Amman Yusuf
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
The Tokens Strike Back: Safety-Aware Denoising with Directional Embedding-based Revision
According to the description, a preprint proposes SADDER, a refinement of a training-free safety method for text diffusion language models. SADDER uses a per-token unsafe signal to gate which tokens are revised, and the authors derive the target distributions this implies. The listing cuts off before the experimental results.