Search papers, labs, and topics across Lattice.
3
0
5
Financial LLMs can now be rigorously evaluated against targeted risks, reducing critical false negatives in safety assessments from 28 to 12.
Style-eliciting prompts can unlock the latent stylistic information in writing representations, outperforming traditional LLM prompting methods in both style recovery and imitation.
Mapping LLM attack strategies onto a multiplex network reveals interpretable vulnerability clusters and dramatically improves red teaming efficiency.
