Safety
On the Hardness of Junking LLMs
DGX agentarXiv:2605.05116v1 Announce Type: new Abstract: Large language models (LLMs) are known to be vulnerable to jailbreak attacks, which typically rely on carefully designed prompts containing explicit sem
Knowledge catalogue
arXiv:2605.05116v1 Announce Type: new Abstract: Large language models (LLMs) are known to be vulnerable to jailbreak attacks, which typically rely on carefully designed prompts containing explicit sem
arXiv:2605.04954v1 Announce Type: cross Abstract: Per-instance algorithm selection (PIAS) takes advantage of complementarity between a set of algorithms by deciding which algorithm to run on a given i