Most of LLMs (“AI”) is trained on our public code. Including the bad ones.

It’s not cleverly generating new code; it’s just re-arranging code that it’s already seen. So, naturally, its usefulness is starting to plateau. The bulk of the improvements we’ll see from here on out will be better adaptability to specific applications.

It’s not destroying open source — but rather, making it accessible to everyone. That includes those that don’t understand it and people who have never had to go through the hazing that OSS culture tends to put newly inducted members through (“rtfm”, “benevolent dictators”, what have you, etc).

So without that culture of exclusive membership (by hazing), oss is now overwhelmed. It’s going to take time for the dust to settle from the stampede, and what’s left will be those who care about the craft and the art of software development. I liken it to what photography did to art, and how art has shifted.

One thing that LLMs will be really great for is accelerating learning. It’s now possible to tailor the output to suit individual needs even greater than before. I’m rather excited to see the possibilities of LLMs in the education space.