This startup is helping gaming companies sell their data to world model labs

In the past few years, we have seen an explosive surge in the popularity of large language models (LLMs) like ChatGPT. Over time, generative AI technology has also branched into other areas like text-to-image and text-to-video, with a notable example being Nano Banana. All these models are quite good at what they do, primarily because they have exposure to an extremely vast dataset, albeit with a few concerns around IP protection. In contrast, a domain that has suffered from a lack of proper training data is world-building, but a startup now aims to solve this problem with a novel approach.

TechCrunch reports that Origin Lab has successfully raised $8 million in a seed funding round led by Lightspeed Ventures. Other investors include SV Angel, Eniac, Seven Stars, FPV, Twitch co-Founder Kevin Lin, and Cruise founder Kyle Vogt.

Origin Lab aims to act as a bridge between video game companies and labs building world models. For those unaware, a world model is essentially a representation of the physical world that can be used to train AI models that interact with the real world. A common example of this is a robot that may be assisting the elderly.

The challenge in this domain so far has been the lack of high-quality and reliable training data. Origin Lab intends to solve this issue by enabling game development companies to sell their digital assets to world model labs. Origin Lab will essentially serve as the middleman who converts the digital game asset into a format that can be used to train AI models. This could be multiple hours of walkthrough footage or simply a render that can be understood and fed as input to an AI model.

Origin Lab co-CEO and co-founder Anne-Margot Rodde is quoted as saying:

The AI systems that are being built now need to understand how the physical world works and how things move. That data essentially lives in video games. It became clear that the video game industry was sitting on some incredibly valuable data, but there was no real way or infrastructure to basically connect AI labs and the video game industry. So essentially, we built that bridge.

The benefit of this approach is quite obvious. World model labs like Yann LeCun's AMI Labs will get the high-quality data that they desperately need, while game development companies will be able to open another revenue channel for the work that they have already done.

Over on its website, Origin Lab currently boasts over 500,000 hours of content from over 250 games. It emphasizes that it does not unlawfully scrape data, and instead, licenses it. It also highlights how the delivered data is typically 2-8 hours long, capturing video, audio, human inputs, camera and spatial telemetry, physics and game state, and scene annotations. It will be interesting to see how this particular market evolves, with regard to game development companies as well as world model labs.