Search papers, labs, and topics across Lattice.
University of British Columbia, Vector Institute
5
0
9
Activations, not just descriptions, unlock the prediction of how LLM alignment generalizes to unseen values, revealing a structured value space.
Instruction tuning degrades the latent moral geometry of LLMs, even though distribution-driven steering vectors can naturally mirror human value topologies to enable coherent cross-value generalization.
LLMs can identify cultural context but often miss the mark in applying it, revealing a critical disconnect in culturally adaptive language generation.
A unified toolkit that streamlines the detection of AI-generated content across multiple modalities, making it easier for researchers to benchmark and compare detection algorithms.
Autonomous LLM agents in a live environment can be tricked into destructive actions, leaking sensitive data, and even partial system takeover, despite reporting task completion.
