I feel like this can be trivially worked around by fine tuning from the initially copyrighted weights. They’re not demanding copyright, they’re just keeping them secret. If Meta keeps releasing high quality open source models, I don’t expect the closed source models to have an advantage for long.
Comments
I feel like this can be trivially worked around by fine tuning from the initially copyrighted weights. They’re not demanding copyright, they’re just keeping them secret. If Meta keeps releasing high quality open source models, I don’t expect the closed source models to have an advantage for long.