(I’m sorry. I can’t bring myself to say “artificial intelligence” non-ironically.)
I am super-optimistic about the advent of large language models and their related technologies. I am also freaking terrified of large language models. I feel like there’s a strong case to be made for both potential outcomes.
On the one hand, I’m really looking forward to June’s Apple Worldwide Developers’ Conference. Bloomberg's Mark Gurman reports (via MacRumors.com here for the stable link) that Apple is going to release their large language model. I know Apple pretty well, so I have reason to be enthusiastic that their take on the technology will be distinct and geared towards solving particular problems. Siri comes immediately to mind, but I’m mostly thinking about Xcode.
We’ve had Github’s CoPilot for some time now. It works by reading your comments and intuiting the code to perform the assigned task. It’s a powerful tool. But what can Apple do with a model tuned on tons of apps, and access to your whole project?
I’m envisioning a new version of Xcode that you can converse with about the giant app that you help manage with a team of five, twenty, or 500 developers. If you’re not a developer, it might be difficult to understand that apps can get pretty complicated! Having an on-board assistant that can identify where certain functionality lives, how something works, suggest improvements, highlight potential memory issues or race conditions (especially in this terrifying new world of Swift concurrency!), could make every developer dramatically more productive.
In this optimistic model, today’s developers use LLMs to simply do more, more-correctly. Like the promise of Swift itself, the result could be fewer bugs, more-consistent implementations, delivered faster. I would love to have that version of Xcode!
The Dark Side
But there’s another side. Today, Cognition Labs introduced Devin, “the first AI software engineer”. In the video attached to that post on X, Scott Wu (who may be positioning himself to be the last human software developer?) shows us their product, an assistant that does all the coding for you. Tell it what you want, and off it goes.
“Devin” is autonomous, using the web to find information to help it overcome roadblocks. According to the company, it resolves 13% of the issues it’s given. That number will surely go up over time.
You probably won’t be surprised to learn the company is funded by Peter Thiel, that arch-asshole libertarian whose goal, as far as I can tell, is to extract every dollar on the planet, solely so he can cackle evilly from his cloud fortress while the world burns below him? I guess?
Anyway, Devin is a good start. Because one fundamental problem Thiel and people like him suffer, is the fact that software needs to be written, and the people who do it are super expensive. It’s almost ridiculous how obvious this is: if we could get rid of the humans who cost so much to make the things that help him burn the world down, then we can make more shitty things, faster. Let a thousand Facebooks bloom. Let a million content farms rise to feed them lies and paranoia. Let a Cambrian explosion of software be thrown at the wall to see what sticks.
Let a colossal swath of today’s software developers go without a job.
So yeah, that’s what I’m worried about. And excited about. Guess we’ll see how it turns out!