- cross-posted to:
- [email protected]
- cross-posted to:
- [email protected]
It’s hard for me how he recognizes all the (social) issues with LLMs and still comes to the conclusion that he isn’t in part responsible for them.
It’s hard for me how he recognizes all the (social) issues with LLMs and still comes to the conclusion that he isn’t in part responsible for them.
Unknowingly using potentially copyrighted code from an LLM in a FLOSS project opens us up to being attacked via being sued for infringement, which most small devs can’t afford to fight, especially in the US’s current extremely pro-corporate courts.
It’s putting a target on your back for down the road when it becomes profitable for patent trolls to use AI to try to scan for copyrighted code on public code bases.
Big tech companies could do the same to squash an open source competitor.
We know for a fact that LLMs have been trained on all available open source codebases.
Do we know that there is a relevant amount of closed source code in the training data?
Additionally to the other answer: Not all published source code is open souce, and I’m pretty sure, the scrapers didn’t stop to read the licence.md files on github
Corporate devs who use a non-self hosted LLM for developing their closed source apps are sending that code to the LLM’s servers, where it can be used as further training data (and considering how many of the corporate AI’s have purposefully trained their AI on pirated copyrighted content (like… A lot of them are doing that), it follows that they would train their AI on user prompts as well).