But what if they deleted the training set also? Actually, it was probably the other way around, first delete the illegal training data, then the model that contains the proof that they had illegal training data.
The volume of text outputs should massively narrow down the weights, expect to a near identical model, as similar as you going to sleep and waking the next day.
But what if they deleted the training set also? Actually, it was probably the other way around, first delete the illegal training data, then the model that contains the proof that they had illegal training data.
The volume of text outputs should massively narrow down the weights, expect to a near identical model, as similar as you going to sleep and waking the next day.