Tree Search for Language Model Agents introduces an inference-time best-first tree search algorithm that enables LM-powered autonomous agents to perform explicit exploration and multi-step planning within real interactive web environments. It is complementary to existing agents and demonstrates significant performance gains on benchmarks like VisualWebArena and WebArena when applied on top of models such as GPT-4o. The project includes a preprint, code, and talk, targeting researchers working on improving LM agent capabilities.
Tree Search sits in PulseGate's AI & ML category. Language models struggling with multi-step reasoning, planning, and environmental feedback in realistic computer tasks. It is built as an open-source project for AI researchers. The project is open source (MIT). It ships for the web.
Carnegie Mellon University builds and maintains Tree Search, and it first shipped in 2024. The project is developed in the open on GitHub with 223 stars. Key capabilities include Tree Search, Multi-step Planning, and Web Environment Interaction.
Summary written by a language model from the project’s public pages.
What PulseGate has recorded for this listing
Same category — not a similarity match