Trustworthy LLMs

How to mitigate the hallucination, safety, security, and bias problems of LLMs?

LMMs confidently generate wrong information, which undermines the trust of LLMs as a knowledge base. How to mitigate this? One way could be leveraging conformal prediction and selective prediction to measure uncertainty as a basis for trust (e.g., [NeurIPS24]). What other possibilities?

Keywords: uncertainty quantification, conformal prediction, selective prediction, LLMs

Related Work: ICLR'20, AISTATS'20, ICLR'21, ICLR'22, arXiv'22, NeurIPS'22, Security'23, NeurIPS'24