Search papers, labs, and topics across Lattice.
1
0
3
Achieve stable and interpretable persona control in LLMs by routing contrastively-trained personality vectors aligned with the Big Five facets, outperforming prompt engineering and other baselines.