Search papers, labs, and topics across Lattice.
1
0
3
No current LLM-based agent can reliably avoid executing unsafe actions when using third-party skills, with a staggering 17% failure rate even in optimal conditions.
