Search papers, labs, and topics across Lattice.
1
0
3
LLMs can still be easily fooled by simple prompt rewrites because current adversarial training doesn't adequately cover the data distribution, but a new method using diffusion models closes this gap.
