Comment by thisislife2 | Hacker News Reader

Comment by thisislife2 | Hacker News Reader

Parent Full thread

I completely share your sentiment about feeling irked about open source code being used to train commercial AI models. However, I think the battle is already lost - the nature of copyright and open source code philosophy (currently) means that there isn't any real way of preventing your code being used to train AI. Look at the legal precedents being set in courts where many of the BigTechs have actually pirated copyrighted media to train their AI, and the court has said "that's acceptable". (Ofcourse, the actual act of piracy - like Facebook did by downloading copyrighted material through torrents - may not be legal, but the courts may be lenient here too as there seems to be an undercurrent of government approval to do anything to win the "AI Race").

And, even if you move your repository somewhere else, can you really prevent anyone from uploading it to Github? To do so, you may have to create your open source license.

> However, I think the battle is already lost - the nature of copyright and open source code philosophy (currently) means that there isn't any real way of preventing your code being used to train AI.

Laws should make it a double-edged sword, make distillation explicitly legal.

Not much else they can do.

> Laws should make it a double-edged sword, make distillation explicitly legal.

Knowledge-distillation is already legal. Current case law says none of outputs of any model are protected by copyright, so one could use model outputs for whatever they want - including distillation. That is why AI companies resort to ToS clauses to block distillation and/or training competing models.

Its completely 100% legal, to ask LLM to write detailed spec of any closed source project based on the code. Then ask another LLM to implement it to make it public domain.

Its already proven based on previous specification laws.

Hilarious as it sounds its true. We can make “any” code a public domain woth LLMs.

Distillation is totally legal and I highly encourage it

Make all LLM output licensed as AGPLv3 - the license is viral, right ? ;-)

Unless you can actually prove what your model has ingested & if you have the license for it in court.

Even if you cannot do anything to prevent it you don't have to make it easy - and you can make it clear that you don't approve by boycotting GitHub over this.

Reply on news.ycombinator.com