1 paper · 1 filter
Jacob Dineen, Silei Ren, Muhao Chen +2
In security-sensitive applications, language-model agents are often required to coordinate without disclosing confidential information. Yet repeated interactions may also let ordin…