The authors recast LLM unlearning as closed-form, null-space-constrained activation steering.
Null-space activation steering enables training-free LLM unlearning without utility loss
By steering forget queries off memorized answers and projecting corrections into the null space of retained activations, Nullify preserves model utility while erasing target knowledge at a fraction of the compute cost.
Top University
Wei Zhai · Xiang Liu · Qiang Huang · Rui Qian · Lemao Liu · Ziwei Li · +3 more
Fudan University · BEDI Cloud · Tianjin University · King Abdullah University of Science and Technology (KAUST) · Zhejiang University
Research Digest··3 min read
The authors propose Nullify, a training-free, non-destructive method for LLM unlearning that operates entirely in activation space during inference.
Why this paper
From Fudan University and 5 others
In one line
Nullify erases LLM memorized data via null-space activation steering, preserving utility without weight updates.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§