Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Well, define "easily". I played with a few "uncensored" models from Hugging Face and they all started to struggle at some point, especially around male/female dynamics.


Abliteration can be done with off-the-github scripts. It's not perfect, but works very well for the heaviest censorship.

Male/female dynamics may be in the corpus too, and even the reality may famously have some perceived biases.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: