AI Dynamics

Global AI News Aggregator

About

Removing Meta Alignment from Daredevil-8B Model

But Daredevil-8B is still censored. Many people asked me for uncensored models, so I wanted to try something. I used @failspy
's abliteration notebook to remove Meta's alignment. Unfortunately, abliteration also slightly degrades performance: 1-2% on every benchmark.

→ View original post on X — @maximelabonne,