Avoiding Data Leakage in Synthetic Data Projects
Synthetic data offers vast opportunities for machine learning model training without real-world constraints. Yet, beneath this promise lies the risk of…
Read more →Worldmodeldata aggregates and structures licensed gameplay data from commercial video games into training datasets for world models, physical AI, and robotics. Its datasets pair 1080p multi-view game capture with engine-tick-aligned action labels and more than 200 structured properties per frame (position, velocity, physics state), delivered in Parquet, HDF5, or JSON-L. Founded by CEO Rhea Loucas, with former Meta VP of Public Policy Richard Allan as chairman and Yann LeCun as senior science advisor, the Cambridge-based company raised a £7M seed round led by Iona Star Capital.
Sensor / Time-seriesVideo