150,000 web navigation tasks were created by researchers from Carnegie Mellon University and Amazon to facilitate internet-scale training of LLM agents. The dataset is intended to reduce reliance on human annotations for agent training. It was uploaded to Hugging Face by the author 'data-for-agents' on May 28, 2025.
Use Cases
- Training agents for web navigation based on the described tasks.
- Benchmarking agent performance on internet-scale tasks.
- Developing methods for automated agent training without human annotations.
- Studying agent behavior in simulated web environments.
Strengths
- Contains 150,000 tasks, providing a substantial scale for training.
- Created by researchers from Carnegie Mellon University and Amazon, suggesting academic rigor.
- Specifically designed to facilitate internet-scale training of LLM agents.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Last updated 2025-05-28 21:21:42; freshness should be verified.
Provenance
- Source
- Carnegie Mellon University, Machine Learning Department and Amazon
- Collection Method
- Created by authors of the paper 'Towards Internet-Scale Training For Agents'.