No LLMs, please
The various large-language models (LLMs1) from pick-your-favorite-company are getting pretty fancy and are capable of some impressive work. Open source projects have been adapting to these new tools, with some like the Linux kernel accepting LLM-assisted contributions while others have forbidden them.
Given that, I wanted to publicly state that I do not and will not use LLM tooling in either personal or professional projects. All my bugs and typos are 100% organic and human-generated. 😄
Why?
In short, LLMs draw from a poisoned well. The various companies scraped the entire Internet with impunity and blatantly violated copyright law. These tainted sources were then used for their training and model refinement, meaning everything they produce is tainted. I cannot ethically use these tools, until and unless specific training sets that respect copyright and licenses are made available and are easily auditable.
But, hey, these companies are producing trillions of dollars of economic activity, so who cares that they’re overwhelming us all with slop and garbage? We sure have come a long way since “You Wouldn't Steal a Car”.
-
I’m specifically not using the term “AI”, since what’s currently available is very clearly not even close to actual intelligence. ↩︎