Comment by gpt5

Slightly tangent question - they said that they have protected the public test set with a strong copyleft license to prevent training private models on them.

Does it actually work? Isn’t AI training so far simply ignores all license and copyright restrictions completely?