| ▲ | broken-kebab 21 hours ago | |||||||
Can you elaborate? I'm not sure I'm getting what is the point here. As I said, legal system rarely deals with 100%, it's not a requirement to make a law enforceable. Not telling the particular law is good or bad, just that "you can't certainly distiguish between human and LLM output" is not a strong argument against it. The conflict emerges when e.g. you are attributed words you have never said, or presented in a fake video, and there ways to argue about such cases in court other than just trying to analyze bit-by-bit, or doing histograms. | ||||||||
| ▲ | Xelbair 20 hours ago | parent [-] | |||||||
Simple - you need to assign blame, and you need to do it in a fair way. Any assingement of blame of using unauthorized LLM generated content could be shifted to external contractor(preferably outside of EU's jurisdiction) where you yourself had no awareness of it. Contract of course explicitly forbids LLM content and requires labeling. This is and extremely basic loophole that gives enough plausible deniability to any sane publisher/content farm. Now you need to prove they were aware of LLM content being generated, as they cannot rely on LLM content detection methods due to high amount of false positives. Even easier if you publish content from people writing under pseudonyms/anonymously. It devolves into he said she said kind of situation, and burden of proof is on accusers side. Basically depending on enforcement this system either: - only punishes small players that cannot afford to setup such chain - is used as an arbitrary excuse to shut some outlets if the system is corrupt | ||||||||
| ||||||||