This story was originally published on HackerNoon at: https://hackernoon.com/why-macos-is-underrepresented-in-public-ai-research-datasets.
MacPaw Research explains why macOS is severely underrepresented in public AI datasets and introduces GUIrilla, a framework for scalable Mac UI exploration.
Check more stories related to tech-stories at: https://hackernoon.com/c/tech-stories. You can also check exclusive content about #macos-ai-training, #guirilla-framework, #computer-use-ai-macos, #macos-api-accessibility, #guirilla-task-dataset, #os-atlas-macos-coverage, #macapptree-python-library, #good-company, and more.

This story was written by: @macpaw. Learn more about this writer by checking @macpaw's about page, and for more stories, please visit hackernoon.com.

MacPaw Research argues that computer-use AI systems underperform on macOS because public training datasets contain almost no Mac interface data. Their new open-source project, GUIrilla, addresses this by automatically exploring macOS applications and generating structured UI datasets at scale. The release includes GUIrilla-Task, a dataset covering over 1,100 Mac apps and 27,000 tasks, plus macapptree, a Python library for extracting accessibility metadata from Mac applications. Together, these tools aim to improve AI agents, UI understanding models, and developer tooling across the Mac ecosystem.