Search papers, labs, and topics across Lattice.
1
0
3
Unlocking fairer vision-language models may be as simple as intervening in the sparse latent space of a sparse autoencoder, enabling targeted bias removal without harming performance.
