The Cathedral and the Bazaar

Open source deserves a lot of credit for creating AI. Is it the future?

Writers and Coders are the same in that they both sit at a keyboard for hours, days, months, building up elaborate creations with only letters and symbols.

But that’s where the similarities end.

Here’s gobbledegook of a few programming languages I’ve dabbled in through the ages: FORTRAN, APL, Pascal, C, dBase, Python, JavaScript, PHP. I’ve barely scratched the surface. 1000s of programming languages exist, along with hundreds of databases and millions of frameworks and libraries that glue it all together.

You only need to know one language if you’re a writer.

You can argue that composing something out of that one language has all the subtlety and complexity of the entirety of Computer Science and I wouldn’t disagree. I try hard every week to write something intelligent in the English Language.

Programming languages and all that surround them are syntactically strict and brittle. Put a semicolon in the wrong place and the big red X in your code editor tells you right away you messed up. So in that way, it’s easier to write code.

That last category of CS — frameworks and libraries — holds the central difference.

Ask any developer — they’re positively thrilled to find a library / framework to build on. Every modern software product is 99% built by artfully stitching together blocks of code written by someone else.

Plagiarism is built in.

Plagiarism is anathema to a writer, which is why AI is universally hated by every author and artist. AI sucked up their hard work and now spits it back out in endless, soulless verbosity. They’re right to be pissed. There are over 100 active lawsuits filed by creatives against the AI companies. Copyright law is clearly on their side, so it will be interesting to watch how these turn out. We’ve already seen a few billion dollar settlements.

AI got good at writing code the same way it got good at writing language. We know where it got its knowledge of language — by hoovering up everything on the Internet, but how did it get so good at writing code?

A clue: the very first of all these lawsuits filed against AI companies was actually by coders, not writers or artists. It’s Doe v Github.

Enter the Bazaar.

In 1997, Eric S. Raymond, champion of the growing Open Source Software movement wrote an essay called The Cathedral and the Bazaar ¹

The illustration from the cover of the Raymond’s book is this 1913 painting by Lyubov Popova titled Composition with Figures. It was released into the public domain in 1999.

Raymond’s essay opposes two styles of software development. The Cathedral is the traditional approach. A single company assembles a crack team of craftsmen (programmers) and they carefully and painstakingly design and build the perfect Cathedral from the ground up.

This was the way to build software since IBM created the market, and although Microsoft was an upstart, programming for PCs not mainframes, the approach was the same.

Enter Open Source Software (OSS). The crown jewel of OSS at the time Raymond wrote his essay was a web server the fast growing Internet ran on. It was called Apache, and it was built by a largely volunteer community of developers.

The name Apache tells the story: ‘A patchy’ web server. A patch is something you release to fix a bug in your software. The OSS community was relentless in finding and patching bugs, one of the hardest parts of maintaining commercial software.

While Apache was big, OSS was building something even bigger — something the world largely runs on today: an operating system called Linux. A project overseen still today by one man — Linus Torvalds. Linux is open source, built and maintained by thousands of developers across hundreds of companies.

The Bazaar was winning and Microsoft and its Windows monopoly hated it.

I joined Microsoft in 2001. A few months later Microsoft’s then CEO Steve Ballmer called Linux a cancer on intellectual property.²

Microsoft was on a Jihad against Open Source. In my first company-wide meeting, Ballmer brought onstage an exec hired to lead a new compete team against OSS. I sat through a dramatized roleplay of our pitch why Windows was better than Open Source.

With my roots in Berkeley and the Bay Area, I thought the whole thing was silly and a bummer. Ballmer’s Jihad turned the minority of computer science academia still ambivalent to Microsoft, against us.

Years later, Satya Nadella would come in to replace Ballmer after Microsoft had missed the mobile wave and now was in danger of missing the cloud wave. In 2014, just months after becoming CEO, Satya headlined a customer briefing in San Francisco in front of a slide that read:

Microsoft ❤️ Linux.

I liked working at Microsoft a lot more when Satya started running the show.

We’re close to the seminal connection between Open Source and AI. Remember Doe v Github?

That’s the thing: GitHub.

Git was invented by the daddy of Linux himself, Linus Torvalds. Legend has it, Linus wrote Git in a furious few weeks to manage all the Linux source code contributions pouring in from the open source community. Linus says he names all his projects after himself and git is Brit slang for a nasty person.

Problem solved, Linus gave the code away and GitHub grew into the largest repository of source code in the world. Satya led its acquisition by Microsoft in 2018.

AI trained on GitHub.

10s of millions of code repositories, 100’s of millions of lines of code. AI learned to code at the same time it learned to write, and in a very interesting discovery, the two are complementary. AI models trained on code along with language have quantifiably better critical thinking abilities.

That’s the beef in Doe v. Github. AI learned from every line of code every open source developer ever wrote and it didn’t ask for permission. Alas, due to the model of open source, judges in the case have found little basis in coders claims because open source explicitly allows re-use.

A ton of intellectual capital goes into writing elegant code. While it differs from writing, it’s still supremely creative and requires a lot of skill. Just like authors and artists, developers should get credit for their work.

But it’s telling that it’s one lawsuit out of over 100. By-and-large developers are on board with AI as their latest coding assistant and it has completely democratized the pool of people who can build with software.

The interesting turn is Open Source is back in the news as a model for AI. Think of the frontier labs: OpenAI, Anthropic and Google as today’s Cathedral builders. Is this the right way to go? Wouldn’t it be better if, like Linux, this foundational technology was available freely for the community to build on and improve?

Now a new version of Open Source is being bandied about and this time not between companies, but between countries. The latest Open Source advocate comes from China.

On July 17th, Xi Jinping, China’s President, opened the World Artificial Intelligence Conference in Shanghai telling attendees to:

seize this rare, historic opportunity to encourage open-source

hmmm

The day before Xi’s keynote, Chinese company Moonshot AI released a new AI model called Kimi K3. Kimi knocked Anthropic’s latest offering, Fable 5, off some key leaderboards, a first for a Chinese model.

Anthropic called foul, claiming Moonshot AI ‘distilled³’ Kimi from Fable.

Moonshot AI went partway open source with Kimi, releasing the model’s weights, so anyone can host and tweak it (although they can’t see how it was trained, etc.).

AI has gone fully global and China is the biggest Bazaar in the world, but is China proposing the same kind of Bazaar that built Linux? Is Anthropic CEO Dario Amodei the new Steve Ballmer railing against Open Source?

For writers, the Bazaar was never for them.

For coders, writing with AI is no longer optional, it’s just the job and they’re used to a job that is defined by change.

What about China? They desperately want to catch up with the US frontier labs so this open source positioning is geopolitical and more different than it is the same.

Like the saying goes, the present sure rhymes with the past, although Dario ain’t exactly Steve and Chinese open weights aren’t the same thing as open source.

Who knows, maybe a Satya is waiting in the wings.

Thirteen years ticked by for me at Microsoft between Steve’s Jihad and Satya’s 180 on open source and I know one thing — it ain’t gunna take another thirteen years to sort it out this time.

1. You can read the full text here: https://archive.org/stream/CathedralAndTheBazaar/cathedral-bazaar_djvu.txt

2. Chicago Sun-Times interview June 1, 2001 exactly five months after my first day.

3. Affectively a form of reverse-engineering.