The Changelog: Software Development, Open Source - Free Dolly, GitHub Accelerator's cohort, improving Tailscale via Apple’s open source & what the heck are passkeys?! (News)

Episode Date: April 17, 2023

Kara Deloss announces GitHub Accelerator's 2023 cohort, Databricks releases the first open source, instruction-following LLM, fine-tuned on a human-generated instruction dataset licensed for research and commercial use, Mihai Parparita writes how he improved Tailscale thanks to Apple's open source & Neal Fennimore asks and answers the question: Passkeys: what the heck and why?!

What up nerds, I'm Jared and this is Changelog News for the week of Monday, April 17th, 2023. Good news everyone! Our companion Changelog Newsletter now has 100% more Dark Mode support, thanks to the many, many readers who hit reply and let us hear about it last week. Don't know the power of the dark side. Okay, let's get into the news. GitHub's Accelerator launched their first cohort last week. First time hearing about it?

GitHub Accelerator is a 10-week program where open source maintainers receive an initial sponsorship of $20,000 to work on their project, paired with guidance and workshops from open source leaders, with an end goal of building durable streams of funding for their work. Funky dope maneuver! Cara DeLoss announced the 2023 cohort, which spans 20 projects and 32 participants from all over the world. Check Cara's blog post for the full list, but here's a few of the repos that will sound familiar to Changelog listeners. HTMX,

Carson Gross' library, which makes AJAX, WebSockets, and more available directly in HTML. Nuxt, the Vue community's framework for building web apps. And Dataset, Simon Willison's open-source multi-tool for exploring and publishing data.

Hello, Dolly! Hello, Dolly! This is Louis Dolly. I plan to give you back where you belong. Earlier this month, the team at Databricks released Dolly, a large language model trained for less than $30 to exhibit chat GPT-like human interactivity, aka instruction following. Two weeks later, they released Dolly 2.0, the first open source instruction following LLM fine-tuned on a human-generated instruction dataset licensed for research and commercial use.

Licensed for commercial use? What, what, what? That seems like a big deal to me. What this means is that any organization can create, own, and customize powerful LLMs that can talk to people without paying for API access or sharing data with third parties. Databricks' announcement post goes into detail on how they made it, including how they crowdsourced 13,000 demonstrations of instruction from their own employees. Fascinating stuff. This is, in my opinion, a huge step toward bringing production-grade LLMs to all, but just how good is Dolly 2 when compared to results you'd get from bailing up to an API from one of the big providers?

The Databricks team says, quote, As a technical and research artifact, we don't expect DALI to be state-of-the-art in terms of effectiveness. However, we do expect DALI and the open-source dataset will act as the seed for a multitude of follow-on works, which may serve to bootstrap even more powerful language models. Tailscale engineer Mihaly Parparita writes how he improved Tailscale thanks to Apple's open source. Mihaly says, quote,

The ability to peek under the hood of the operating system can be a powerful tool for debugging. A recent blog post by Daniel Jalkut struck a very familiar chord with me, reminding me of times I spent poring through WebKit internals at my previous employer. I've been able to continue that pattern at Tailscale, now more focused on Darwin, the operating system at the core of macOS and iOS, its kernel, and its userland tools, end quote. He goes on to tell a tale of how he fixed two network interface-related bugs by reading Apple's implementation of ifconfig. ifconfig, that's what I call it at least, is a tool for configuring network interfaces

on Unix systems. Some people probably call it interface config or IF config. Mihaly's post served as a solid reminder to me just how cool it is to be able to peek under the hood of the platforms we build upon. To learn, to adapt, and sometimes to just get the job done. Up next, a solid write-up by Arnaud Laray at Postman on how to enhance your API-first design process. What makes this post different from the rest? It's sponsored content. Yes, for the first time ever on Changelog News, we're making money. Now, if you despise sponsored content, have I got a deal for you. Check it out at changelog.com slash plus plus. On the other hand, if you're

cool with sponsored posts, as long as they're actually good and interesting, let's get into it. Arnaud says API-first design should be reusable, interoperable, modifiable, user-friendly, secure, efficient, pragmatic, and crucially aligned with the organization's goals. He then explores how you can achieve an API-first design by integrating the following five actions into your design process. One, use natural language to analyze and challenge needs. Two, observe context and identify constraints. Three, fully describe and document your APIs.

Four, leverage existing APIs and guidelines. Five, integrate automated and human feedback loops into the process. Read Arnaud's full post for the details on those five actions. And for even more, check out the book they're writing on the subject. Thanks to our friends at Postman for putting out good content for the developer community and for sponsoring this episode of Changelog News. If your organization could benefit from a Changelog News sponsorship, please do get in touch. Pass keys. What the heck and why? That's the question asked and answered by Neil Fenimore on a recent CSS Tricks post. Neil says, quote, They were a main attraction at W3C TPAC 2022,

gained support in Safari 16, are finding their way into macOS and iOS, and are slated to be the future of password managers, like 1Password. They are already supported in Android, and will soon find their way into Chrome OS and Windows in future releases. End quote. Sounds like this is probably something you want to learn more about. In fact, we had an episode request a little while back asking us to do an interview with somebody from Apple on the topic.

I tried hooking that up, but no dice. If you want that to happen, let us know and I'll work on it even harder. Anyways, back to passkeys. In brief, passkeys are a specification that is built on top of web offend, which allows for public key cryptography to replace passwords. Instead of presenting something you know to prove your authenticity, you present something you have, which is a hardware device holding your private key. The challenge with this historically is that sometimes you lose that thing you have, which means you don't have it anymore, and you're locked out. Passkeys build on WebAuthn by providing cloud sync of your credentials

to other devices that you do have. Hence, iOS and Android play a big role here. That's as far as I'll go into it on the pod, but Neil's post goes into extreme detail. It's a great primer for anyone who's curious. That is the news for now. Of course, there are plenty of noteworthy things going on in the software world, which is why today's companion email has an additional 19 links for you to comb through. You can read it now by following the link in your show notes,

or get it delivered directly into your inbox by popping your email into the form at changelog.com slash news. Seriously, do it, do it. On this week's episode of The Change Log, Adam and I are joined by Zach Lada. Zach dropped out of high school after his freshman year to work in tech and had over 5 million people using his software by the time he turned 17. Not bad. He has since gone on to found Hack Club,

the program he wished he had in high school. Have a great week. Share ChangeLog with your friends if you dig it, and we'll talk to you again real soon.