Cambridge Startup Worldmodeldata Secures £7 Million to Develop AI Training Database
Worldmodeldata, a new startup based in Cambridge, has successfully raised £7 million in seed funding as it steps into the public eye. The company is focused on creating a comprehensive database of training data for artificial intelligence, sourced from video games.
The core idea behind Worldmodeldata is the concept of “world models.” These models help AI systems understand how the world functions. Instead of merely reacting to immediate inputs, the world models developed by Worldmodeldata learn about visual aspects, interactions, and changes over time. This knowledge enables them to predict future events and plan actions safely in complex environments. However, the effectiveness of these models heavily depends on the quality of data used for training.
To tackle the challenge of data scarcity, Worldmodeldata collects and organizes rich datasets from modern video games. These datasets encapsulate real human behaviors and interactions within dynamic environments. The company provides essential data for clients, including research labs and robotics companies, who require models capable of understanding dynamics and making safer real-world decisions.
The data is obtained from actual gameplay in popular games built on engines like Unreal and Unity. Unlike many firms that scrape data from the web, Worldmodeldata acquires data through formal licensing agreements, allowing both developers and the gaming community to benefit from their gameplay.
The funding round was led by Iona Star Capital, a venture capital firm based in London that supports early-stage companies at the crossroads of AI, data, and technology.
Worldmodeldata has set an ambitious goal to create a data library containing one million hours of gameplay data by the end of 2026. The funding will be instrumental in furthering this objective, aiding product development, expanding the team, and securing more data-sourcing partnerships.
Rhea Loucas, the founder and CEO of Worldmodeldata, emphasized the significance of their work. “World models signify a major shift in AI, but to make progress, we need vast amounts of data to help these systems make predictions and operate safely in real-world settings,” she said.
According to Loucas, such an extensive dataset is currently non-existent, but video games provide a safe environment to generate the necessary data for training the next generation of AI at the scale required. Worldmodeldata is committed to filling this gap.
