Showing Your Work with Ben Welsh

By • The Data Journalism Podcast in Spotify

Recording

Listen on the original podcast site

Show the timestamped transcript
  1. [MUSIC]
  2. >> Hello, and welcome to the Data Journalism Podcast.
  3. I'm Scott Klein.
  4. >> I'm Alberto Cairo.
  5. >> And I'm Simon Rogers.
  6. >> This week, we're joined by a returning guest, Ben Welsh from Reuters.
  7. Ben is a long time proponent of open source and
  8. published lots of great open source projects.
  9. And so he and I paired up a couple of months ago to take a look at
  10. open source activity and news and came out with a piece on the source blog from Open News.
  11. Since then, I have been tracking open source activity in news and
  12. I've actually launched a newsletter at openjournalism.news.
  13. And frankly, I've been watching open source activity in news a lot since then.
  14. And there's a lot to be happy about, a lot of encouraging activity.
  15. A lot of really good repos coming out, a lot of good open source activity.
  16. And I've gotten a lot of great feedback from the community on the work.
  17. So this interview, we talk kind of much more generally about open source activity in news.
  18. >> Amazing, and the thing about Ben is that I don't think modern data journalism
  19. would look the same without him.
  20. He does such incredible work around opening up data and making it more available.
  21. And it was a real pleasure to spend time with him.
  22. He is actually a kind of regular fixture at Nycar where we recorded this interview.
  23. And he often does this thing like the full Nycar where he basically presents every
  24. day constantly.
  25. So he's a bit of powerhouse.
  26. We were kind of lucky to have some time with him.
  27. >> Ben is also a returning guest.
  28. >> Yeah. >> He visited the, yeah,
  29. the podcast more than four years ago.
  30. So it was great to hear what he's been up to since then.
  31. He has a new job.
  32. And as always, I found his observations about where the industry is going,
  33. where we are, and where we are heading.
  34. I think that they are spot on.
  35. >> Okay, let's get into it.
  36. [MUSIC]
  37. All right, we are live from Nycar in Indianapolis.
  38. First time ever, and we're so glad to be here with Ben Welch.
  39. You've got a new job since we last spoke.
  40. Tell us what you've been up to at Reuters.
  41. >> Thank you for having me.
  42. It's a pleasure to be back here at Nycar and talking to you guys again.
  43. I now live in New York City, money making Manhattan,
  44. where I walk to work every day at Three Times Square and
  45. work at the Reuters Newsroom.
  46. Reuters is an international news agency with more than 2,700 journalists
  47. all around the planet, and one node in that big network is right there in New York,
  48. where we cover markets, economics, business, and really anything else.
  49. And I lead a small team who are working on automating data for breaking news.
  50. >> Tell us a little bit more about that.
  51. What does that actually entail day to day for you?
  52. >> Sure, if you think about what's going on on Wall Street,
  53. as the charts move up or down, there's an opportunity to cover that with data.
  54. And a traditional approach would have been to assign somebody to go and
  55. download a spreadsheet and reformat it and stick it into a charting tool.
  56. And then get an editor to look at it and push it through.
  57. And that worked great, it just wasn't necessarily the quickest thing and
  58. can only cover so many topics at one time, your throughput was limited.
  59. And what we really see is an opportunity for automation to really attack that type
  60. of breaking news coverage of market moving information in a way that gives us
  61. greater speed and scale to talk like the corporate denizen that I am.
  62. And so if you think about it, that data that's coming out of the Korean stock
  63. market, the oil prices, the whatever else, the futures market is updating minute by
  64. minute and what our team has done is set up hundreds of automated pipelines that
  65. are drawing from all these different data feeds identifying when new records are
  66. available and automatically publishing charts that are ready to go within seconds.
  67. And so when the oil prices ran wild on Sunday night after the attack on Iran,
  68. we had the live chart up on the homepage within minutes before the tech story
  69. about the increase was even ready to show that and
  70. then it can develop live as the new numbers come in.
  71. We're trying to take that approach to as many different topics as we can,
  72. not just commodities and markets, but also public opinion polls.
  73. We routinely publish one of the broadest American public opinion polls.
  74. And by automating the charts in coverage,
  75. we're able to just get it out quicker and faster.
  76. So on that Sunday, we had the first public opinion poll of what Americans thought
  77. of the attack on Iran and again, the chart was ready before the tech story
  78. because we had it all wired up.
  79. >> So there are two really exciting things about it, at least.
  80. One is that you didn't create your own graphics stack,
  81. you didn't kind of invent the entire thing.
  82. But you're in fact wrapping it around data wrapper,
  83. which is something that really lots and lots of newsrooms are now using.
  84. I know the New York Times has switched back to if they're still using that code.
  85. And it's become one of maybe two with Flourish data wrappers become one of
  86. the two sort of main graphics engines that newsrooms are using.
  87. Which leads me to the second thing that's exciting is that you actually also
  88. publish these pipelines as open source code, right?
  89. So newsrooms could adopt this and be able to, I mean,
  90. not that many newsrooms need to be as quick as Reuters.
  91. But newsrooms could adopt these to be useful for
  92. their communities in different ways, right?
  93. So one thing we released last year that has been super useful to us, and
  94. I suspect will be to others, is a Python library that kind of acts as a wrapper or
  95. a helper for automatically making charts with data wrapper.
  96. So most of us encounter data wrapper in our browser as a point and
  97. click interface where with relative ease you can make a really great chart, right?
  98. But it also has an API that allows a computer or
  99. a machine to make a chart by posting into that system the instructions for
  100. how to make it.
  101. But to do that requires a good amount of technical know how and
  102. you kind of have to decode the kind of dozens of little options that configure,
  103. make the bar this color, set the axis this format.
  104. And most of that's undocumented.
  105. So what we've done is written a Python library that fully maps out all of those
  106. options for eight or ten of the data wrapper chart types and
  107. fully documents how they all work.
  108. And so now with a very small amount of Python code, you can really easily
  109. configure and automate a line chart, a bar chart, a column chart.
  110. And that's now open source and available for anyone.
  111. And we also took it one step further, not to get too nerdy here,
  112. but I feel like this might be the venue.
  113. >> This is the venue for this.
  114. >> Yeah, we connected that to what's now called a model context protocol server,
  115. which is kind of an emerging standard for making a kind of pluggable system for
  116. connecting outside data sources and tools into an AI chat bot like OpenAI or Claude.
  117. And because of the way the Python code is written,
  118. it's actually kind of perfectly suited to kind of connect with the chat bot.
  119. I don't know if folks have done this, but there's an emerging standard as well for
  120. sort of teaching the bot how to deal with different data structures.
  121. And the system that we use to help ourselves make the chart is the same
  122. system that the AI use, so we've released an MCP server.
  123. It's still experimental, but it can be plugged into any chat bot.
  124. And you can then say, hey, chat bot, here's a spreadsheet.
  125. Could you make me a bar chart?
  126. And boom, it'll talk to DataRapper and bring it right back.
  127. >> That's so helpful.
  128. I mean, the charts that AI tends to make have not been great.
  129. Certainly nothing to the level of what DataRapper is capable of.
  130. So that's exciting, I'm gonna try that.
  131. >> Yeah, I just talked to the Politico crew here at the conference, and
  132. they're installing it in their local newsroom too, which is pretty exciting.
  133. >> This is gonna lead into what we're really here to talk about.
  134. But why do you make Open?
  135. Why make it available to everybody?
  136. Why isn't it like secret, commercial, confidential information?
  137. >> Well, I think we have a lot to benefit from making it open.
  138. Within days of me releasing it, I received a contribution from a developer
  139. in a newsroom in Germany who I've never met, who contributed support for
  140. a DataRapper map and automating how to configure that that we hadn't done yet.
  141. And so by inviting other people into the project,
  142. we got something back out of it, right?
  143. And so to me, there's definitely the reciprocal part of it.
  144. For me, I think open source really demands a higher standard of coding and
  145. quality, just even knowing it's gonna be public.
  146. Sometimes it pressures you to make it a little better.
  147. And I think that that's what an editor I used to have liked to call good pressure.
  148. And so I think that it puts a little good pressure on it there.
  149. And I just think it's also the right thing to do, not to sound too corny.
  150. We're not releasing the stuff that really makes our system break news and
  151. beat everybody else.
  152. We're releasing one component of it that I think is kinda shared across the field.
  153. And then I think I don't really see much of a downside for us in doing it.
  154. >> So what Ben is talking about and Simon is talking about is a piece that Ben and
  155. I published earlier this year that came out of some of the research that I've been
  156. doing for years and Ben joined and was doing some physical analysis over the last
  157. year that was really looking at the prevalence of open source activity
  158. in newsrooms in America, and really kind of chasing a hunch that we had.
  159. A sort of feeling we had that the activity has really tailed off in recent years.
  160. And it actually turned out that I had a data set that was ready to go on this
  161. because I had, back when X was called Twitter, I had a data set of all of
  162. the newsrooms open source repository, their open source accounts on GitHub.
  163. And I was chatting with Ben one day and we decided to take a look at the,
  164. use this data set to take a look at the activity to see if it really had
  165. tailed off.
  166. Ben, do you wanna talk just for a second about kind of what we found and
  167. what the piece is about?
  168. >> Yeah, so we took Scott's list of I think about 350 news organizations on
  169. GitHub and wrote a little Python code to go download all the repositories that
  170. have been made by those organizations ever.
  171. Just to kind of build a data set of all open source activity,
  172. at least on GitHub, by this kind of very large broad set of news orgs.
  173. And then when we started analyzing and looking at it, we found that not only
  174. was Scott's hunch that activity was going down right.
  175. It was, the trend was much more dramatic than we had expected.
  176. If you just take a look at something like the number of new open source
  177. repositories, that number peaked at 2,000 new repositories or
  178. more than that in the year 2016, and it has just been in a consistent and
  179. really quite dramatic decline since then.
  180. Each year stepping down to fewer and fewer with a kind of a low point reached
  181. in 2024, I think of around only 400.
  182. So we've seen just in less than a decade, an 80% drop in the number of open source
  183. repositories published by news organizations.
  184. Which is a, as a long time data journalist, I said, wow,
  185. that sounds like a nut graft to me.
  186. >> Yeah, yeah, and it wasn't just open source repositories.
  187. We wanted to, that's an easy thing to count.
  188. And so it was a good proxy in that sense, if not an imperfect proxy like all
  189. proxies are.
  190. But we also looked at posts to the nikar-l mailing list, so we're at nikar.
  191. So I think it's just natural for us to talk about the nikar-l list.
  192. And there was a time when the nikar-l listserv was the must read email list for
  193. data journalists.
  194. I remember when I let it buzz my phone when a new nikar-l post would come.
  195. I would get a notification on my phone cuz I wanted to read it as soon as it came
  196. out, and we found that there was a similar decline on the nikar-l mailing list.
  197. I think that was down 89% from its peak at around the same time.
  198. So really a culture of sharing that really peaked in 2016 has tailed off
  199. considerably, not just in the stuff that's easy to see, but
  200. I think in the stuff that's kind of harder to measure as well.
  201. >> Definitely, I think one of the things that always attracted me to data
  202. journalists originally was this culture of sharing.
  203. Because often data journalists, you're kind of isolated where you are.
  204. This idea you're part of a bigger community, but actually also that it can
  205. make your work better because people are gonna come back to you and
  206. improve what you do, just like the example you gave.
  207. So obviously, the peak is 2016.
  208. That is a different world now in terms of data journalism.
  209. Do you wanna talk about how different that was?
  210. How data journalism's changed since then?
  211. >> Well, how different was in terms of just number of outlets?
  212. We have BuzzFeed doing a ton of data journalism.
  213. 538 were incredibly prolific contributors to GitHub.
  214. >> Without a doubt, the first thing that came to our mind,
  215. we're asking the question, what could explain this?
  216. Why has this gone down so much?
  217. And I think if you compare today to then,
  218. the thing everybody who's paying attention thinks of first is economics, right?
  219. We just have all these places that were large contributors.
  220. You just listed them.
  221. And if they go out of business, they can't keep contributing, right?
  222. It's kind of a pretty basic explanation.
  223. And I think that's just true, you know what I mean?
  224. You definitely see in our data set that there's a strong cohort of places.
  225. They're just gone, and that accounts for some, but not all of the decline.
  226. And that's where it started to get interesting for
  227. us in kind of reporting out the data set.
  228. Cuz what we saw is that there were other cohorts or
  229. other groups of news organizations within the data who are also declining,
  230. who didn't have those economic problems.
  231. You know, I'm not here to name names or call anybody out, but
  232. we can all think of the very successful news organizations of the Internet era.
  233. And if you look for them in our data set, you're gonna see they're dropping too.
  234. So it can't all be the money, right?
  235. So something else must explain it.
  236. And that's what set Scott and I off on a little bit of a reporting expedition.
  237. You wanna share what we did, Scott?
  238. >> Yeah, so as journalists, as data journalists, we didn't just stop at
  239. looking at the data and publishing what the data findings were.
  240. We wanted to talk to people to see if they had a hunch about what changed and
  241. why the numbers had gone down.
  242. So we talked to more than a dozen people.
  243. These were practitioners, people who we felt were big advocates for
  244. open source or just big open source releasers back in the day.
  245. And talk to them about, first of all, did they agree with the hunch?
  246. We let them know what the data showed.
  247. And asked them to help us understand what changed in their newsrooms.
  248. What were their motivations for open sourcing in the first place?
  249. And did those motivations go away?
  250. What changed about their workplaces that might have led to a decline,
  251. a sort of zeal to open source everything.
  252. >> And in addition to economics, we really heard two other major themes kind of come
  253. up in different ways from different perspectives as we went through it.
  254. One was the idea of technological maturity, you might say, or
  255. like the we fixed it theory.
  256. And several people we interviewed said one reason you might see less open source
  257. is we just don't need it as much.
  258. Some of the problems have been solved.
  259. You look at the explosion of open source in that period is associated with
  260. the development a lot of web frameworks for publishing data on the web.
  261. Be it Django or Sveltekit or what you name it, D3.
  262. And now that those problems are quote unquote solved,
  263. maybe there's less need, right?
  264. So that was one theory we heard from folks.
  265. And I've heard that concurred by others.
  266. And then a third theory we heard, and
  267. this especially came from the people in the large successful newsrooms, I would say,
  268. is the idea of kind of a cultural change that's happened in the winners of
  269. the internet era.
  270. It's articulated by one person we interviewed as saying when your team has
  271. moved to the center of the newsroom, it's no longer the boutique
  272. kind of experimental R and D nerds in the corner.
  273. But they're now part of the capital P product and the quote unquote main thing,
  274. right?
  275. Your bosses are different people with different outlooks, right?
  276. Your incentives are different.
  277. You're focused on meeting the OKRs and
  278. the quarterly goals, not kicking up something cool to see if the Internet is
  279. a real thing or not, right?
  280. And that that cultural change accounts for why, in their theory,
  281. why some of these large newsrooms have dropped.
  282. >> But while that's happened,
  283. there have also been like really interesting startups,
  284. things like The Pudding or City Bureau, Bellingcat,
  285. which are kind of pioneering new types of reporting.
  286. Do you think that's making a difference, Scott?
  287. >> Yeah, I mean, there definitely are newsrooms, especially places that are
  288. startups like The Pudding, Bellingcat, there are a few others that
  289. are really punching above their weight when it comes to sharing.
  290. So The Pudding puts a lot of things open source.
  291. There's a terrific newsroom in Harrisburg, Pennsylvania called Spotlight PA.
  292. I think they put their entire everything.
  293. I think they put things up on their GitHub before they publish it.
  294. I mean, they are really, really doing kind of keeping the old ways, in a way,
  295. doing really terrific work.
  296. >> Yeah, I think that's right, cuz our findings are really just in the aggregate,
  297. right, we're talking about news at all.
  298. But there are some real good news stories inside the data set when you dig down.
  299. And I just think we really can't say, in my opinion,
  300. enough good stuff about the success of the open source intelligence movement,
  301. as it's known, who I think are one, modeling excellence in journalism and
  302. hard work and creativity and inclusiveness, but also sharing.
  303. And I think that in data, my little niche, my little corner of that big curve,
  304. we would do well to pay attention to how they've been succeeding and
  305. try to emulate them.
  306. >> Yeah, I mean, data journalism was hardly new in the early 2010s, but
  307. it sort of felt like we were bringing something new to newsrooms.
  308. It was a lot of people coming from tech or from outside news were coming in,
  309. bringing some of the sort of open source fervor.
  310. These are people who worked at places that open sourced without even,
  311. you didn't need a recent open source, of course you open sourced everything.
  312. So I think there was a lot of energy coming in, it was a fairly new,
  313. at least the way we were practicing it, and thinking about it was fairly new and
  314. growing.
  315. And I think on some level, that's what OSINT is now,
  316. what open source intelligence is now.
  317. And I think, in a way, Ben talked about maturity.
  318. It's almost, it's a good thing that data journalists would have to come to
  319. a conference like NYCART to see people who really understood what they did.
  320. And to sort of gather meaning and standards and sort of edit each other in
  321. a way that they didn't necessarily get from their newsrooms.
  322. Where we all sort of worked in places where we had bosses that were very bright and
  323. very good journalists, but not necessarily people could help us with our Python
  324. problems or decide between pandas or R or whatever, right?
  325. And so we kind of turned to each other for that, and in a way,
  326. that's what created the culture of sharing.
  327. And now a lot of our newsrooms have become much more sophisticated,
  328. much more data literate.
  329. And we are turning inward, and we talk all about the inward shift in the story.
  330. Where there are actually people in our newsrooms that we can turn to,
  331. to get the answers to our Python questions and to edit our math,
  332. as Amanda Cox used to put it.
  333. And in a way, we kind of didn't, we didn't need each other quite as urgently.
  334. And so the culture of sharing became more of a conscious effort than the default.
  335. >> I love that you also included a piece about Django, because Django is so
  336. important and something that a lot of people perhaps don't know about now.
  337. But without it, the web would not look how it looks.
  338. Do you want to explain what Django is for people?
  339. >> Yeah, I mean, we've all used Instagram, unfortunately, right?
  340. And believe it or not, the first edition of Instagram when it was launched as
  341. a startup was built using an open source web development toolkit called Django.
  342. And that toolkit, which launched this billion dollar business that now
  343. dominates the attention of many Americans,
  344. was invented at a small local newspaper in Lawrence, Kansas.
  345. Where at the Lawrence Journal World, a team of ragtag nerds decided to build
  346. a better newspaper website, and in doing so, created a set of tools for
  347. building websites out of databases that then grew into one of the most popular
  348. ways to make a website of any kind.
  349. And it's just one of, I think, a whole list of stories and
  350. examples of where really significant and
  351. world changing kind of technology was developed in newsrooms.
  352. And these are stories, I think, that we know in our little world, but
  353. might not be more broadly known.
  354. >> Yeah, you sort of think about news as a legacy business, but
  355. Django is far from the only example.
  356. D3, the JavaScript framework that helps create graphics was incubated,
  357. I would say in a newsroom was sort of invented before Mike Bostock got to
  358. The New York Times, it was really incubated there.
  359. Backbone underscore sort of a lot of JavaScript behavior frameworks.
  360. >> Svelte.
  361. >> Svelte, speaking of JavaScript frameworks,
  362. very much came out of Rich Harris's work at The New York Times and elsewhere.
  363. And I found out in writing this piece that the varnish proxy
  364. write through cache was written on behalf of a newsroom in Europe.
  365. So there was a time, and in many ways, it can still be here,
  366. where newsrooms were not just open sourcing code for other newsrooms to use,
  367. but open sourcing code that people far outside news were using.
  368. >> And Django, named after the Jazz guitarist Django Reinhardt.
  369. >> Yes. >> Cuz Adrian Hollabarty,
  370. massive Jazz fan.
  371. I always felt that when you put data sets online or put anything online,
  372. the numbers of people it would attract would be very small,
  373. often like 10 people, one person even.
  374. But to me, that didn't matter because it was a sign of transparency and
  375. the sign of kind of openness, which is important,
  376. especially in this age of trust deficit that we see.
  377. How important is it, do you think, like emotionally almost,
  378. for news to be part of this open kind of culture?
  379. >> Well, I think it's important to do just for its own sake, as you said, for
  380. transparency in the scientific spirit of Phil Meyer that really launched
  381. this whole movement.
  382. I think it's part of being credible and convincing to your audience.
  383. But I wouldn't write off its appeal as a product either.
  384. I think we can point to a number of data products that are incredibly
  385. successful in the commercial marketplace and with readers.
  386. If you look, for instance, this week we had the first midterm elections of the year
  387. in the United States, in Texas, and elsewhere.
  388. And the live election result pages on all the different news sites that night were,
  389. I'm sure, drawing absolutely stupendous readership.
  390. And those are data products, just like anything else.
  391. And they are made by data journalists, including ones at this conference.
  392. We can look at the coronavirus story.
  393. We can look at, this one might be a little controversial to some.
  394. We can look at the WikiLeaks story and many others, and
  395. see cases where publishing data on the web has had really dramatic readership.
  396. >> Yeah, Simon, you bring up a really good point,
  397. which is the idea of sort of reproducible research.
  398. And when I was at ProPublica, one of the things that we really paid attention to
  399. was the rise of the reproducible research movement within academia.
  400. And the idea that it was incumbent upon us,
  401. if we were doing a sophisticated data analysis, to publish, initially,
  402. white papers where we explained, we're sure, a very sort of academic light,
  403. although never peer reviewed, sort of research papers to help people understand
  404. our analysis.
  405. And then later, when Jupiter Notebooks, then called IPython Notebooks,
  406. came out to actually publish or reproduce what you could run.
  407. Now you could do it right within GitHub, but you could rerun our analysis and
  408. see where we might have screwed up.
  409. We did that very much in the spirit of journalism,
  410. where we said, here's our methodology down to a T.
  411. You could reproduce it and do the exact same thing if you have the same data set
  412. that we do, which often we published as well.
  413. And you could do it, and if we did our math wrong, we want to know.
  414. Just like every journalist wants to know if they get it wrong, and
  415. then want an opportunity to get it right.
  416. And so that was very much part of our impetus for
  417. open sourcing all of our data stuff.
  418. And on some level, that is still happening.
  419. We're going to be talking at this conference with the Myre Award winners.
  420. And in both of the cases of the folks we're talking to, ProPublica and
  421. the Baltimore Banner, they published very detailed methodologies,
  422. explaining all of the decisions that they had to make, because the data's imperfect.
  423. And the code out there that's available to do this work is imperfect.
  424. And all of the decisions that they had to make that brought them to the analysis.
  425. So that too, it's not just creating Svelte and D3 and open source products.
  426. But also simply as part of, it's how we are transparent as data journalists.
  427. We have an opportunity to be very transparent about
  428. our methodology down to the line of code.
  429. And why would we not take advantage of that?
  430. >> But how do we fix this?
  431. Because I know you guys actually spent a lot of time talking to people about
  432. how to make things better.
  433. So what are your recommendations?
  434. >> Well, we asked everyone, we interviewed that same question.
  435. We also convened a panel at the News Product Alliance conference last fall in
  436. Chicago and asked the whole room full of journalists the same question.
  437. And we've had, I have to tell you, dozens and
  438. dozens of conversations with different data journalists since the story's come out.
  439. And we've asked all of them, how do we fix it?
  440. Same question.
  441. I don't think anybody knows the perfect answer.
  442. And I don't think there's one answer.
  443. There's probably a lot of different things that need to happen.
  444. But ideas that came up, I would be happy to share.
  445. One is the idea of trying to make open source more of a norm within the project
  446. cycle.
  447. One big reason a lot of projects were open in the past, we heard from people,
  448. is that was how you started a project.
  449. If your project started open, it was easier to keep it open rather than
  450. waiting till the day before you published and
  451. having to mop everything up and shove it out.
  452. So kind of better defaults was one idea that they'd come up.
  453. Another is trying to create in these sort of growing and
  454. strong institutions that are succeeding right now.
  455. Maybe a position is focused on this.
  456. And open source editors, is that what they called it, Scott?
  457. >> Mm-hm, mm-hm.
  458. >> What other ideas did we hear?
  459. Hm.
  460. >> Well, one of them, I mean, one of them was just like, give it the office.
  461. Like, if you want open source to come back, open source your stuff, right?
  462. Start with you, was one of the- >> Pull up your pants.
  463. >> Yeah, just do it.
  464. So that was definitely one of them.
  465. Assigned somebody to coordinate, we heard, make an open source award.
  466. So sort of best open source project of the year.
  467. We also heard from people who were talking about, really among the people who
  468. said, you know, what changed in my newsroom is that we matured not just as
  469. journalists, but as product managers.
  470. So as product development became more sophisticated and
  471. more formalized in newsrooms, you needed an OKR, you needed a goal.
  472. You know, it had to link to some corporate strategy to open source things.
  473. And so you just could never prioritize it.
  474. So one of the things we heard is, you know,
  475. the bosses should make it one of the goals that we should try to open source for
  476. projects this year.
  477. Or we should try to document, you know, five projects this year so
  478. that we can't open source it.
  479. So we also heard from people that if one of the reasons that this might have slowed
  480. down is that formal product management kind of methodologies came in and
  481. dissuaded people from spending time on this.
  482. How do we then use that to help persuade people to do this?
  483. I think we need to welcome in the vibe coders.
  484. What do you guys think about this?
  485. 100 percent.
  486. No question about it.
  487. There's just this growing group of people who are able to participate in coding in
  488. this dawning era of AI assistance.
  489. And these are people that have kind of been locked out up until now.
  490. And I see in them the ones I meet so much hunger and curiosity and desire.
  491. And also as a numbers person, I see a growing denominator, just like more
  492. potential participants, more people to collaborate with.
  493. And I just kind of wonder how can we capitalize on this moment?
  494. And I'm I feel like I don't know what it will look like, but I sense that there
  495. must be the GitHub moment coming for vibe coders where they all are able to not
  496. just make something in private on their computer with Claude, but share and
  497. collaborate with each other.
  498. There's a moment of connection that's happening that must happen, I think, or
  499. logically follows.
  500. And I think that there's a way in which GitHub is part of why we had our peak and
  501. this smaller niche.
  502. And I think that's why I'm thinking of it as the GitHub moment for vibe coders.
  503. And I can't wait for it to get here.
  504. No, I couldn't agree more.
  505. I mean, how many projects have I looked at where I said, oh, gee, I wish it did
  506. this thing a little bit differently, or I wish it supported the CMS that I use, or
  507. I wish it could do this, but I don't have the time or the inclination or I don't
  508. know this programming language, or I don't feel like I've got the chops to do a
  509. pull request for this from GitHub, so I'll just live without it, or I'll do
  510. something else, or worse, I'll make my own.
  511. And so what vibe coding can do, I mean, vibe coding is, it gets better every
  512. couple of days.
  513. I mean, it is already a spectacular at doing software development.
  514. And one of the things it's particularly good at is reading an existing kind of
  515. code repository and making changes, including making security changes, you
  516. know, making speed changes, just sort of looking at it and looking to look at the
  517. Git history and understand why things were the way they were so it can be
  518. careful about what it does.
  519. And so the idea of, you know, fixing a repo that's out there so that it works
  520. better for your newsroom, or let's say somebody takes a look at your, at Ben at
  521. your code for, you know, that wraps data wrapper and says, well, how do I do this
  522. with flourish? Well, you can vibe code, you can have cloud code change so that it
  523. works with the flourish API instead of the data wrapper API, or let's say that
  524. data wrapper changes this API, because if it's undocumented, that means I'll be
  525. going to change it without telling you.
  526. And then when that happens, it's something that really probably you can vibe
  527. code. Now, interestingly, there is some concern about this in the open source
  528. community. So there's a developer who created a system called vouching, where
  529. they are so concerned about vibe coded pull requests on GitHub, that there's now
  530. a system where you can vouch for other developers. So you can say, well, I know
  531. that Simon is a real developer, he didn't vibe code this, so you should let him do a
  532. pull request to your code base. Whereas no one knows Scott, so don't take his
  533. code because he probably does vibe coded that. So it's kind of interesting that
  534. there's sort of, you know, antibodies being deployed to stop this. But I agree
  535. with you, Ben, I think that it's going to open it up to all sorts of really smart
  536. technical journalists who may not be able to code at the level that you would need
  537. to do a real pull request, who can add really critical features that you just
  538. don't need or you don't see, but but actually make your stuff better.
  539. It's like back in the day, back when we all started, there was a kind of opening
  540. up almost democratization of day journalism because anybody can make a
  541. chart. And now you've kind of got the same thing happening again, right?
  542. There's a new technology, it's opening it up. Anybody can do it. It does change how
  543. we do what we do, but it's an incredible moment, I think.
  544. Without a doubt, they'll be good and bad, but you know, who cares, you know?
  545. Yeah, I guess I'm still the hopeful one. You know what I mean? To me, I find
  546. myself at this AI moment, feeling, using the language of my new employer, feeling
  547. kind of bullish. You know, I feel kind of up about this. And my hope is, is that
  548. this next AI chapter is going to turn that curve back around that we made in
  549. our story. And we're going to see more people sharing in the years to come.
  550. Ben, thank you so much for joining us. Great to have you.
  551. Thank you for having me.
  552. Thanks for listening to the Data Journalism podcast. We're hosted by Alberto Cairo,
  553. Scott Klein and me, Simon Rodgers. Special thanks to Lauren, Chris and the whole
  554. NikeR team for all of their support with this episode.
  555. Our producer is Aussie Linus Goodman.
  556. The music you can hear is the sound of data made with two-tone, an app that turns
  557. numbers into tunes. This week, we used Google searches for the term "data
  558. journalism" since 2004. Do you have a new data driven story or project coming up?
  559. Tell us about it at dayjournalismpodcast@gmail.com and we might even feature it
  560. on the show. Subscribe to make sure you never miss an episode.

Downloads

Recording audio · Timestamped transcript