The Exchanges

Q So, so tell us about the, the, the whole thing then. When, when did you guys start, um, and what has the process been? Because you have six million of those pictures?

A Yeah, we think. Like, we don't really know, honestly. There's a whole, there are piles on top of the filing cabinet, so it's unclear, but we think it's about six million. Um, I mean, the digitization process started in, uh, last spring, I believe. Um, but the, the, we've been talking about this with Google for at least a couple of years, And honestly, it's a project we've been trying to do for like, 15 years, maybe, you know, maybe 20. Um, And what I like best about this project actually is it, to me, it sort of captures a moment where, um, this kind of elusive project, which is kind of a nice-to-have project, like this was not a critical, like, business need for us, but it was something that was eating at us all because these photos are, like, moldering in the basement, and we're, you know, we know what a great asset it is, and how it's just fascinating they are. Um, it's a moment where suddenly that becomes economical to do, and And not only does it become economical, but it, like, it, then it starts to feed this cycle where, you know, we, like, all of this data becomes accessible, um, you know, because of the cloud and the, and, uh, access to compute power and storage and so on, um, and that, and then becomes potentially tractable to machine learning. At the moment, we're just doing a little bit, but as that allows us to develop techniques to work with, like, very large bodi…

AI assessment note: “the digitization process started in, uh, last spring, I believe.”

Q And for now it's for internal use, right? It's for the newsroom to access the archives, or is that going to be exposed to the public eventually?

A Um, we, we're thinking about it. We might expose it to the public. Um, I think, uh, I'd love, I personally would love to see that happen. Um, there's rights issues like there always are, like some of those notes scrawled on the back of those photographs, you know, to know who owns the image, who took the image, and there's some complexity. Um, but, uh, but for now it's, it's for internal use, but it's being used actively to tell stories to the public. So we're routinely now running these backward looking series, um, uh, that are, that we're reminding the archive to tell specific stories. The best example is the overlooked series where we're going back and running obituaries of famous women who didn't get obituaries, uh, printed in the paper. Because we were stupid. So, um, that's a great, like, moment. It's a great way to, like, retell, um, you know, our past stories in a way that reflects the current moment. Um, so it's fun.

AI assessment note: “We might expose it to the public... but for now it's for internal use”

Q And you mentioned mention, uh, machine learning, a little bit of machine learning. So what, what, what do you use it for now?

A Well, for the digitization processes, I think it was Alan who mentioned it, um, uh, we're doing some, uh, uh, text recognition on those notes on the back, and as we're learning more about the, like, quote, structure of those notes, which is very, like, evolves a lot over the years and is very haphazard, um, that helps us extract more of that data off the back of the pictures, um, as well as doing facial recognition and place recognition on the content of the photos to pick up unexpected I wish I had a good anecdote. I was trying to think of one coming over. We have found, like, unexpected people in the backgrounds who later became famous and notable and, like, recognizable. Um, so that kind of thing to extract either on, you know, information we didn't know was in those pictures or, um, that's encoded in this, you know, kind of loosely on the back of the photos. That's what we're doing today. Um, but it's particularly that ability to, um, Work with and correlate, ah, the imagery across time that I think is pretty cool going forward.

AI assessment note: “we're doing some, uh, uh, text recognition on those notes on the back”

Q So that's actually a perfect, um, segue into talking about the, uh, this infrastructure at the, at the New York Times that you, that you oversee. Uh, so when, when did that migration start? Where, what, what was the situation before? You guys had your own data center, or how did that work?

A Yeah, we were behind, you know, it's significantly behind the curve. Um, we had four, Of our own least, you know, data centers, a lot of stuff in, in AWS, some of it in the VPC, some of it's just laying around in like AWS classic, um, and not really a strategy around what, how, why we were doing what. When I came in, we started to look at it and sort of impulsive, like I had an eye. This was like late-thousand-fifteen, early-thousand-sixteen. I've been watching Google Cloud for a while. I'm like, when are they going to get serious, blah, blah, blah. The obvious thing was to plow ahead to AWS. Then at the last second, we were like, well, let's take a look, see how they're doing. We looked at the data infrastructure first. Um, took a look at BigQuery, and we're like, holy shit. So, and from, so we're like, okay, data, we're gonna go all, um, all Google, which is like PubSub, Dataflow, Dataproc into BigQuery. And then from there, um, we continued to get excited and ended up going pretty deep.

AI assessment note: “we had four, Of our own least, you know, data centers”

Q And, uh, what else? Are you guys a big, uh, open source shop, or what, what do you, uh?

A Uh, I mean, every, everybody is, to some degree, what, what, like, it's, what is open source these days anyway. It's sort of confusing and weird, but, um, Uh, you know, we, a lot of the things we dabble in are pretty familiar on the front end. It's, it's React, um, and a little bit of Node, and a, you know, GraphQL, um, sort of around the, you know, that's sort of, pardon me, sort of API to the content. Um, we did a big Kafka implementation as well, which just runs our main publishing pipeline, so all content flows through Kafka on its way to the front end. Um, we, actually, there's a, there's a New York startup that we're, uh, working with, Um, called ActionIQ that is like a marketing campaign management platform that's sitting on top of our BigQuery data lake that we're pretty happy with because that was, that's been a real sore spot for us historically. It's like running our, our increasingly, um, large pre-scale marketing operations.

AI assessment note: “React, um, and a little bit of Node, and a, you know, GraphQL”

Q Do you want to elaborate on that? I thought that was super interesting.

A Um, yeah, I think, um, If you read the post, the way that this really happened was sort of accidental. Like, I showed up and started to work with the team, and we were kind of like, You know, inevitably, in software teams, like, culture is important. Like, how do you have people thinking in the right way, collaborating in the right way? How do we, you know, create a culture that people want to join? And we just started to think, like, well, ok, what, like, that's a hard conversation. That's weird, and culture is hard to, like, grab ahold of, but what's one thing we can grab ahold of that we think would help? Um, and we started, I think it was career ladders. So we're like, ok, let's, like, take that. Let's write down what we think it looks like now. And then, you know, let's go through a process to, like, iterate on a little bit, and let's, like, test it, and then let's try to get it out there, and, like, let's keep looking at it. Um, and, and we repeated this process, like, over and over looking at every kind of thing that we thought we could materialize that would have some impact on, you know, on the way it felt to be a software developer at the New York Times. And it was only after we'd been doing this for a while that we were kind of like, oh, surprise, surprise, we're all software engineers, this looks a lot like a software development process, you know. Um, but, but I th…

AI assessment note: “modularizing what culture meant.”