Overleaf

We've got some very well-known companies under our remit.

Overleaf is probably the most well-known.

It's a 30 million researchers around the world.

And Figshare is part of the portfolio.

And so people also have this pushback about giving too much control to the frontier models.

And we've seen this case very close to my work is...

I mentioned Overleaf before as one of our products.

It's a collaborative writing tool for researchers.

She takes another sip.

Back in Sagan's apartment, Phoebe debugs code and edits Overleaf documents and folds paper cranes instead of curling in a ball and crying.

The dining table has turned into a field of cranes, with a small space cleared at the end to take meals at.

She calls her mother on Sunday mornings and tells her about how well her weeks have been going.

You know, for my papers that I write, most of my papers are written with my students and my collaborators, um, and we use a hodgepodge of different things.

We are usually writing our LaTeX documents using Overleaf, which is the standard way of sharing, um, LaTeX documents, which is the scientific typesetting system.

I've also-- I've recently-- You know, I've-- It took me a while since coming to Johns Hopkins, but I finally built up a s- a good-sized research group.

And so we're moving into the modern era with-- where we use Slack to communicate like real businesses do.

To you.

The problem is that AI co-pilot tools are available for many of the typical productivity tools used by scientists, such as GitHub and Overleaf.

It's become as seamless as autocomplete and spell checking.

So I just don't see how journals could realistically police such rules.

And then I wanna talk about ways of integrating kind of that process into something you might work on with a, with a collaborator.

So, like, Overleaf is the one that I'll talk about, but of course Dropbox, that would work potentially well too.

Um, and so we're gonna have these three parts, and they're really gonna be about kind of setting these best practices for thinking about the verification step and the iteration that happens kind of at a, at a, at a higher level.

One level of abstraction above the coding and the-

The way to think about it, I think the joke that everyone says is that really what it is, it's a committed set of historical versions of work that you've done.

And, you know, the obvious one that we've all done, right, is you have like [chuckles] workingcode_v1.dofile, right? And then you add v2_final-final, right? And obviously what you'd rather have is you have one file, but somehow you can kind of keep track of all the way that it's changed.

That's what you do, for example, when you do a Word document, right, is you do track changes and you might have stuff in the background, Overleaf as well.

Yeah.

I'm gonna now move to taking some of those prompting philosophies, but in a-- implementing them in a much more powerful way through s- tools that are designed for it.

So the way I worked until a year ago is mostly Overleaf, sometimes Dropbox.

All my co-authors and I would share a repository in, in this kind of, um, tool.

My sophisticated colleagues, uh, who do empirical work often use GitHub, which is, uh, the industry standard for software.

So one thing that, that you probably, um, you probably know is that, or many of you know, but, uh, I'll remind those of you who don't, is that you can connect these two.

Um, and the main reason that's useful is that GitHub is, uh, since it's the meeting place of software engineers, the software companies that the s-- and the AI native environments that are designed primarily for software engineers are great at linking to GitHub.

So GitHub can be sort of a waypoint, um, in linking our ac-- Uh, you know, you can bring it in from anywhere, but I'll show you how I brought it, brought my stuff in from Overleaf, um, through GitHub.

So here's the process.