversational, but then you're like- Well, it's a parlor trick at- Yeah... some level. I mean, even the Perplexity.

131
00:22:56.516 --> 00:23:06.606
Perplexity has an AI, like, you know, summary news pod- podcast product, and it's kinda dull if you ask me. Uh, but- Mm...

132
00:23:06.636 --> 00:23:19.986
look, I think a lot of this points to, and this is where I wanna get at with, with how y- you see newsrooms and publishers using th- this technology because, look, it's good at summarization- Mm...

133
00:23:19.996 --> 00:23:23.306
and it's good at versioning, and to me... and it's good at pattern matching.

134
00:23:23.396 --> 00:23:33.276
So if you look at the professions, the areas that have been affected the most, obviously developers, you know, that's the sort of home run use case. It has not replaced the writers.

135
00:23:33.516 --> 00:23:42.636
[laughs] Weirdly, it is, it, it is- Right. Yeah... you know, putting pressure, downward pressure on, on developer jobs. It's good at coding 'cause it's good at pattern matching, right? Mm.

136
00:23:42.776 --> 00:23:55.056
But when it comes to, you know, content, speaking broadly, it's, it's good at summarization. Like, I mean, I- Yeah... after, after we do this podcast, I'll download a transcript and I'll ask it to summarize it.

137
00:23:55.196 --> 00:24:06.116
It will not write it for me. I think a lot of times people get to the output, but like honestly, like my own summarization skills as I get older and I do more things like are, are, are not as good as, as they used to be.

138
00:24:06.216 --> 00:24:15.796
And so it will catch things that I would probably not remember or have to go back, and so that saves me, I don't know, like, you know, it's like 15%, 20% of time.

139
00:24:16.376 --> 00:24:21.936
And a lot of, a lot of content that is created is some form of summarization at the end of the day.

140
00:24:22.396 --> 00:24:35.046
Or the existing product, i- i- you know, the, the atomic unit of news, which is still like the sort of article for, for many publishers, you know, can easily- Mm... be summarized. That's why...

141
00:24:35.096 --> 00:24:45.347
that's what I see a lot of publishers, they're pointing at it to, to do the three bullet points.Which isn't exactly inspiring [laughs] Yeah, it's like everyone's using it to kind of do Axios almost, right? Yeah.

142
00:24:45.358 --> 00:24:55.088
Where- whereas Axios kind of like almost reduced the atomic unit to something arguably even smaller than, than an article. Yeah. And- Yeah, 400 words like and bullet points in the, in the bulk.

143
00:24:55.688 --> 00:25:12.568
Yeah, but what I, I think NotebookLM suggests to me is that the, the, the, the innovation of doing it with the data and then doing this, whether it's a podcast or w- you know, whatever your preferred way to consume that data or get an overview of that data is, the data can be whatever you want, you know?

144
00:25:12.608 --> 00:25:15.988
So what if it's like all the things that you actually do care about?

145
00:25:16.028 --> 00:25:26.548
Like, basically bridging the idea of like, oh, all the things you follow, somehow putting that in a sort of NotebookLM idea and then just getting it summarized for you.

146
00:25:26.588 --> 00:25:33.548
And that's, that's the agentic future where you have these sort of news agents on the consumer side of this going out, like- Yeah...

147
00:25:33.568 --> 00:25:41.288
give me all the stuff that I care about, summarize it all together, and give me that, that daily brief, uh, however it is. And we're kind of there, right?

148
00:25:41.348 --> 00:25:50.528
Like, I mean, uh, uh, OpenAI just came out with Operator, which is their computer use thing. So it's essentially it can take over your computer or, or even- Yeah...

149
00:25:50.548 --> 00:26:04.848
work in the background and just get stuff for you, whether it's, you know, ordering the pizza or the flight or, hey, summarize all these stories on newspapers or whatever, the Times and, and the Journal and, you know- Well, you can like set it.

150
00:26:04.988 --> 00:26:15.838
You can say, "Hey, every morning I wanna have like a summary of, you know, this topic, this topic and this topic. Use these, these sources, et cetera- Right... or, or just leave it up to them."

151
00:26:15.928 --> 00:26:26.068
And I can see that being a very compelling use case for most... for not most, for a, a large chunk of consumers.

152
00:26:26.128 --> 00:26:37.828
I think sometimes those in the news business that are, particularly those that are extremely online and are constantly consuming news, don't recognize that it is simply not a major part of most people's lives.

153
00:26:37.888 --> 00:26:46.488
You know, they would... we would call these people like low information voters, whatever. Guess what? [laughs] They're the majority. You know, I... and in some ways it's kind of sensible.

154
00:26:46.668 --> 00:26:56.568
You know, the news product itself is, to me, it's not in a great shape. You know? Mm-hmm. Yeah. The user experience is, is, is horrendous on a lot of these sites.

155
00:26:56.668 --> 00:27:07.988
A lot of, you know, these, these stories are like too narrative driven for, for many people. I don't think a lot of people want the, your storytelling. Mm. A lot of people just wanna know what's going on in the world.

156
00:27:08.088 --> 00:27:21.658
And real or perceived, there's, there's, there has been a, a lack of, of trust in some of the bias that many people believe has, has cropped into the, the news product itself. So I wonder- Yeah...

157
00:27:21.708 --> 00:27:30.508
how AI can be used and, and specific examples you see to improve the, the, the product, right? Like, so- Mm...

158
00:27:30.718 --> 00:27:44.348
I, I talk a lot about how publishers are downstream of tech, and I guess what I see out there, and maybe it's, maybe, uh, I'm not seeing a-as much as I should, I don't see publishers sort of looking to innovate.

159
00:27:44.988 --> 00:27:57.248
You know, putting a chatbot on your site, to me- Yeah... is like, okay, great. Like, I mean, honestly, like who's using them? Mm. When you start to ask for usage, you know, then, then the conversation gets changed. Yeah.

160
00:27:57.298 --> 00:28:06.768
Yeah. Because I... to me, the obvious thing is like, yes, it's very good at summarization and piecing things together, but that just points to multi-source. Like- Mm-hmm...

161
00:28:06.908 --> 00:28:17.598
outside of a few use cases, particularly like around finance publishers, you know, a Bloomberg or something that has unique data sets, I honestly don't see why I would...

162
00:28:17.648 --> 00:28:27.397
if I wanted to have a summary every morning of Trump executive orders, I don't understand why I would want it just from The Washington Post. Like, why wouldn't- Yeah. Right...

163
00:28:27.397 --> 00:28:38.908
I want Politico and all these other sources, Punchbowl, et cetera? Yeah, and that's kind of where I don't know if that user experience innovation can even be achieved at the publisher level.

164
00:28:38.918 --> 00:28:45.848
Like, yes, you're right, you're gonna, you're gonna see these chatbots. Some are gonna be better than others. Some summarization's gonna be better than others.

165
00:28:45.888 --> 00:28:57.358
But there's a couple of barriers to, to it being like this great experience. One is, just like you said, the, the availability of content. You know, you kind of want a more, a broader view from, from multiple sources.

166
00:28:57.908 --> 00:29:08.008
Another is that publishers are rightly very concerned about accuracy, and there is a still... like, uh, hallucinations are just inherent to the technology.

167
00:29:08.408 --> 00:29:24.668
And yes, you can reduce them, and yes, there are, you know, safeguards you can put in place, but generally those safeguards, on the publisher side anyway, have been so restrictive that, you know, the Washington Post chatbot just won't say anything if you go anywhere near like outside of the, the guardrails it has.

168
00:29:25.148 --> 00:29:39.088
Times AI, which I actually like, I actually like, like their UX and, and how they've presented on Person of the Year, but the thing is, those summaries that they did of the Person of the Year, you could tell th-they weren't generative.

169
00:29:39.228 --> 00:29:46.268
Or they... in other words, they were generated, yes, probably by AI, but they were also vetted by an editor, 'cause every time you would go to the- Yeah...

170
00:29:46.348 --> 00:29:54.068
small or medium version, it was exactly the same word for word, which says, oh, it's not actually doing this on the fly, which means it can't scale.

171
00:29:54.128 --> 00:30:03.968
You know, you couldn't do it in the way that we're talking about where it's totally tailored to whatever the user is doing at any given moment. Now, tech companies are much [laughs]

172
00:30:03.978 --> 00:30:11.958
, well, shall we say, less concerned with accuracy- More letter rip... unless they're bullied into it- Letter rip... like, like Apple was. [laughs] You know, Apple was bullied into dialing back- Well-...

173
00:30:11.968 --> 00:30:15.508
their notifications because of the BBC... yeah, but like also, like that's a great example.

174
00:30:15.608 --> 00:30:26.628
You know, Apple was using these AI summarizations of news and it was screwing it up and, you know, they just like kind of rolled it back. They didn't get like a ton of like blowback that I could see. Like- No...

175
00:30:26.638 --> 00:30:36.648
I mean, like things just kept, kept going on. Like, if like The New York Times did that, it, there would be like a million VCs with pitchforks that were blaming DEI or something for it.

176
00:30:36.658 --> 00:30:42.820
[laughs] So it's like you can't win at some point if you're a publisher.Yeah. So, uh, there's kind of a level of...

177
00:30:42.840 --> 00:31:01.700
And also, like, the, the idea of hallucinations, the fact that it's not a human doing it, like it's, it's, it's similar to the, the self-driving car problem where even if a self-driving car is demonstrably safer than humans driving cars, and I don't know if we have the data on that, but that's the supposition, right?

178
00:31:02.500 --> 00:31:13.700
The fact that a self-driving car even kills one human, you know, even if the aggregate is lower, because of this sort of phase we're in and it's new, it, it's, it's much more serious.

179
00:31:13.740 --> 00:31:24.420
And this is part of the reason, like, Uber dropped their self-driving after the accident, I forget where it was. But so, so I don't know if, like, hallucin- like the-th-it's an analogy. Like, you get- Hmm...

180
00:31:24.440 --> 00:31:36.700
what I'm getting at is if, if a publisher screws up with hallucinations, again, even if it... the error rate might be lower than humans, the fact that it's this AI doing it and there's sort of this mistrust, I don't...

181
00:31:36.740 --> 00:31:45.140
Again, I don't know if it's ever gonna be able to cross that threshold of being accurate enough that they're gonna apply it at scale and be confident. Yeah. You know?

182
00:31:45.239 --> 00:31:51.520
That's, that's kind of the thing that might hold back, like, them being able to sort of control their own destiny.

183
00:31:51.550 --> 00:32:02.760
'Cause like I said, the platforms will apply this at scale, and the next one that I think will be really interesting is Amazon because- Oh, really? Yeah. Is that different?

184
00:32:02.840 --> 00:32:13.160
Well, it's, uh, because Alexa is such a bad product compared to ChatGPT right now. Oh, it's terrible. And wh- again, it... I'm not sure what they're gonna release.

185
00:32:13.180 --> 00:32:22.530
They're obviously gonna release an AI version of Alexa at some point. Maybe it's, it's gonna be, like, a big question, is it gonna be more like Apple, where it's very underwhelming and doesn't really do much? Yeah.

186
00:32:22.540 --> 00:32:35.740
Or are they actually gonna go for it? Are they gonna go for, like, a real ChatGPT interface? And if they do do that, I could definitely see generative summaries of the stuff you care about in a conversational way.

187
00:32:35.820 --> 00:32:44.140
Like, like, that could suddenly become a real thing people are doing. 'Cause a lot of us still have these microphones all over our house. Yeah. Like I do. I, I've never...

188
00:32:44.200 --> 00:32:49.940
It's funny 'cause, like, I remember wh- when they came out with that, we were doing, like, flash briefings at the time for... Yeah.

189
00:32:49.949 --> 00:32:57.680
And the idea was, like, people were gonna be ordering their groceries off these things and, like, I think they're just sitting there and are, are just, like, speakers- Giving us the weather... at, at this. [laughs] Yeah.

190
00:32:57.720 --> 00:33:07.040
It's like parlor trick. You know, my parents are always, like, you know, asking, like, Google to, to tell them what the weather's gonna be. But, like, beyond that, it just seems like- Yeah...

191
00:33:07.080 --> 00:33:17.180
kind of an underwhelming product. And then we just sort of move on and forget about all [laughs] the, the hype that it was gonna change everything. But with- Mm... with which, which publishers do you think...

192
00:33:17.240 --> 00:33:30.840
Like, give me examples of publishers that are actually using AI to create compelling products at the end of the day. Well, products. Well, the Post and the Times are both using it- They can be features.

193
00:33:30.900 --> 00:33:39.950
I'll even take a feature. [laughs] I think the Post and the Times are doing some good stuff with investigative. That's not quite a product, but being able to process- Sure... a lot of data and do- Yeah...

194
00:33:39.950 --> 00:33:53.140
a better investigation, that's good. Again, it's, like I mentioned, Time AI, I think, is, is a good experience because it makes it a nice sort of Chipotle-style menu of what you're getting, as opposed to this chatbot.

195
00:33:53.200 --> 00:34:03.200
I don't think chat... Even as much as I'm a pro-chat person, I don't think it is the thing that's gonna turn, you know, publisher experiences into something amazing.

196
00:34:03.800 --> 00:34:10.900
I do think it is helpful to have chat so you know how to play in AI and get some good user data on sort of what- Hmm...

197
00:34:10.970 --> 00:34:20.310
people who are actually coming to your website care about, and, and that'll help you sort of play in these bigger summarization engines. But I, I agree, it's not, it's not the best product for everything.

198
00:34:20.989 --> 00:34:30.680
But that said, I do think a certain amount of... I don't know if anyone's really doing this, but it-it's, like, chat on the article level, you know?

199
00:34:30.700 --> 00:34:43.400
Where it's like, oh, like, what are the places I can go deeper and get sort of follow-up questions about, you know? Like, like, that's... that seems like the place for chat for me. You know?

200
00:34:43.430 --> 00:34:52.720
It's like, oh, okay, I've, I've read this article and I actually wanna go a little deeper on it, and rather than going back to Google, maybe there's something here that points me somewhere. I'll, I'll give sort of a...

201
00:34:52.740 --> 00:35:01.140
Can I give an anti-good product? So yeah, I guess- Yeah... which would be- Let's do it. So there was a lot of hype around Particle- Oh, yeah... which was this news app. I was just looking at that. Yeah, yeah.

202
00:35:01.320 --> 00:35:08.640
I mean- I mean, that's, like, a former Twitter executive started this sort of AI news summer- summarization app.

203
00:35:08.740 --> 00:35:16.080
I mean, news aggregation apps, I don't know, there's, there's a graveyard somewhere that's, like, filled with, like, Circa and all, all of them.

204
00:35:16.200 --> 00:35:27.630
They, for whatever reason, they, they make a ton of sense, but they've never really taken off outside of existing aggregators just putting one. Like, if you have...

205
00:35:27.790 --> 00:35:39.140
If, if you're already aggregating, if you're one of the choke points of the internet, if you're, you know, Google with Google News, okay. You know, Apple, Apple News is, like, has emerged as just a massive force.

206
00:35:39.180 --> 00:35:53.020
But that's, you know, that's just the second-order impact of controlling distribution. Hmm. And Particle doesn't have that. Right. Yeah. And I find... I just find, as a design, like, it's, it's really busy, you know?

207
00:35:53.080 --> 00:36:01.170
Like, it's almost like the design is anti-AI in the sense that AI was supposed to simplify things and summarize things- Yeah...

208
00:36:01.170 --> 00:36:10.370
and I tap into a Particle summary, and it just has, like, a bunch of different fonts and colors, and I'm, I'm like, well, what's the path to the story that I actually wanna read about in this summary?

209
00:36:10.420 --> 00:36:16.800
And I can't find it [laughs] like, every time I look, like- It's very branded... I just go back to Google. I guess. Yeah. Like, I would say it's very branded rather than just- Hmm... like, bare bones.

210
00:36:16.820 --> 00:36:32.910
And you look at, like, what is... I mean, even the sort of positioning of a lot of these AI products is, is very... I mean, it's very... Like, ChatGPT is, like, incredibly bare bones, right? And, like, it's also- Hmm...

211
00:36:32.940 --> 00:36:42.372
you know, I don't know if this is intentional or not. Like, they don't abstract the, like, the, the speeds and feeds part. That's what we used to, like, call it, like, in tech.... coverage.

212
00:36:42.512 --> 00:36:57.952
But like you know, it's like I gotta choose which model and like, you know, you go and you're just like, I, I can understand like to me it's gonna hold back adoption. Like I mean convenient always wins- Mm...

213
00:36:57.962 --> 00:37:11.582
and, and simple always wins, and the idea that I need to choose which, which model I'm gonna use, I mean that's like CB radio land. Like, that's not like, you know, people don't wanna do that. I think I feel- Right...

214
00:37:11.612 --> 00:37:24.352
fairly confident in speaking for humanity. [laughs] Oh, totally. Yeah, yeah, yeah. Yeah, that's not gonna work. But yeah, like features, you know, I mean, the spoken articles thing was kind of a, a level one- Yeah...

215
00:37:24.482 --> 00:37:37.912
of what's happening, and again, I'm not sure how comfortable publishers will be at doing this, but like in applying like a NotebookLM style to a batch of coverage, I think that's sort of the logical next step- Yeah...

216
00:37:37.921 --> 00:37:46.372
of like I don't wanna hear just like someone speaking this article to me, I want like a conversation and, and sort of a summary and overview of that. I think that's kind of inevitably gonna be coming.

217
00:37:46.872 --> 00:37:55.932
It's just a matter of like can they scale it? Will it, will it, will it be- Yeah... something that, that just becomes a, a standard way we consume something. Yeah.

218
00:37:55.942 --> 00:38:06.292
And I think one of the, the, the big, one of the big outstanding questions, right, is how AI is gonna drive productivity, right? And you can look at productivity two ways, [laughs] right?

219
00:38:06.612 --> 00:38:13.912
One, you need productivity to drive economic growth. It is just how... Labor productivity is how you drive economic growth. So you're not against...

220
00:38:14.152 --> 00:38:18.792
If you're a de-growther, you can be anti-productivity, but I think most people are not de-growthers.

221
00:38:19.372 --> 00:38:27.482
And on the other hand, there's a, a lot of unease because, you know, driving labor productivity means, you know, fewer overall jobs.

222
00:38:27.532 --> 00:38:48.032
That's generally why, you know, historically, you know, for instance, like unions have tried to hold back automation and all kinds of different technological advances that, you know, from a, on an economic level drive, drive productivity, so that's a good, net, net good because g- it means that you can deploy those labor inputs to more productive endeavors.

223
00:38:48.052 --> 00:38:57.252
At least that's the theory, right? But generally, you know, people do not like change when it is visited upon them. They usually like change- Mm... when it, when it affects other people, I've [laughs] noticed.

224
00:38:57.292 --> 00:39:02.972
It's kind of like accountability. It's much better for other people than ourselves. Who is like... So how do...

225
00:39:03.032 --> 00:39:11.062
Like, obviously, news organizations in particular, but publishers overall are under pressure to do more with less. I mean, we've seen like cuts like nonstop, right?

226
00:39:11.112 --> 00:39:26.672
And you know, AI has, you know, at least the theoretical, you know, benefit of doing just that. I mean, I think I would say for me it is like a 10 to 15% these various tools added up together increase- Sounds right...

227
00:39:26.772 --> 00:39:37.462
in, in productivity. It's not like... I think a lot of times, again, like I say, people go to like the writing of the stories, they wanna just hit a button and things like happen and like- Right... I don't...

228
00:39:37.512 --> 00:39:44.012
Maybe I just haven't found that button or just- [laughs]... been able to hook things together, but that ain't the case. Like it will... It, it is...

229
00:39:44.152 --> 00:39:57.612
It speeds the, you know, the like the production of this podcast, right? Mm. Like that it absolutely, you know, helps do that. It helps with, you know, if you use Grammarly for like copy editing. So- Yeah...

230
00:39:57.672 --> 00:40:00.312
who ends up- These sort of subtle things... I mean, you've written about this before.

231
00:40:00.712 --> 00:40:14.992
It would seem like all of these things, 'cause we're seeing already like this exists within the context of, you know, organizations are, are slimming down, particularly in the, in the middle management layers, right? Mm.

232
00:40:15.302 --> 00:40:26.152
And you don't have as much of, of that in, in most publishing organizations just because, you know, the business is, is not such that you can run it like Google for sure. [laughs] Right.

233
00:40:26.992 --> 00:40:37.152
But like it would seem like this would empower the people who are making the content to do more, right? To be able to, to, to be more productive, right? Yeah.

234
00:40:37.212 --> 00:40:51.292
But that some of those coordination, I saw it like described as glue jobs, will end up being affected quite a bit by, by AI tools and technology.

235
00:40:51.832 --> 00:41:01.512
Glue jobs like asse- assistant editors, copy editors- Yeah... that sort of stuff So like, you know- Yeah... you're moving data around, you're doing some kinda coordination role.

236
00:41:01.592 --> 00:41:09.842
Obviously, anyone who's been inside a news organization can tell... Like you know, the people actually making the content- Yeah... are a minority in, in these companies, right? Yeah.

237
00:41:09.872 --> 00:41:22.312
And that's because there are a ton of coordination roles. If you're gonna take subscriptions, you're gonna have a saves team. You're gonna- Yeah... have a growth team. You're gonna have all of these, and, and it adds up.

238
00:41:22.352 --> 00:41:32.252
It's a lot of people to run a publishing business. I think that's what the sort of reaction to that is, you know, in the area we're both broadly in, right? Mm.

239
00:41:32.272 --> 00:41:40.012
Like it's like you s- you slim it down to like the bare essence [laughs] and you know, you can, you can...

240
00:41:40.032 --> 00:41:47.532
You might not be able to do as much as, you know, people with 75, 100, 200 people, but like you can cover a lot of ground these days. Yeah.

241
00:41:47.872 --> 00:42:00.672
Well, I also think in, in a lot of places there's going to be sort of a, a sel- well, not a select few, but people l- in the newsroom who are really embracing AI tools are kinda gonna build their own glue.

242
00:42:00.712 --> 00:42:08.752
So I know a bunch of people who are in newsrooms who are kind of... They're not coders, but they're maybe AI productivity enthusiasts, right? Yeah. And they...

243
00:42:09.272 --> 00:42:25.312
What, what I find interesting is that it's almost this merger of editorial and product now, and I think that's kind of a reality going forward, that those two things are going to be much closer and in some roles almost a dual role.

244
00:42:25.352 --> 00:42:32.232
Like I know a bunch of editors now that spend a lot of their time building tools and, and maintaining those tools.

245
00:42:32.292 --> 00:42:43.652
So f- this one publication that it's more regional and a lot of the reporters who are covering various things like the courts and, and recalls and whatever are...

246
00:42:43.892 --> 00:42:59.152
used to spend a good chunk of their time in the morning like re-You know, figuring out what's going on in all these things and, you know, go to the right government website, et cetera, and, like, what's changed, what piece has been updated and now they, alongside some product stuff, have, have built these tools.

247
00:42:59.212 --> 00:43:01.132
Again, like, no code tools. They don't know code.

248
00:43:01.192 --> 00:43:14.542
They're just taking some off-the-shelf platform, plugging in some generative stuff and doing it in sort of a left brain kind of way, and now that's, that's done for them, right? And, and again, like, not, not...

249
00:43:14.692 --> 00:43:20.052
uh, no- normally, like, you know, a few years ago you'd get your product team to do it if it even fits on their roadmap.

250
00:43:20.112 --> 00:43:29.812
You're like, "Yeah, we'll do that someday," maybe it's six months, a year down the road, et cetera, and now there's a lot more sort of empowerment of, of just sort of doing that yourself.

251
00:43:29.892 --> 00:43:36.352
So, so that's kind of a neat development. It does mean for certain roles you're gonna be, like, you...

252
00:43:36.592 --> 00:43:49.472
that, that ten percent efficiency isn't necessarily getting realized in the, the story building, it's more in sort of the, the infrastructure building of, of the place, but I think it sort of generally levels up- Mm-hmm...

253
00:43:49.532 --> 00:44:01.232
of everybody. But yeah, I keep thinking the big question is, like, what do you do with that 10 or 15%, right? Like, you know, in, in roles I've had, you know, do, do you want the people to go deeper on certain things?

254
00:44:01.292 --> 00:44:11.412
Do you want more stories? Like, what's gonna have the most impact to, to the publication? And, like, in sort of in t- today's environment, I think it's, like, going deeper, you know?

255
00:44:11.492 --> 00:44:18.522
Like, it is, like, you don't wanna necessarily want more stuff out there- It depends... you want the better stuff for your audience. So yeah, it kind of- It depends...

256
00:44:18.522 --> 00:44:36.552
depends on what you're doing, but- You know, like, I mean, for instance, like, I'm doing an online forum later this week with Newsweek and, you know, they've used AI and other, you know, tools to double their output basically with the same, with the same amount of staff and, you know, that, like, you...

257
00:44:36.612 --> 00:44:46.892
I gu- I think, you know, I'm gonna talk to, to them about it with the chief product officer. But, you know, the reality for a lot of these models is you need... you... they still need to, to...

258
00:44:47.152 --> 00:44:51.112
they still need traffic- Mm-hmm... in their models. Sure.

259
00:44:51.132 --> 00:45:05.892
Like, and it's easy to say, "Ah, it's about depth, not breadth and stuff," but the reality is, like most businesses, you're operating your existing business and you're trying to build the next business, and you can't, you can't just put the existing business on pause.

260
00:45:06.162 --> 00:45:18.032
Totally. You're gonna have to... So I, I have a lot of sympathy because there's a lot of pressures in, in the marketplace, obviously, and, you know, it depends on your model, but a lot of models still need to...

261
00:45:18.072 --> 00:45:21.422
efficient ways to, to create, to create traffic. Mm-hmm.

262
00:45:21.452 --> 00:45:40.312
So which, which leads me to the search thing, and I wanna just, you know, close with a little bit on that because, you know, obviously the big concern for, for publishers is, is the change to search and, you know, I think sometimes it can be both underplayed and overplayed.

263
00:45:40.852 --> 00:45:49.682
Underplayed in that the search is so central to many, if not most, publishing models.

264
00:45:49.812 --> 00:46:01.232
It has been the, not just the dominant, but the most stable algorithmic distribution source for, for, for audience, and, uh, oftentimes very high quality, right?

265
00:46:01.352 --> 00:46:10.052
Compared to a lot of the kind of nonsense audience that you, that publishers would get from- Yeah... from like Facebook. Its, its death has been exaggerated, uh- Right, because it is still...

266
00:46:10.652 --> 00:46:20.312
Google, I saw one, you know... the stats are kind of difficult to get in there, but I saw one saying, "Oh, Google f- finally dipped below 80% share for the first time in, like, nine years." Yeah.

267
00:46:20.832 --> 00:46:27.492
You know, and it was, like, 79.9 versus... and it's like, okay, that seems- Yeah... pretty dominant. [laughs] I think it was actually 90. You know, it was like, you know, there's...

268
00:46:27.552 --> 00:46:31.892
it's, it's just so dominant, like, it's like, okay. You know, so we- we're...

269
00:46:31.932 --> 00:46:50.632
I mean, like, we talk, we use, like, Perplexity in these kind of things, but, you know, regular people are not, uh, to, for the most part, I mean, they are, yes, but, like, in the broad sweep of how dominant search is, this is, this is chipping away and- I also tried to switch to an AI search as kind of my default.

270
00:46:50.692 --> 00:47:01.672
I couldn't do it, and I think it's 'cause if you're extremely online and you, you've just developed this rhythm with Google as, like, you know, firing up a new tab and getting links and- Oh, it's everywhere...

271
00:47:01.732 --> 00:47:09.392
reading certain terms. And so we'll see if the... we'll see how, how the Trump administration, uh, deals with some of the leftover antitrust cases. Oh, yeah. True.

272
00:47:10.072 --> 00:47:18.632
But it is kind of a, just a thing in terms of what I want when I search the internet or want something from the internet. Most of the time I'm...

273
00:47:18.672 --> 00:47:29.171
that processing that AI does, I can actually still do faster with a, a re- search results page with just my head. Like, oh, here's the link in... that I was actually looking for in the first two.

274
00:47:29.212 --> 00:47:35.432
It ki- kind of depends if you're looking for information or something very specific like a destination. Yeah. But when, I mean, the...

275
00:47:35.472 --> 00:47:42.532
I think the big sort of concern that I hear all the time is, you know, search has gotten less reliable. Sure. If AI...

276
00:47:42.892 --> 00:47:57.772
the, the whole promise of these chatbots is to return just what you're looking for and not, like, the publisher webpage on a search result is often just an intermediary to what people, you know, want.

277
00:47:58.012 --> 00:48:07.692
Like, you know, I, I- Yeah... did the search yesterday about- My, my use case is definitely not... which stre- streaming the Eagles game. Yeah. I ended up on a USA Today, yeah, horror show of a page.

278
00:48:07.772 --> 00:48:15.592
[laughs] No offense to my friends at Gannett, but, like, that is not a page I would wanna take to [laughs] the sort of UX hall of fame that's just filled with...

279
00:48:15.632 --> 00:48:27.472
it was basically a Fubo ad, but it was just filled with junk to try to get through to, to... and I get it. I get everyone has to pay bills, but nobody's gonna, uh, nobody's gonna protest if that stuff goes away.

280
00:48:28.212 --> 00:48:41.352
[laughs] Like, I don't- Yeah... think so. I mean, agreed. It's like, you know, it's an ex- existential thing for publishers that AI search, you know, is going to essentially substitute what they're doing and I...

281
00:48:41.682 --> 00:48:53.980
but I think once you get sort of to a more granular level of what they're actually serving up, you find that some, some types of content is m- are more at risk-For being substituted than others, right?

282
00:48:54.180 --> 00:49:01.840
And Yeah You know, like, that's kind of what I was getting at with sort of the sameness of things earlier in, in terms of these AI summarization engines sort of...

283
00:49:02.110 --> 00:49:09.070
And that's partly due to the licensing deal, but also partly, like, what AI is good for. Like, we're always gonna want to, on Saturdays,

284
00:49:10.100 --> 00:49:18.400
go, you know, settle in with a podcast or a feature article or our favorite writer or our favorite substacker and just, you know, kind of go deep on that.

285
00:49:18.940 --> 00:49:29.330
And AIs might be good at sort of pointing you to that, and to some extent even some discovery within a, a summary if you find something interesting that you wanna go deeper on.

286
00:49:29.340 --> 00:49:37.920
But for mo- for the most part, if you're just kinda like, "Okay, I'm busy, but I, I need to know what's going on," you know, AI, AI is an amazing tool.

287
00:49:38.000 --> 00:49:45.620
And, you know, [chuckles] like, to your point of, like, the publisher experiences on a lot of these websites, I mean, it's, it's kind of a better experience if you- Yeah...

288
00:49:45.660 --> 00:49:59.179
if you don't need to do anything deeper than, than what just a, a summary of what's going on. So I think it's absolutely gonna become a greater and greater part of just how people get information.

289
00:49:59.260 --> 00:50:15.980
And it-- the reversioning of it I find interesting because if you think about how attention is now so spread out over various apps, and putting aside, like, the TikTok ban sort of notwithstanding, you know, some people just, like, wanna scroll all day on TikTok and kinda get their noobs that way.

290
00:50:16.000 --> 00:50:20.970
And, you know, whether that's YouTube Shorts, whether that's, you know, podcasts.

291
00:50:21.000 --> 00:50:28.390
And, and then this is something I've written about too, and I don't have the answer to it, like, how much does human authenticity even mean in that experience?

292
00:50:28.480 --> 00:50:40.670
Do you care that that avatar you just saw on TikTok was actually AI if it's, you know, something relatable and, and gives you the information and experience you're looking for? Maybe.

293
00:50:40.980 --> 00:50:50.230
I mean, it, you know, again, it, it, it... I think in terms of the way people consume things and what I was just talking about, I think it matters a lot less for your day-to-day stuff- Yeah...

294
00:50:50.230 --> 00:51:01.900
and matters more for that stuff you're doing on Saturday. Yeah. I mean, already Pew found that 39% of people 18 to 29 get their news regularly from, from TikTok. Hmm.

295
00:51:02.009 --> 00:51:14.480
So, I mean, the patterns are changing already, and AI to me is just a giant accelerant to a, to a lot of this. What about the lawsuits, right? I mean, is 2025 the year when...

296
00:51:15.240 --> 00:51:24.120
Not just the lawsuits, but, like, coming to some kind of economic understanding, right? Like, I mean, there-- we've seen- Mm... the smattering of, of AI deals.

297
00:51:24.240 --> 00:51:31.500
Like, these are just with, you know, a few, o-obviously publishers, not the mass number of, of publishers out there.

298
00:51:32.140 --> 00:51:45.940
I regularly ask publishers and, and about using the, the, the, the different, like, toll bit or pro rata, like, in order to, you know, basically position themselves to, to get some kind of economic value.

299
00:51:46.040 --> 00:51:54.520
Most of them are just, like, deploying the tool because it's free, the tools because they're free, and, like, sort of monitoring, which means they're not doing anything about it.

300
00:51:54.620 --> 00:52:16.260
And, you know, there is, again, we are always left in a could, might in the future, you know, we're gonna be able to, you know, wring some kinda economic value out of these, these AI answer engines, because they're breaking, in some ways, the bargain that underpinned Google Search, which was- Mm...

301
00:52:16.280 --> 00:52:19.560
we're gonna let you crawl our webpage, webpages.

302
00:52:19.980 --> 00:52:38.420
We're gonna let you scrape them, even if Google won't claim it's scraping, but they're scraping data from those, those webpages, and they're putting it on their website in the search results, and the bargain is you're gonna get traffic that you can monetize, and by the way, you're probably gonna use our tools to monetize it, so we're gonna get a taste.

303
00:52:39.020 --> 00:52:46.570
That's off. I mean, the, you know, Perplexity- Yep... for instance, said, "Hey, robots, robots.txt," which is- [chuckles]...

304
00:52:46.570 --> 00:52:57.400
you know, the, the, the little code that basically told Google and other crawlers, "Stay away from these pages." You do it, like, for transaction pages or pages with personal data, et cetera.

305
00:52:57.480 --> 00:53:00.460
But, you know, you always had that option. You could opt out of the system.

306
00:53:00.690 --> 00:53:10.060
And, you know, Perplexity, which interestingly, their CEO has kinda turned into a, a little bit of a heel in this drama, uh, which I think- Mm... is just where Silicon Valley is now.

307
00:53:10.310 --> 00:53:15.300
I know something about that guy's off. Yeah. It, you know, th- they're just saying, "Hey, we're not gonna pay attention to this."

308
00:53:15.380 --> 00:53:33.120
So w- are we gonna see any sort of understanding at all this year of how the economics to all this work? Because a lot of these AI, you know, the vision for these things break the, the economic model, right?

309
00:53:33.240 --> 00:53:40.720
Mm-hmm. And, and look, Silicon Valley, that's their thing, you know, move fast, break things, et cetera. They love breaking other people's economic models, not their own.

310
00:53:41.300 --> 00:53:52.540
[chuckles] So are we gonna start to see some kinda understanding about how, how the economics work out? Because, you know, you can get into the doom loop of, like, well, there's no economic incentive to create content.

311
00:53:52.580 --> 00:53:58.640
Like, it will not be created, and then it's gonna be filled with junk, e- et cetera. So one thing...

312
00:53:59.000 --> 00:54:19.360
So I'm not, obviously I'm not 100% sure how this is all gonna work out, but one thing I am sure of is that there is a consensus building that might actually already be taken over, which is that if an AI engine is going to ingest, uh, publisher content and then summarize it,

313
00:54:20.420 --> 00:54:32.410
that is... The industry is deciding that's more like syndication than search. And this is-- I'm talking mostly about the real-time stuff, right? Not putting training data aside. Like, if you're summarizing news. Yeah.

314
00:54:32.440 --> 00:54:41.980
If you're summarizing stuff that's out now. Yeah. They need direct payments. Yeah. The indirect stuff just not gonna work here. Yeah. Because it's just, it's, like, I don't know, little citations, I don't...

315
00:54:42.060 --> 00:54:47.680
It's not gonna work the same as search, I don't think. I know Google- Yeah... says that people, you know, they, they're driving as much traffic.

316
00:54:47.740 --> 00:55:00.092
I just, it, it's just hard for me cons- using these, these services, the entire point of them is to not have to go to all these websites. Right. I mean, like, come on, that's the, like, that's the USP.Yeah.

317
00:55:00.512 --> 00:55:04.532
We're starting to see, like Google's kind of stayed away from news so far.

318
00:55:04.912 --> 00:55:13.332
It's left it to ChatGPT and Perplexity, but we've saw the first hints of them, they are moving in that direction, and, and obviously that's gotta be on the roadmap.

319
00:55:13.372 --> 00:55:25.432
That this-- they're gonna apply generative summaries to, to news and, and Google News at some point. And, you know, that's obviously gonna be a bomb that [laughs] that, that lands in the industry when, when that happens.

320
00:55:25.452 --> 00:55:33.012
But they'd started d- at CES, they announced for their Google TV platform, they would start summarize-- giving sort of some basic summaries.

321
00:55:33.072 --> 00:55:37.492
I don't think that product's out yet, but that, that was one of their big announcements. Yeah.

322
00:55:37.512 --> 00:55:47.972
Wait till they start to stitch together like, you know, video of, you know, different newscasts and, and YouTube- Uh-huh...and like, ah, that sounds like a pretty good, and sounds like a doable product.

323
00:55:48.132 --> 00:55:54.472
You know, I'm not an engineer, but it seems, doesn't seem so far-fetched. Well, you imagine those NotebookLM people as, as avatars, right?

324
00:55:54.612 --> 00:56:05.792
They're just gonna be people that are just chatting at a, at a news desk or, you know, on, in the field somewhere. But so, so the thing is, with these deals and the lawsuits, so far it's only the big players.

325
00:56:06.432 --> 00:56:13.172
If you're a small player, yes, you can go to a Tolbit or Adapt or any of these other sort of startups that are trying to create this.

326
00:56:13.252 --> 00:56:28.852
The problem is that they haven't really gotten the attention of the big AI guys yet in, in a major way. And my sense is they won't until the sort of people at large demand it.

327
00:56:28.992 --> 00:56:40.832
In other words, like, someone's gotta crack that user experience where, "Oh, yes, I want my news this way and I expect it from you, ChatGPT, and everyone else," Perplexity.

328
00:56:40.872 --> 00:56:44.812
Again, Perplexity is sort of a different animal because they don't do licensing, they do rev share.

329
00:56:45.272 --> 00:56:52.762
Obviously, they've gotten into legal hot water over that, but with News Corp, and, you know, we'll, we'll see how that all plays out.

330
00:56:53.412 --> 00:56:59.452
I gotta say, given my, what I sort of just said about the consensus emerging, I, I, I worry a little bit about Perplexity.

331
00:56:59.482 --> 00:57:14.092
I do think kind of their, the sheen of that service, which was kind of a darling about a year ago and everyone was admiring, has kind of worn off, and they're kinda throwing spaghetti at the wall with- Well, this TikTok thing-...their shopping...that's, that's, that's where I, I like, I, I kind of like- Oh, yeah [laughs]...I'm out on this guy.

332
00:57:14.192 --> 00:57:24.692
I mean, there was two things. One was when the, the CEO during the New York Times tech worker strike said, "Oh, we'll take over your server." I'm like, "Eh."

333
00:57:24.732 --> 00:57:34.472
I, I-- That just struck me as just this Silicon Valley buffoonery from all the money being thrown around there. I was like, "This guy's too big for his britches. He's not focused on the right things."

334
00:57:34.532 --> 00:57:54.352
And then there's this talk of, of taking over TikTok and, and it being half-owned by the government because obviously Perplexity does not ha- control a distribution choke point, so it's- Mm...just some random app that people have to download and, and the, just the, the deck is stacked against it.

335
00:57:54.652 --> 00:58:02.642
Yeah. Yeah. It's, it's a, it was an odd thing to do, obviously. I think it was more of a publicity stunt. They- That's what I mean...get their name- Like people are doing-...talked about it...publicity stunts.

336
00:58:02.972 --> 00:58:06.542
Uh, it's just like I, I listened to a podcast, the, the, uh, it was actually Prescient.

337
00:58:06.972 --> 00:58:18.792
It was a, a more or less the podcast, and Sam Lessin was saying, "Anytime people are getting excited about spending money versus making money, it's like- [laughs]...be very, be very concerned."

338
00:58:18.902 --> 00:58:28.572
[laughs] So I think we're seeing that in the stock market today, but, uh- Well, the thing is, I wrote about this too. I, I do see kind of a, a, I guess a synergy there to, to, to pick a terrible word.

339
00:58:28.632 --> 00:58:45.392
It was kinda like, you, you know, if, if TikTok has this AI connection, which, you know, doesn't really have an AI strategy right now and has this connection to Perplexity that can, you know, its users actually kind of find useful, and I'm not sure what that would be.

340
00:58:45.472 --> 00:58:48.372
I could see that, that sort of working out for them in some way.

341
00:58:48.412 --> 00:58:59.162
Like, 'cause so they sort of instantly have kind of a, an AI strategy that's specifically about surfacing good summaries and then pointing people to the right information if they wanna go deeper on something.

342
00:58:59.182 --> 00:59:05.092
And then obviously TikTok, the benefits to TikTok are, are huge 'cause, you know, you're just exposed to all these users. Will it happen?

343
00:59:05.482 --> 00:59:10.352
Um, probably not, but it's, you know, it's, uh- Yeah...it's an interesting thing to think about. Be a Frankenstein product.

344
00:59:10.502 --> 00:59:21.382
[laughs] Even if it's not, even if it's not Perplexity, it's kinda like, well, is there, is there some AI solution that takes TikTok to some other level? Anyway, TikTok's got a lot of problems. Yeah.

345
00:59:21.382 --> 00:59:31.332
Doesn't, I don't think it has to worry about its long-term strategy until it's sort of figured out this, this whole situation with the ban. Okay. All right. Pete, let's leave it there. This was enjoyable as always.

346
00:59:31.352 --> 00:59:43.832
We covered a lot of ground. Just really appreciate it. Yeah, my pleasure, Brian. Thanks for having me on. [outro music]