I just learned that Oracle, which is shoveling unbelievable amounts of cash into getting everybody to use LLMs, also has a policy forbidding contributors to OpenJDK from using LLMs to generate code, citing code quality, security, and legal concerns:
-
they rightly point out that machine-generated code cannot be copyrighted
That's not quite the issue here. The reason Oracle requires contributors to own the rights to code they contribute is so that they are in a position where they can grant co-ownership to Oracle, by signing a legal agreement (the "OCA").
The worry is that an LLM may regurgitate from its training material verbatim portions of a work copyrighted by a third party, that the contributor has no ownership over -- and therefore cannot transfer it Oracle, nor ensure compatibility with the GPL.@clayfoot @ubernostrum But the overwhelming concern for the projects maintainers is that they don't want to get drowned in PRs full of "plausible-looking code, with plausible-looking tests, which is nonetheless incorrect or, even if it is correct, is poorly designed and therefore difficult to maintain", a.k.a. slop.
OpenJDK has a very thorough reviewing process for contributions, which ensures a very high degree of quality, but puts a large burden on reviewers. This makes the project particularly vulnerable to being drowned under a wave of LLM-fuelled garbage contributions.
-
J jwcph@helvede.net shared this topic
