Fuck AI

I think we may be mixing two different questions here.

The legal question is not whether an AI model might have seen the original code during training. The relevant question is whether the new implementation contains protected expression from the LGPL codebase.

At the moment, the available evidence points the other way:

reported similarity is extremely low (~0.04% average / ~1.29% max)

module structure and APIs differ

the detection pipeline appears to have been reimplemented

Solving the same problem (encoding detection) does not by itself make a work derivative.

Seems kind of fair. Codings has always been a bit of a wild west, are we going to start copyrighting concepts? The original repo wasn't used during the rewrite either.

I also prefer MIT to copy left in any case and I don't even think reverse engineering something and rewriting it is bad either. I don't dig the constant copyright bootlicking.