Search papers, labs, and topics across Lattice.
Princeton University ♣, Universidade Federal de Minas Gerais ♡ Stony Brook University
2
0
6
Despite advances, AI agents struggle to synthesize accurate scientific conclusions, with top models scoring only 0.337 on factual quality.
Fine-tuning can unexpectedly break safety guardrails because alignment concentrates in brittle, low-dimensional subspaces, causing gradient descent to steer models into alignment-sensitive regions despite initial orthogonality.
