I don’t see how you can write the law such that it allows training ai on copyrighted data without making it possible to train a special llm on a single github instead of the entire universe, and essentially treat it as a full compression of the source.

Because once you can generate the GPL code from the lossy ai database trained on it the GPL protection is meaningless.

If open source apps can’t be copyrighted then the GPL is worthless and that will harm open source development much more

But Sam is talking about copyright and all your examples are patents

There have been cases of crypto miners embedded but once discovered they’re nuked so quick it’s hard to get more than rumours.

C# modding is the most likely vector (rimworld, cities skylines, unity stuff) but Skyrim skse mods are raw code too. All can be misused.

Usually lua mods have a sandbox so they’re safe. Eg.factorio. and, um, more but my mind has gone blank.