Search papers, labs, and topics across Lattice.
1
0
3
4
Turns out, those fancy open-weight LLM safeguards are easily bypassed with simple jailbreaking tricks, proving pre-trained nastiness is the real problem.