Their open weights are what makes the model any good! You don't have their data and you don't have their training code. I'm sure their research will be replicated in time, but you can't just "change the weights" to make an equally good model that doesn't suppress anti-CCP info.
Comments