Reproduce “$τ$-Rec: A Verifiable Benchmark for Agentic Recommender Systems” 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
Bharath Sivaram Narasimhan, Karthik R Narasimhan
τ-Rec evaluates movie-recommendation agents through conversations, catalog tools, and programmatic checks of constraints and policies. It controls when user preferences are disclosed and measures pass^k across trials. Across nine configurations, pass^4 is generally below pass^1; DeepSeek V4 Flash reaches 0.571 pass^1 with max thinking and 0.383 pass^4 with high thinking. Results cover a 60-task, 153-movie benchmark, not deployment.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.