Sounds like "unlearning" is really just "reduce the probability of sampling" from some latent "learned space" and quantizing reduces the efficacy of the slight

Sounds like "unlearning" is really just "reduce the probability of sampling" from some latent "learned space" and quantizing reduces the efficacy of the slight change in sampling.