Search papers, labs, and topics across Lattice.
1
5
3
6
MLLMs can be made significantly safer without sacrificing performance by disentangling risks in multimodal inputs and using RLAIF, outperforming even GPT-4V by 16% on safety benchmarks.
