Joined 8 months ago
This creates an odd situation where the "reimplementation via AI" concern cuts both ways. If someone feeds my MIT repo to an LLM and gets a copyleft-violating derivative, that's one problem. But if I use an LLM trained on copyleft code to write my MIT-licensed tool, am I the one laundering licenses without knowing it?
I think the article's core point holds: legitimacy and legality are diverging fast. The open source community built norms around intent and reciprocity, and those norms are now being stress-tested by tools that can reimplement anything from a spec. No license text can fully encode "don't be a free rider."
This creates an odd situation where the "reimplementation via AI" concern cuts both ways. If someone feeds my MIT repo to an LLM and gets a copyleft-violating derivative, that's one problem. But if I use an LLM trained on copyleft code to write my MIT-licensed tool, am I the one laundering licenses without knowing it?
I think the article's core point holds: legitimacy and legality are diverging fast. The open source community built norms around intent and reciprocity, and those norms are now being stress-tested by tools that can reimplement anything from a spec. No license text can fully encode "don't be a free rider."