It seems so easy for just one poisoned model (trained with test data) to infect a ton of finetune model mixtures...it could happen without intention. Under thi

It seems so easy for just one poisoned model (trained with test data) to infect a ton of finetune model mixtures...it could happen without intention. Under this scenario, the ones that achieve the top performance are the closest relation to the poison model?