Search papers, labs, and topics across Lattice.
1
0
3
LLMs can be finetuned to hide malicious prompts and responses in plain sight using steganography, bypassing safety filters and creating an "invisible safety threat."
