Search papers, labs, and topics across Lattice.
1
0
3
4
Agentic LLMs can be taught to refuse harmful actions with up to 50% greater success, even zero-shot across diverse models and tasks, by explicitly learning when *not* to act.
