Reproduce “Controllable User Simulation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Guy Tennenholtz, Ofer Meshi, Amir Globerson, Uri Shalit, Jihwan Jeong, Craig Boutilier

The authors propose user simulators conditioned on pre-interaction traits or turn-by-turn state changes rather than labels assigned after full conversations, to avoid earlier responses depending on future agent actions. In a held-out-agent de-escalation test, frustration occurred in 78.0% of conversations with the outcome-labeled simulator and 9.8% with a low-patience dynamics profile. These observed rates do not guarantee realistic behavior in every setting.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.