@jackcorleone19
AI/独立开发干货|实战笔记
Science
The edit reportedly changed just 0.016% of Qwen3.5-0.8B’s weights without gradient training in the researchers’ experiment.