Resources
Jul 12, 2023 |
53 min
|
1.8k views
|

Why RStudio is now Posit (J.J. Allaire | Posit CEO) - KNN Ep. 158

Today, I had the pleasure of interviewing J.J. Allaire. J.J. is the founder of RStudio and the creator of the RStudio IDE. He is an author of several packages in the R Markdown publishing ecosystem including rmarkdown, flexdashboard, learnr, and distill, and also worked extensively on the R interfaces to Python, Spark, and TensorFlow. J.J. is now leading the Quarto project, which is a new Jupyter-based scientific and technical publishing system. In this episode, we learn about why RStudio has now repositioned itself as Posit, how it maximizes its open-source nature as a B Corp, and how J.J. as an open-source advocate views the private nature of many LLMs. I really enjoyed this conversation, and I hope you will as well!

Posit - https://posit.co/

Podcast Sponsors, Affiliates, and Partners:
- Pathrise - http://pathrise.com/KenJee | Career mentorship for job applicants (Free till you land a job)
- Taro - http://jointaro.com/r/kenj308 (20% discount) | Career mentorship if you already have a job
- 365 Data Science (57% discount) - https://365datascience.pxf.io/P0jbBY | Learn data science today
- Interview Query (10% discount) - https://www.interviewquery.com/?ref=kenjee | Interview prep questions

Listen to Ken's Nearest Neighbors on all the main podcast platforms!
On Apple Podcasts: https://podcasts.apple.com/us/podcast/kens-nearest-neighbors/id1538368692 (Please rate if you enjoy it!)
On Spotify: https://open.spotify.com/show/7fJsuxiZl4TS1hqPUmDFbl
On Google: https://podcasts.google.com/feed/aHR0cHM6Ly9mZWVkcy5idXp6c3Byb3V0LmNvbS8xNDMwMDQxLnJzcw?sa=X&ved=0CAMQ4aUDahcKEwjQ2bGBhfbsAhUAAAAAHQAAAAAQAQ

MORE DATA SCIENCE CONTENT HERE:
🐤My Twitter - https://twitter.com/KenJee_DS
👔 LinkedIn - https://www.linkedin.com/in/kenjee/
📈 Kaggle - https://www.kaggle.com/kenjee
📑 Medium Articles - https://medium.com/@kenneth.b.jee
💻 Github - https://github.com/PlayingNumbers
🏀 My Sports Blog - https://www.playingnumbers.com
🗣️ 66DaysOfData Discord Server - https://discord.com/invite/4p37sy5muZ

image: thumbnail.jpg

Transcript#

This transcript was generated automatically and may contain errors.

You get in a lot of these, like there's like language wars and oh, we used to use this and now we use this, or this is so much better than that. Any programming language that's popular has like a really good reason for being.

It's funny, I was listening to another interview with John Carmack, who did all the Doom and Quake and all the id stuff, and more recently worked on the Oculus. And he's legendary for getting every single ounce out of the hardware, doing crazy stuff with like video buffers and assembly and total maniac. And the interview, he was like, I kind of need to like realize that sometimes like Python is great. There's a lot of cases where it just doesn't matter. You know, the best programming language is the one that works generally that you're currently using.

I'm very agnostic in how I think about all these tools. I think any tool that gets adoption, I have respect for. And I'm kind of curious, like, well, why is this getting adopted? And why do people use this and care about it?

This episode of Ken's Nearest Neighbors is powered by Z by HP, HP's high compute, workstation grade line of products and solutions. Today, I had the pleasure of interviewing JJ Allaire. JJ is the founder of RStudio and the creator of the RStudio IDE. He's an author of several packages in the R Markdown publishing ecosystem, including R Markdown, flexdashboard, learnr, and Distil. And he also worked extensively on the R interfaces to Python, Spark, and TensorFlow. JJ is now leading the Quarto project, which is a new Jupyter-based scientific and technical publishing system, which I'll be doing a project on shortly.

In this episode, we learn about why RStudio has now repositioned itself as Posit, how it maximizes its open-source nature as a B Corp, and how JJ, as an open-source advocate, views the private nature of many LLMs. I really enjoyed this conversation, and I hope you will as well.

JJ, thank you so much for coming on the Ken's Nearest Neighbors podcast. You bet, happy to be here. Yeah, you've done such awesome stuff in the open-source communities, and obviously working with R, and now transitioning into Posit. I'm so excited to hear what your journey's been like, and also maybe a little bit about what the future of open-source, as well as the continuation of a lot of technical concepts through the platforms you're building. You got it, yeah.

JJ's introduction to data

First of all, welcome. Second, I'm interested in how you first got interested in data. It might have been a series of careers ago, but to me, that is one of the kernel questions. Absolutely, it was a long time ago, actually. I was 13 years old, and I was a huge baseball fan.

Like a lot of 13-year-olds, I watched a ton of baseball on TV, read all about it in magazines, and I got exposed to a writer named Bill James, who many listeners have probably heard of, but not necessarily everybody. So he wrote this series of books called the Bill James Baseball Abstract, where he used data analysis to understand the game a lot better. And one of the things he did was he actually exposed the fact that a lot of the conventional wisdom about baseball, how players were valued, how games are won and lost, was actually wrong. And a lot of the reasoning that went on was just basically based on people's intuitions and subjective experience, but not based on data.

And that was just a huge, and it turned out Bill James, when he started, was a heretic. He had a 28-page hand Xeroxed newsletter. It turned out he ultimately revolutionized all of baseball. If people have seen the movie Moneyball, it's based on Bill James' work. And now sabermetrics is an entire discipline, and all the sports now are very quantitative. And that was really in large part due to Bill James.

Anyway, he got me to see, wow, if all these experts running around saying these things that are not true and are so easily debunked by using data, that that's a really fundamental thing that is clearly probably going on outside of sports. And my real sort of main interest was political science and politics and economics. And I went to study that in college. And I was very attracted to the empirical research side of that because it was the same thing. Wow, we're making these really consequential public policy decisions. Are we informing those decisions based on data?

And so in college, I really focused a lot on quantitative analysis, I ended up using a bunch of tools that some of which are still in wide use today, like Excel and tools like SAS and Stata and Jump and SPSS. And I was really fascinated by it. I really enjoyed that work.

And that took a detour in my career where I kind of got into computing. And I spent 10 or 15 years building other types sort of development tools, programming tools, web servers, authoring tools, et cetera. And when I kind of got done with all that, I really was very keen on working in open source software. And I found out about, I heard about R, which was an open source statistical computing system. And that was very, very compelling to me. Because that was sort of where my heart started in my professional life was like, how can we use data to understand the world better and make better decisions?

And R at the time was early. And it was designed by statisticians. So it's extremely well suited to its purpose, but it also needed a lot of tooling. And so I started working on open source tooling for R. So that's kind of how I got into this.

So I love that. Obviously, I have a massive connection to the Bill James story. I mean, my work is in sports analytics. And the seminal person in golf is Mark Brody. I got to meet him a couple of times last year. And the way that we view that game has completely transformed more recently than with Bill James, right? I mean, for those that don't know, one of the cool things about sports analytics is everyone looks for a singular metric that you can evaluate everyone against apples to apples. And in baseball, that was wins above replacement. In golf, that's strokes gained. And it completely changes how these sports are viewed.

I want to share an anecdote about that. Just remind me when we get on to talking about some of the other stuff, about literate programming and notebooks and things, I've got a Bill James-related tie-in based on an interview he did very recently that I want to share. So just make a note of that. Okay, I made a note of that.

What drew JJ to open source

So we're going to get into all of that, which I'm very excited about. But before that, I'm interested in what particularly drew you to open source. You know, that's something that obviously has been a massive benefit to my career. I think every data scientist out there that's worth their brain assault is using open source tools. And it's really cool for me to see how involved and how much you grow in open source communities in your own. Can you just tell me about like, what was the kernel that drew you?

Well, part of it is I think in science in particular, it's very important that people are able to share and reproduce results with each other and innovate on methods. And I think that open source does those things very, very well. If I do some analysis and I want you to evaluate it, you don't need to purchase a software. If I want to reproduce the analysis in 10 years, I don't need to make sure I still have a license or make sure the software still runs. If a group of people like in the sports analytics community wants to do a bunch of innovation on methods, then they don't have to lobby a proprietary software vendor and say, oh, hey, pay attention to sports analytics, it's really important. Could you support this, the following new cutting edge methods?

So a lot of things, I think, specifically about science and inquiry and trying to develop pools of knowledge in communities are really, really well served by open source. So I think that's one of the reasons why it's really important in data science and science. But for me personally, I also like the idea of making contributions that are durable, contributions that can be around in several decades that aren't sort of tied to the fate of a given or the fate or the whims of like a single company. And so that's very attractive. I think a reason that I wanted to work in open source is these durable contributions. But it turns out, I think, for data science and science, it's a pretty fundamental requirement.

I think a reason that I wanted to work in open source is these durable contributions. But it turns out, I think, for data science and science, it's a pretty fundamental requirement.

Yeah, something you've mentioned in some of the talks you've given is that your goal isn't just to be around this year or next year, it's to be able to produce something that's still working in a hundred years. Yeah, that's right.

I think that philosophically, that's so dramatically different than what you described might happen in a private organization or maybe in a startup. Can you talk to me about how, in some settings where things are built privately, they might actually break down because of the incentive structure. If you look at the history of proprietary software, typically there are like transition points where software is like phased out, formats are discarded. Customers are forced to migrate from one thing to another. Customers are abused because they have a dependency on the software and really have no choice but to keep doing it.

So I think it's really important for, again, this concept of durability and this concept of trust that we're trying to create something that you can rely on for a very long time. That hundred-year idea is this is trustworthy and durable. But it's also, I think, the idea that if we can create a company that has a synergistic relationship between sort of building a business and being able to invest at scale and creating open-source software, that company is worth having around for a hundred years. Traditionally, software companies don't work that way. There are definitely examples of open-source software companies.

But even like the classic example that people point to is like the paragon of successful open-source software companies like Red Hat. And I just saw a news item yesterday where Red Hat's gonna make it much, much harder for sort of non-Red Hat distributions like Oracle Linux or Amazon Linux, or there's various flavors that are derived from Red Hat. They're gonna make that much harder. So it's, you know, the ability to do something that people can trust for the long-term, I think, is the big driver behind that.

Posit as a B Corp

So how can you create a system for that, for example, at Posit to be able to do that for the long-term? I mean, go ahead. No, no, I think that we talk about it as sort of a virtuous cycle where, you know, our mission is to create open-source software that's usable by anyone regardless of economic means, that's extremely useful to a whole broad range of people. And so we do that, you know, that's our packages, our IDEs, our web frameworks, reporting tools, all the different things we create. And by making them open-source, they tend to get very widely adopted. And what happens is that people in large enterprises adopt them and then want to do more with them. They basically want to, like, integrate them with their authentication systems, integrate them with their compute grids, and, you know, integrate them with their sort of, like, audit requirements. And there's lots of things that come up in a larger setting when you're scaling. And our idea is, what we've done, is we've sold products to larger enterprises who are starting to take these platforms seriously. And that builds a business for us, and that internalizes us to reinvest in open-source and ultimately create more opportunities for other folks to adopt the tools. So that has worked pretty well for us.

I'd like to take a quick second to thank one of our sponsors, PathRise. Job hunting can be daunting, particularly in today's unpredictable world. PathRise matches you with a dedicated mentor in your field to provide tailored support throughout your job search journey. From optimizing your resume and application process to mastering your interview skills, their expert guidance prepares you to secure your dream job and to negotiate a competitive salary. The cherry on top is that you don't pay until you secure that high-paying job that you've been aiming for. Don't hesitate. Visit pathrise.com slash Ken G to get matched with an expert mentor today.

This episode of Ken's Nearest Neighbors is brought to you by Z by HP, HP's high-compute, workstation-grade wire products and solutions. Z is specifically made for high-performance data science solutions, and I personally use the ZBook Studio and the Z8 workstation. I really love that the Z workstations can come standard with Linux or WSL2, and they can be configured with the Data Science Software Stack Manager. With the Software Stack Manager, you can get right to the work of doing data science on day one without the overhead of having to completely reconfigure your new machine. Now back to our show.

I wanted to talk about this a little later, but since we're on topic, I think it, I want to touch on it now. What is, I notice that a Posit is a B Corp. Can you explain a little bit what that means, but also how that sort of ties into what we're talking about? Yeah, a B Corp is a benefit corporation, and in short, in very short, the nature of corporations in the United States is that they're really, their exclusive ultimate responsibility is to their shareholders. So they have a fairly narrow definition of what's ultimately important, and that can lead to companies, again, as I said, selling themselves, becoming abusive, and that's problematic because companies really have a bunch of stakeholders. They have their customers, the broader community, their employees, and what a benefit corp is is it sort of broadens the idea of what a corporation is for, and it broadens the accountability to say it's accountable to multiple groups of stakeholders. Further, public benefit corporations define a public benefit, a specific mission,

that they're also obliged to pursue. In our case, it's open source software for science. So it's a new type of company that has been developed really in the last 10 or 15 years, and we made the transition.

It actually didn't exist in a robust form when we started our studio, but we made the transition, I think, a few years ago, three or four years ago, to being a benefit corp, partly because we wanted to have it in, that idea of durability and who we're accountable to, we wanted to be in our charter, and we wanted people to understand that we were trustworthy and we meant what we said.

Meaning what you say, I think, is so important, and it also ties in very interestingly to, I think, technology communities, particularly engineers, because the philosophy of open source has always been interesting. How do you make that work when there's these individual, individual capitalistic mottos of each individual person? And I think software engineers feel so much like, hey, this isn't how I want it, why don't I just do it myself? Screw it, I'm gonna build it. And you feed into that, but you also feed into the ownership and the community aspect.

And psychologically, to me, there's very few case studies like that where people will actually willingly work together, they build something largely for free. Yeah, it's amazing. Yeah, there's been writing about, there's a little bit of a gift economy dimension where I think we are fairly advanced industrial civilization that's fairly affluent, and so people start to turn to other considerations like what can I give, what can I contribute, how is my work used? We're sort of above the providing basic food and shelter.

And I think what you're saying, a lot of data scientists and engineers that I interact with kind of feel that way. They actually do want to create something bigger, bigger than themselves, and make a contribution to other people in their field.

Yeah, I mean, realistically, I see that making a meaningful contribution to the field, for a lot of people I've talked to, it means a lot more than, for example, money does. It does, it does. Just being able to say, hey, look what I've enabled these other people to build. That absolutely does, yeah.

I see that making a meaningful contribution to the field, for a lot of people I've talked to, it means a lot more than, for example, money does.

Especially I mean for- I view, go ahead. No, you go ahead, go ahead, sorry. I was gonna say, I view a lot of the, for example, the content that I create in very much the same way, where I'm probably not gonna come up with the most, anything that inherently creates a tremendous value for humanity over the course of my career. I might be selling myself short, but I feel like that's pretty realistic.

I mean, I think letting people, having conversations with all the people you have conversations with, opens up a lot of kind of intellectual pathways for a lot of people, so. Well, that's what it is, is if I can inspire someone that actually does make that difference, to be more interested in the field, if one of my videos helps them get started, to me that is just as valuable. Absolutely. And it's in my own way, right? Absolutely, yeah, yeah.

JJ's path to founding RStudio

So let's go back a little bit more to your specific story. Yeah. You obviously fell in love with open source. You started building in this. How did you get involved with RStudio? How did you get to the part where you're leading Posit now?

Yeah, so RStudio started, I think it was like 12 or 13 years ago. And originally it wasn't started to design, intended to be a company. It was just, hey, let's make an IDE for R. You know, I had done programming tools. I'd done some IDEs. I'd done writing tools. And so I said, I think R could really benefit from having a nice IDE. And so I actually ended up working on that with Joe Chang, who was the co-founder. He's the CTO now.

So we worked on RStudio. We worked on that for a couple, three years. We released it. It kind of started to get momentum. We ended up meeting Hadley Wickham, who sort of became another co-founder, who's now our chief scientist. He had been working on, you know, ggplot and dplyr and a bunch of really, really popular R packages for doing data analysis and visualization. And so we said, hey, let's all work together. Let's all see what happens if we all work together. And that was sort of the birth of it being more than just a single project, because by that time we had four or five people and we kind of had to decide, you know, are we going to try to make this a small company? Are we going to try to grow it?

And we decided that there were a lot of other folks we knew who could make contributions and it would be really valuable to try to grow the company. And so, and then we teamed up with Tarif Kowaf, who's someone who's now our president. So he's been here for also for over 10 years. And he kind of came on with the responsibility of figuring out, you know, how could we create this virtuous cycle? How could we create and grow a company that supports open source? So that was the genesis of RStudio.

And Joe actually came up with a product called Shiny, an open source web development server, web development framework called, for R, for sort of data-centric apps called Shiny. And so, you know, we had RStudio, we had Shiny, and then we started working on, that's okay, let's see if we can create some sort of enhanced commercial version of these things to sell. And that ended up working out. And so we get the company grew slowly at first, you know, to 15 and then 20 and 30 and 50. And, you know, we ended up, it ended up working really well.

Becoming Posit: expanding beyond R

And we kind of ended up, I would say about three years ago, we had sold our products to lots of customers, but we observed that every data science team that we were selling into was doing, we started off with like these sort of very R-centric groups but basically everybody was using R and Python. Or, you know, maybe there'd be some exclusively R and some exclusively Python groups, but the vast majority of groups were using both. And we were selling, you know, enterprise software to help those teams be productive and deploy things. And so it was like sort of not, it didn't make sense to sell them an R-only thing. It just didn't compute at all.

So we started, we added support for Python to sort of all of our products. And that was well-received, but it's quite difficult for customers to get their head around buying like Python solutions from a company called RStudio. It was like, just really like, wow, how am I going to explain this to the CIO? Like we're buying, you know. And so we decided a couple of years ago that we really need to change the name of the company so that people understand the kind of the bigger breadth of what we're doing.

And also at the same time, kind of change some of the focus of our open source work where we had really been really primarily aimed at doing open source in the R ecosystem. And we decided that when we do new projects, we need to make them multi-language. And so we ended up doing Shiny for Python. That was announced like a year ago. Quarto, which is a project that I'm working on, was Python from the beginning. So we're essentially trying to evolve the company to one that creates open source data science software for both Python and R, sells commercial products for both Python and R. And so we renamed the company Posit. That's about a year ago to reflect that.

You know, renaming the company, I think that that is so in line with the open source nature. You know, it's, hey, we're a community, we're trying to build things that are most valuable to people, and we're gonna take this in the direction that the group is seeing this, or like that the users are seeing this. And I feel like that's something that most companies in a more like traditional profit-motivated like zone or would really struggle to do, because it's like, oh, this is our identity, how do we change?

It's hard because I mean, our studio had a really terrific brand, had a really good reputation. And the easiest thing to do is just keep turning that crank and see what happens. And I just felt like, I actually felt like bigger picture, like going back, we'd become a benefit corp. We'd set up this sort of capital structure and group of people capable of delivering a much broader set of things. And I wanted to go for that. You know, even if it was, it's obviously quite inconvenient to change the name of the company. Now, all of a sudden you've gone from, everyone's heard of you. Even most Python users have heard of RStudio. No one's heard of Posit. You know, that's kind of a hard reset, but I think it's worth it. You know, it'll take time, but eventually we're hopeful that folks will have heard of us. It'll be really, really have been a very worthwhile transition.

Where R thrives and why

Let's talk a little bit about maybe where R is most valuable or who the R users are and how that fits into the broader community. Because to me, that's something, you know, for example, in sports, R, especially with the, I think it's the NFL, R is the prominent language that they're using. Maybe it's changing a little bit, but there's specific communities that seem like R is their first choice. And some of them are begrudgingly adopting other tools. But, you know, why is it so prominent in some areas and then some other areas are just like.

I mean, some of it is, I would say that R plays very well. Certainly, if you look at like people who are like in an undergraduate or graduate setting, if you're studying software engineering, you are gonna gravitate to Python. If you learned about machine learning in your computer science department, you're absolutely using Python. If you are in a company and you're doing data engineering, you're using Python. If you are in a statistics program, an undergraduate or graduate statistics program, you're very likely using R.

So there's a lineage of individuals and subfields where if you're coming from a heavy statistics orientation, you're likely using R. And conversely, if you're coming more from a computer science data pipeline, data engineering, you're using Python. So that's one explanation of where that comes from. The other thing though is that many people who do work with data do not principally identify as a software engineer or software developer. They identify as like, I'm a biologist, I'm an economist, I'm a sports analytics person. And so a lot of them, it gets driven by like, what's the package ecosystem look like for doing the work that I care about?

And so you will see subfields. My guess is in the NFL, some of it's historical, but there's probably packages that are extremely useful and would be really annoying to do without. And that makes it more sticky there. I mean, you can conversely see in deep learning in Python, like all the places, all the ways in which you wanna do deep learning and all the libraries are all in Python. So you're just in Python if you're doing neural nets. And so I think there's some of that that goes by field. We see a huge preponderance of R in insurance, and in pharma, those are places that are like, actuarial work or research in a clinical setting is like very statistics intensive. And so you see it more in those settings. So I think part has to do with educational background, part has to do with package ecosystem, how much statistics is a big part of things versus machine learning. So those are some of the factors.

I have to think in terms of factors, open source actually plays a really large role in the communities. I mean, just as you said, with the NFL, I think NFL scrape R was one of the best ways to get NFL data. And so if everyone's scraping their data using R, then probably will analyze it using R too. That's right. Exactly. And to me, that's such a cool way that open source can drive adoption, unity, those things. Well, that's what I was talking about the long term. There's no way, like Stata was not gonna come up with like an NFL scraping feature. They just wouldn't have done it. Whereas the community says, hey, we care about this, we wanna do this. So here's NFL scrape R.

Yeah, well, there's other things like that where you have these communities that are pushing towards sort of this more, more specific use cases. And I think one we talked about previously was PCA. Yeah. Like principal component analysis in R in the libraries that are available there, I find to be so much more effective than the ones in Python. It's just more interpretable. And if the brunt of your work is doing PCA, maybe in a recent- Yeah, you're gonna end up in R. Public Result. That's right.

LLMs and language agnosticism

So that's what we see. Yeah. Amazing. Well, one thing I think is really cool, and we talked a little bit about this offline, is that I see the world becoming more and more programming language agnostic. So with the expansion of LLMs, I now can just take my Python code and say, hey, turn this into usable R code. Or I could say what I want and say, hey, give it to me in JavaScript or Julia or Mojo when it's more widely accessible.

You know, I think that there's a huge democratization of programming languages, and it's gonna be more about the tooling or the libraries or the accessibility that we can get across any of these platforms. Do you have any kind of broader thoughts on that? Yeah, I mean, I definitely agree that using LLMs lets you bootstrap into a new language in a way that's really pretty revolutionary, you know? Because a lot of, you would normally spend two or three days like fumbling around trying to figure out which way is up with a new language. But you know, you can say, hey, I wanna do whatever, you know, I'm gonna do a PC analysis, I wanna visualize it this way, and it'll give you roughly what you need to do, and then you're gonna probably have to tweak it and fix some problems and whatnot. But I think in terms of accessibility of different programming ecosystems, it is pretty profound.

You know, I think, I will say that I do think that there's a lot of like skill that's developed by data scientists or programmers

ends up being fairly deep knowledge of the language and language ecosystem and vagaries of it and workarounds. And so I think, you know, the investment people make in a given language is still pretty sticky. You know, I don't know that people are gonna be like just flitting around between languages because you are giving up a lot when you switch languages.

So yeah, we'll see how that, I definitely feel like people are really feel empowered to try things that they don't understand that well because they can get going really quickly with LMs for sure. Maybe it'll be more like, you know, people who maybe currently use SPSS or Excel or something that's a little less code oriented will feel more empowered to say, hey, wow, I got started, I created a notebook that does kind of like what I was doing in Excel, it all works and it's really flexible and now I'm off and running.

Yeah, I think the sort of cold start problem is very difficult for a lot of people. It's, you know, hey, I know that this would probably look better and be more adaptable if I was using JavaScript, but I've written five lines of JavaScript code in my entire life. Yeah, yeah. All five of those being Hello World. Yeah, yeah, yeah, yeah, yeah, yeah, yeah.

And to me, it's like, well, the barrier to entry to be able to be competent in that, but then to expand that knowledge is so much lower than it used to be in the past. And I think I'm super interested in where that goes. I also think in particular, coding tools being accessible to people who are doing research is really interesting. It is, it is.

You know, you were talking about political science. Political science research is now enabled by the ability to code more effectively, to apply more quantitative rigor, because, you know, Python isn't as much of a barrier, R isn't as much of a barrier, any of these tools aren't as much of a barrier. You know, you're really focused also on research and publishing with Quarto. Tell me a little bit more about that.

Quarto and communicating data

Yeah, so actually, the open source project that I've been working on for the last few years, it's called Quarto. And really, it's a platform, or platform's probably the wrong word. It's a framework for publishing reports and research that have data at their center. So it's for creating reproducible documents. It's for creating artifacts from data analysis that go beyond just like a single metric, but, you know, a suite of data visualizations, narrative that accompanies visualizations, notebooks that you created that dig into a topic and make them available as a website, or make them available inside a presentation. So Quarto is really about communicating.

And I wanted to bring a little bit of the motivation for this. And, you know, it really, one of the fundamental capabilities of Quarto is that it can take Jupyter notebooks and can publish them in a huge variety of formats. So you can create entire books or, you know, a blog or a presentation, a PowerPoint presentation, a Word document, a really attractive searchable website. So you can take Jupyter notebooks and actually bring them to a lot of people.

That's one of the fundamental capabilities. And I think if you go back to like what's interesting about Jupyter, and we were talking a little bit earlier about Bill James and Sabermetrics, and there was an interview he did, I think about a month ago with Michael Lewis. And he said that he, I had not heard him say this, but he has become sort of disillusioned with the contemporary sports analytics movement because there's so folk, you mentioned that wins above replacement, like we need to get one number, you know, that represents the player's value.

And Bill James's work always had context and narrative, and, but, you know, consider this. And he sort of became, he says become disillusioned because it's always just about like getting it, reducing to a single number. And he said, there's just a lot more going on when you're thinking about players and teams and careers and all that. So I think that what Jupyter notebooks allow you to do as a publishing medium, and by extension, Quarto, is marry narrative to visualizations, to those reductions that you get in terms of numbers and things. So oftentimes there's just a lot of context assumptions, qualifications, other branches to consider in a rigorous analysis. And part of the focus of Quarto is helping you communicate those.

So I think that what Jupyter notebooks allow you to do as a publishing medium, and by extension, Quarto, is marry narrative to visualizations, to those reductions that you get in terms of numbers and things.

That's so interesting to me, because I see that so much in my own work is that every data scientist wants to qualify their work. We say this holds true when X, Y, Z. Right, right. Every business person just wants, hey, what is this number to tell us how to work? What is this number? Yeah, and there's so much storytelling, there's so much background in that that is very difficult for us to convey because data scientists, engineers, we tell the story from the beginning. And business people, they want the story from end to, basically they just want the end or the conclusion. And that's not, you can't make great decisions with just the end. You need to understand the context.

I recommend, listen to that interview if you can. It's really fascinating, him kind of reflecting on his career and kind of where things have gone.

There's another worthwhile thing to read, which was Edward Tufte wrote a pamphlet, people are probably familiar with some of his books, the visual display of quantitative information, called the cognitive, it was, I think it's called The Cognitive Style of PowerPoint, called Pitching Out Corrupts Within. And he basically describes how trying to reduce decisions and analysis to like a single graph, a single three bullets, a single number leads to oftentimes like terrible consequences. That's true. Yeah, yeah, yeah.

So he actually showed, one of his examples was he showed like the ill-fated space shuttle mission. There's like a PowerPoint deck where they basically were talking about, should we do this or not? And there's some slide, it's just got like four bullets and there's a bunch of other detail that needs to be explored and it's just, it ended up on one slide. Like a really important thing was just sort of meshed over and the idea of like let's reduce it down. Yeah, and they are, you know, the whole, you know, the Wall Street value at risk, you know, it's like we have our value at risk metric and that's gonna tell us like how exposed we are. But there's qualifications on value at risk, there's like tail events that happen and it's not, that's not actually your value. Your value at risk is not $10 million, it's actually like $200 million for the tail events. Well, you know, tell me more about that but that's not what people wanna hear, they just wanna know what's our value at risk, you know? Yeah, what should we do? I'm really passionate about that, of bringing that narrative and storytelling into communicating about data and that's a lot of what Quarto is about.

Jupyter notebooks and their place

Yeah, you know, you mentioned how Jupyter Notebooks are perceived. I mean, when I first started using Jupyter Notebooks, I was like, why would anyone do this? Because I was coming from more software engineering, I was saying, okay, if I wanna productionize my code, I shouldn't use a Jupyter Notebook, I should just use a Python file, I should build it into a pipeline or a WebP. And then I started teaching online and I said, oh goodness, Jupyter Notebooks are perfect for this, I can run the cells, I can go and individually talk about what's going on, I can tell a story, I can provide context at the relevant places, I can interface markdown so it looks good, I can put in these links.

And this, to me, is an extension of that, is okay, well, I have this really great analysis, what if I wanna publish it in a book and any of these types of things? That's right, yeah. To me, that is just making all of this so much more accessible to everyone who might not be able to go in and run a notebook and understand what's happening. That's right, yeah. You don't have to be able to run the notebook in order to get a lot of value out of it. And you don't have to present even the code in the notebook, you can present just the visualizations and metrics and commentary and things and links and things like that.

Yeah, so notebooks, they're very good for teaching, they're very good for storytelling, they're very good as an interactive environment for just playing around with things. So they have their place and then also more traditional workflows based on Python source files certainly have their place too.

I mean, that is the biggest thing, or one of the biggest things that I'm taking away here is that there are so many things out there that have their place. You know, there's a place for people to program in R and it is the best tool for this specific use case. There's a place for me to use Jupyter Notebooks, which is in teaching, there's a place for me to just, you know, run Python notebooks. One of the things I always kind of reflect on is like you get in a lot of these, like there's like language wars and oh, we used to use this and now we use this or this is so much better than that. Any programming language that's popular has like a really good reason for being. You know, you can make fun of, you know, PHP for web development, but like it has a really, there's a really good reason why PHP exists and there's a group of users for whom it's amazing. You can make fun of Java, you know, but there's like a really like important use case at CAP.

So I'm very agnostic in how I think about all these tools. I think any tool that gets adoption, I have respect for. And I'm kind of curious, like, well, why is this getting adopted and why do people use this and care about it?

Yeah, I think it's also important to note the other side of that is that every single programming language tool that we use, there's something inherently thought about it, right? There's some way that another language does something better. That's exactly right. Well, what is your use case? If your use case means that the deficiency in that language, that doesn't matter for that use case, then use this one. If that's gonna be a massive problem, use this other one.

Yeah, it's funny. I was listening to another interview with John Carmack, who, you know, did all the Doom and Quake and all the id stuff and more recently worked on the Oculus. And he's legendary for getting every single ounce out of the hardware, doing crazy stuff with video buffers and assembly and total maniac. And in the interview, he was like, yeah, you know, I kinda need to realize that sometimes Python is great. Like, I don't need, there's a lot of cases where it just doesn't matter to soak all that performance out. You know, I have a script that runs in three seconds. I run that script twice a day. Like, why do I care? Now, obviously, if he's building Oculus and he's trying to maximize, reduce the flicker, yes, that's one story. So even he was just like, yeah, I have a renewed appreciation. He was trying to learn as much as he could about machine learning and AI and stuff. So he ended up doing a bunch of Python. And he really, I think, got more of an appreciation for like, yeah, a lot of people should just be using this.

What makes products succeed or fail

So, you know, you mentioned John Carmack, correct? Yep, yep, right. I mean, he's building out, you know, games. Yeah. He's building out systems. Yeah. You've been involved in a lot of projects, you know. And to me, there's something that we talk about offline where projects, some of them fail for no good reason. You have a good tool, you have a good community, all these things, and they, it's a little thing sometimes that make those things fall out of over, not be as successful as you think. Yeah. Can you tell me a little bit more about the little things that are-

Yeah, I mean, John Carmack's a good example in that a lot of the early success he had with like, you know, Beyond Castle Wolfenstein and Doom and Quake is he did, he would stack like 120 little things on top of each other that added up to like, wow, I didn't know you could even do that. That's unbelievable. And I think, so that's on the one side is like, a lot of times the whole of, is this a product I'm excited about? Is this a product I want to use? Actually has to do with all these little things that accrue together to say, wow, this just makes sense. It just works.

I think similarly, you can have like a really good idea that, you know, is really well executed, but then there's other reasons why it might fail. Like a good example, back to the, you know, the Oculus like thing that you put your phone in the goggles. They did some crazy engineering to make a Samsung phone be able to do VR. It just turns out people don't want to strap a thing on their face. You know, they just, people just don't want to do that. So, you know, or, you know, there could be things like, early on, we're seeing more of this now, but you know, like a cloud-based IDE, like an IDE that runs in the browser. You know, 10 years ago or 15 years ago, people did that and it was like, well, that's really inconvenient for me. You know, really good job, great idea, well executed, don't care.

And so there's a little bit of like a timing thing sometimes. So it's, there's just so many factors that affect whether a product is successful. Some of which are very much in control of the people creating it and some of which are out of their control. So. Yeah, it's a very weird thing. I relate this almost to dating. Yeah. Where you could meet the perfect person, but the timing isn't right. Yeah. And it's like, well, it should have worked. It should have worked, yeah.

And that's how I always feel like, I always feel like whenever I create a new product, I'm just like, I'm, I don't know if this is gonna, I don't know. I mean, I can try to stack everything I can in my favor, work on relevant problems with the right tech at the right time and get all the details right. But I sort of have this open-mindedness to like, maybe it won't add up for people. Maybe they won't, maybe they won't connect with it. You know, it's just, you can get surprised a lot of the time, so.

So a little bit of a segue. I mean, you talked about adding up all these little small things and creating something that people didn't think was possible. I think one of the best case studies of that right now is with open AI, particularly with, you know, GPC4 or whatever they're training now. Yep. You know, they're not using technology that's necessarily new and transformers. Yep, yep. But I think it's Sam Altman who was talking about how, hey, like the advances we've made are a lot of these little, little kind of things.

Yeah, it's not a brand new model architecture or, you know, it's actually, we just found all these spots where, ooh, that's an unfortunate side effect or consequence, can we mitigate that? Or, oh, you know, just, I think he said literally something like that, hundreds of little things, you know?

Yeah, well, so I'm interested, you know, you being a major proponent of open source. Yeah. Regarding open sourcing large language models like that. Yeah, yeah. What is your philosophy? You know, they're, I believe, a nonprofit, if I remember correctly. Yep, yep. But, you know, is that enough for them to have that at all?

Well, I mean, OpenAI has not open sourced their models and nor did they actually, I think Sam Altman would say, we don't think the models should be open sourced because they are, in some cases, they're kind of like weapons grade, you know?

The ability of a large language model to produce disinformation at scale is a weapon.

The ability of a large language model to produce disinformation at scale is a weapon.

And so I think that changes the discussion about open source. I think it's, we don't have open source, like nuclear power plant designs, you know? There are things that are dangerous and need to be used with responsibility and may even need to be used with some oversight. And I think if GPT-4 isn't that, you know, future iterations of it certainly probably will be. So I'm a little more cautious about the idea of open sourcing.

I would agree, like, I think if people haven't seen it, I think watching the congressional hearing that Sam Altman did and Gary Marcus from NYU as well was there. And he said, you know, for language models like bigger than GPT-4, we might want to have licensing for that. Like, we might need to know about that. So it's, you know, and things like we might need to have a testing, a sort of third party that can verify that the model can't self replicate, for example, stuff like that.

So I think, you know, in light of the B Corp in the light of creating a fair ecosystem, regulation to a certain extent helps the first movers in a space. It helps OpenAI, it helps Google because they already have the infrastructure. If someone was creating a startup, someone was building something that they wanted to impact, that is overhead or infrastructure that they wouldn't have access to. That's right, yeah.

And to me, that's the natural conflict that you're kind of also facing, right? Is that we want this to be open and free and to be able to create valuable things for people, but inherently we're only helping the elite few to continue to move towards additional profits, which can be a little bit scary. You're talking about, yeah, in the case of if you say, we're gonna do licensing of LLMs and we're gonna make it so it's fairly expensive to experiment with these things. Yeah, that's a tough one.

And I know in Sam's testimony, he made it clear like we don't think this should impact open source and startups. He was trying to say, we're not trying to do the regulatory capture lock-in thing.

But some of that is inevitable and I think it has to do with like how dangerous these models can be. Like I think saying like we wanna foster maximum innovation, okay, but if it's like we want maximum innovation in like development of like new bioweapons, we wouldn't say that. Like let's have maximum innovation. Let's have as many people as possible getting in there, making bioweapons, trying out different types of bioweapons. No one would be like, oh yeah, it's gotta be open.

So I think that's an extreme example, but as the capability of these things improves, they do become more dangerous. And so I don't think it's just an absolute, like we wanna have as much innovation as possible. We wanna have as many applications as possible. So it's like at some point we might say, yeah, it's actually kind of good that there's like four people who do this. We might.

Yeah, I tend to agree with you. I think playing devil's advocate a little bit.

These large language models inherently are not a weapon. You're going in and saying, hey, this is a bio, like we know this is a non-biome weapon doing research. There's this weird gray area that we operate in where I don't think that there are many technologies that are as ambiguous, maybe nuclear is, where, okay, if we make advances in nuclear, we can power things more efficiently, but it also has this destructive capability. Classic dual use, yeah.

To me, it's fascinating because there are so few dual use technologies like this that have so much- Bio is interesting and things like gain-of-function research in order to understand better how pathogens work also has the downside of- The downside of? Yeah, so there's a few, but yeah, you're right. There's not that many.

I think you're right that, I think the most dangerous thing about LLMs right now is their ability to produce disinformation at scale, I think, to so disharmonious political and cultural dialogue in social media and other places is like the worst thing that they can do right now.

I don't think that none of these other kind of like really more physically frightening scenarios are in play or already close to in play, but we do need to think about it, the fact that Sam Altman said, we need to make sure they can't self-replicate, right? That is looking toward a future where, huh, oh, they're gonna self-replicate, what?

Distrust, disinformation, and the future of media

It's interesting. This is my personal take. I'm interested in a world where we can't trust any news. I think it's an inevitability. Yeah, I think that's probably right. And I don't know if that's necessarily the worst thing because right now we're in this weird limbo state where the people or the models are good enough to trick most of us, but on the other hand, there's some news that we can vaguely trust still.

If we live in a world where we can't trust any news, that changes our relationship with media. It's like maybe we go to specific people, maybe we go back to what journalism was. I do wonder if it leads to just like everyone just says, well, I'm not going online anymore. It's just like I'm literally interacting with machines all the time and I'm getting like deep fakes and I'm just like, I'm not, like I'm just gonna have like a book club here and in our village and just talk to each other. I mean, that wouldn't be terrible. And you can't trust the book if it was published recently. You can't trust the book, yeah, exactly.

No, I know. I think some shared sense of reality is a helpful basis for like civic dialogue, but we are, I think, gonna be beyond that if we aren't already beyond that.

Yeah, my optimistic hope is that distrust online means that communities and in-person things have a resurgence in value. Yeah. Because, you know, you can, there's something inherent about being able to like reach out and talk to someone. Absolutely, yeah, no, I think you may be right. That may be where we end up and that would not be a bad thing. I completely agree.

Yeah, my optimistic hope is that distrust online means that communities and in-person things have a resurgence in value.

Closing thoughts

So JJ, those are all the questions I had. Do you have any final thoughts? Anything you, you know, any way people can learn more about you? Anything along those lines? Happy to share those.

Yeah, I mean, I think you can put the URL for the Quarto project. I think it's very much worth, if you're a data scientist, it's absolutely worth checking out. That's, I'd say the main thing, again, that I'm focused on now and I'd want people to go take a look at.

Excellent. This was amazing. We touched on so many fun and exciting topics. So I'm glad to be able to share this with the audience. Absolutely. Thanks very much for having me on.