Tensorwire
Products & tools · first seen 15 Sep, updated 15 Sep

One coordinate breaks abliteration on Gemma-3

1 outlet Gemma Llama

TL;DR: Refusal direction ablation, known as abliteration, using the standard diff. of means approach established in literature produced no feasible candidate for Gemma-3-12b whereas it did so fine for both similarily-sized Qwen and Llama mo…

Summary from LessWrong.

Coverage 1 article · 1 outlet

  1. LessWrong
    One coordinate breaks abliteration on Gemma-3