←back to thread

Heretic: Automatic censorship removal for language models

(github.com)
745 points melded | 1 comments | 16 Nov 25 15:00 UTC | HN request time: 0.206s | source
Show context
syntaxing ◴[16 Nov 25 19:31 UTC] No.45947712[source]▶
Amazing. I’m eager to see what the results for GPT-OSS is like. It’s a great model but the “safety alignment” ruins it
replies(1): >>45947835 #
1. pehtis ◴[16 Nov 25 19:47 UTC] No.45947835[source]▶
Specifically for GPT-OSS I had great success with this: https://old.reddit.com/r/LocalLLaMA/comments/1ng9dkx/gptoss_...