WorkFactory: Turning Professional Intents into Executable Work for Scalable Agent Training
What it is
WorkFactory is a framework that generates diverse, executable professional tasks from high-level intents to create training data for working agents. It uses information-level planning and probe executions to ensure task coherence and reliability, producing 112,686 tasks across 612 occupations to fine-tune the WorkPilot model family.
The problem it solves
The lack of diverse, coherent, and executable professional tasks with reliable expert demonstrations for scaling supervised training of working agents.
What you could use it for
- Generating large-scale synthetic datasets for training general-purpose working agents
- Fine-tuning LLMs to perform complex, multi-step professional tasks
- Creating reliable expert demonstrations for supervised fine-tuning
Why it matters
Enables the scalable training of agents that can handle diverse professional workloads, significantly improving model performance on professional benchmarks like GDPval-AA.
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.